>
Sovereign compute, Arabic models and national champions: what is truly sovereign, and what runs on imported silicon.
Volge Quarterly is Volge's signature series, published every quarter for enterprise customers. Each edition takes one subject, sets out where it currently stands on the published record, and makes a set of dated, falsifiable judgements on where it is going, written to give a buyer something firmer than a forecast: every figure carries its source and its date, every judgement carries the date on which it resolves, and the following year's edition grades each one hit, missed or partial.
In 2027 Gulf sovereignty moves from mandate to architecture. Residency rules, sovereign cloud regions and national control are rebuilding the regional stack layer by layer, and national data-classification tiering is the single mechanism that connects them, deciding for every workload whether it runs in a public region, a sovereign region or an air-gap.
Figures carry their basis and their date. Where published estimates disagree, the range is given and the basis named rather than averaged away, and where a widely quoted number does not survive checking it is corrected here rather than repeated. Every chart plots a figure stated in the text, and nothing is projected beyond what a named source projects.
2027 is the year Gulf digital sovereignty stops being a policy mandate and becomes an engineering specification.
For most of the past decade, sovereignty in the Gulf was a statement of intent. A ministry declared that government data would stay in-country; a cabinet decree announced a cloud-first policy; a national vision promised an artificial-intelligence economy. What the region did not yet have was the machinery to make those statements operational: the physical regions, the certified operators, the key-management systems and, above all, a common way to decide which data could sit where. That machinery is now largely in place, and 2027 is the year it starts to bind. Sovereignty ceases to be a clause in a strategy document and becomes a set of choices made, workload by workload, inside a procurement.
The mechanism that connects every part of this story is national data-classification tiering. Each Gulf state has, or is completing, a scheme that sorts data into tiers: Qatar's runs from C0 Public to C4 Top Secret [NCSA National Data Classification Policy v3.0, 2023]; Saudi Arabia's and Kuwait's turn on four levels with the top two localised [CST CCRF, 2018-19; CITRA Cloud Framework Res 112, 2021]. The tier a dataset carries determines the hosting model it is permitted: a public hyperscaler region, a sovereign region operated by vetted nationals, or, for the most sensitive material, an air-gapped enclave. Classification is therefore the switch. It is the single control that translates a political preference for sovereignty into an instruction a cloud architect can follow.
Two models sit either side of that switch. The localisers (Saudi Arabia, Kuwait and Oman) hold sensitive and government data inside national borders and permit export only under tightly drawn conditions. The open hosts (Bahrain and Qatar) invite foreign data onto their soil, and in Bahrain's case guarantee that it remains governed by the customer's home law [Data Embassy Decree 56/2018]. Both models are answers to the same question, and both are being tested in 2027 by the same force: a build-out of cloud and artificial-intelligence capacity larger than anything the region has attempted.
The Editors, Volge Quarterly · 2027
That build-out carries a paradox at its centre. The Gulf is assembling sovereign compute at extraordinary scale (the United States has agreed to license the United Arab Emirates up to 500,000 advanced NVIDIA chips a year [US-UAE AI Acceleration Partnership, 15-16 May 2025]), yet every one of those chips, and much of the security operating the sovereign clouds around them, depends on United States export licences and United States-operated controls. What the region secures is the sovereignty of its data and its model outputs, not the sovereignty of its supply chain. The chip cap is, in effect, a real-time throttle held in Washington. Capital is not the constraint; chips, power and talent are.
The Calls in this edition put dates and confidence levels on how these forces resolve through 2027. They are written to be graded a year from now. Read them as the firm's considered judgement on a region at an inflection, and as a standing invitation to mark our homework in 2028.
A cloud and compute build-out crossing tens of billions of dollars, governed by six data-protection regimes and one shared idea: that the tier a dataset carries should decide where it is allowed to live.
The reconciled basis used throughout this edition, on one page. Where credible sources differ on scope, a range is given rather than a false point estimate; every figure is sourced and year-stamped.
Reconciled from Mordor Intelligence and Arizton market sizing, Knight Frank data-centre tracking, the US-UAE AI Acceleration Partnership readout, and sovereign-fund disclosures (2024–26). Ranges reflect genuine source disagreement on scope. Current to July 2026.
A capital-rich, power-cheap region buying compute capacity faster than it can staff or supply it, where the binding constraints are chips, power and talent, and the decisive variable is a United States export licence, not a Gulf balance sheet.
Two models of digital sovereignty have emerged in the Gulf. They are opposite answers to one question (may foreign or sensitive data leave the country), and both are operated through the same instrument: national data-classification tiering.
The first model is localisation. Saudi Arabia bars the export of government data and requires its two most sensitive data levels to be held in-country [CST CCRF, 2018-19]; Kuwait requires its private-sensitive (L3) and highly-sensitive (L4) tiers to remain on Kuwaiti soil, exempting only Defence and Interior [CITRA Cloud Framework Res 112, 2021]; Oman keeps sensitive cross-border transfers subject to Cyber Defence Centre approval [PDPL Royal Decree 6/2022]. The localiser's premise is that control follows geography: data held on national territory, operated by vetted nationals, is data the state can protect.
The second model is the open host. Bahrain's Data Embassy law is the purest expression of it: under Decree 56/2018, foreign data hosted on Bahraini soil remains governed by the customer's home law, and Bahraini authorities are barred from accessing it. Qatar, too, positions itself as a regional host with a mature classification and cloud-policy stack. The open host's premise is the inverse of the localiser's: that a credible legal guarantee, not a border, is what makes data sovereign, and that a small state competes by offering law rather than land.
What both models share is the switch. Every regime sorts data into classification tiers, and maps each tier to a permitted hosting model. The lower tiers may sit in a public hyperscaler region; the middle tiers require a sovereign region operated under national control; the top tier is confined to an air-gapped enclave. On top of geography sit four technical layers that harden each rung: data residency; national operation by vetted local staff; customer control of encryption keys through bring-your-own-key or hold-your-own-key schemes; and confidential computing (hardware-isolated execution using Intel TDX, AMD SEV-SNP or NVIDIA Confidential Computing) for data in use.
| Model | Premise | States | Instrument |
|---|---|---|---|
| Localiser | Control follows geography | Saudi Arabia, Kuwait, Oman | In-country residency for top tiers; export controls |
| Open host | Control follows law | Bahrain, Qatar | Legal guarantee over hosted foreign data |
| Shared switch | Control follows classification | All six | Tier → permitted hosting model (public / sovereign / air-gap) |
Sources: CST Cloud Computing Regulatory Framework (2018-19); CITRA Cloud Framework Res 112 (2021); Bahrain Data Embassy Decree 56/2018; Qatar NCSA National Data Classification Policy v3.0 (2023). Current to July 2026.
The models are diverging in law but converging in mechanism. Whichever premise a state adopts, the operational question a buyer faces is identical: what tier is this dataset, and which rung of the stack does that tier permit? That is the question 2027 procurement is built around.
The hyperscaler regions are the physical substrate of Gulf sovereignty. Their announced dates and disclosed investments are set out below; where a region is committed but not yet live, it is marked, because the gap between announcement and availability is itself a 2027 story.
Sovereign cloud in the Gulf is being built along two tracks at once. The global hyperscalers (Amazon Web Services, Microsoft, Oracle, Google) are opening in-country regions and, increasingly, dedicated sovereign or air-gapped variants operated to national requirements. Alongside them, national champions (Core42 in the United Arab Emirates, Center3 and HUMAIN in Saudi Arabia, Beyon in Bahrain, Oman Data Park) provide the sovereign operating layer that vetted local staff and customer key control demand. The table records the disclosed commitments; the caption records the caveats.
Sources: provider announcements and Volge tracking (2024–26). Kuwait's Google and Microsoft regions remain committed but not operational as of July
| Provider | Location | Status / date | Disclosed investment |
|---|---|---|---|
| Amazon Web Services | Saudi Arabia | AI Zone / region build-out | >$5.3bn [AWS, 2024] |
| Oracle | Saudi Arabia | 10-year regional commitment | $14bn [Oracle, 2024-25] |
| Amazon Web Services | UAE | Region | ~$5bn [AWS, 2024] |
| Microsoft | UAE | To 2029; plus $1.5bn into G42 | $15.2bn [Microsoft, 2024-25] |
| AWS (first ME region) | Bahrain | Live since 2019 | First hyperscale ME region |
| Google Cloud | Kuwait | Announced 24 Jan 2023, not live | Undisclosed |
| Microsoft Azure (AI region) | Kuwait | Announced 6 Mar 2025, not live | Undisclosed |
2026: the basis of Call 9. Disclosed hyperscaler GCC commitments exceed $13–15bn in aggregate.
Two features distinguish the Gulf build from a mature market. First, the sovereign variant is now the default ask, not a premium option: Abu Dhabi's Digital Strategy 2025-27 commits AED 13bn and targets a government running on 100% sovereign cloud, AI-native by 2027 [Abu Dhabi, 2025]. Second, the economics still favour the region as a place to build (power at $0.05–0.07 per kilowatt-hour is among the cheapest available to any hyperscaler [analyst estimates, 2025]) even as the sovereign layer adds a 15–30% premium over public cloud for the buyer.
The substrate is largely committed; the variable is delivery. A region announced is not a region live, and in 2027 the difference determines whether a state's sovereignty runs on hyperscaler steel or on telco-and-NVIDIA improvisation.
The Gulf's two largest economies are each building national artificial-intelligence platforms of a scale that has no precedent outside the United States and China. Both rest on imported silicon, and both pair the compute with a national large language model.
Saudi Arabia's vehicle is HUMAIN, a Public Investment Fund company launched on 12 May 2025 and chaired by the Crown Prince, positioned as a programme of the order of $100bn [PIF, 2025]. Its compute commitments are extraordinary: NVIDIA announced on 22 October 2026 a concrete pipeline of up to 84 MW and about 27,000 GPUs in two phases through 2027, against HUMAIN targets of about 3 GW by 2030 and over 6 GW by 2034; the earlier framing was an initial 18,000 GB300 units and several hundred
thousand GPUs over five years across 500 MW; AMD a $10bn, 500 MW commitment; an AWS AI Zone of more than $5bn; Oracle's $14bn; and, from November 2025, a 500 MW xAI-HUMAIN build [company announcements, 2025]. Center3 targets 1 GW by 2030. The national model is ALLaM, SDAIA's Arabic large language model, which has powered HUMAIN Chat since August 2025.
The United Arab Emirates runs a parallel programme through G42, chaired by Sheikh Tahnoon bin Zayed and led by Peng Xiao, which took a $1.5bn Microsoft investment and a Brad Smith board seat in April 2024 after divesting its China and Huawei exposure in 2023-24. Its flagship is Stargate UAE, announced on 22 May 2025: a 1 GW cluster within a 5 GW Abu Dhabi AI campus, first 200 MW live in 2026, backed by OpenAI, Oracle, NVIDIA, Cisco, SoftBank and G42 and built by Khazna: a build the UAE's AI minister put above $30bn in January 2026. The national models are the Technology Innovation Institute's Falcon family (Falcon-H1 and Falcon Arabic, both May 2025) and Jais, from G42, Inception, MBZUAI and Cerebras.
Four sovereign Arabic models now compete for government adoption: ALLaM (SDAIA, Saudi Arabia), Falcon (TII, UAE), Jais (G42 and partners, UAE) and Fanar (QCRI, Qatar). Falcon-H1 is a hybrid architecture reaching 34B parameters with a 256K-token context window across 18 languages [TII, May 2025]. The strategic point is not benchmark leadership but provenance: a model trained, hosted and governed in-region is itself a sovereignty claim, and its adoption into a government production workflow is the test that Call 8 sets for 2027.
Two national compute programmes, four national models, one dependency. Every layer above the silicon is genuinely sovereign; the silicon itself, and the security operating much of the sovereign cloud around it, is licensed abroad. That tension is the subject of the next page.
The paradox is that the same arrangement that unlocks the chips also encodes the dependency. A flow that can be made licence-free can be made licence-bound again; a Country Group can be re-tiered; a
real-time enforcement framework operated to United States requirements is, by construction, a control the United States can tighten. The chip cap is therefore best understood not as a ceiling but as a throttle: a variable held outside the region. Senator Warren's criticism of the G42, MGX and Trump-crypto ties is a reminder that the arrangement is politically contingent as well as commercially structured.
The central tension of Gulf sovereign AI is that its supply chain is not sovereign. Every advanced chip in the region enters under a United States export licence, and much of the security operating the sovereign clouds runs on United States-defined controls. What is secured is the sovereignty of data and output, not of supply.
The mechanism is explicit. Under the US-UAE AI Acceleration Partnership of 15-16 May 2025, the United States agreed to license up to 500,000 advanced NVIDIA chips a year into the Emirates: roughly a fifth to G42, the remainder to United States hyperscalers operating there. The Biden administration's AI Diffusion Rule, which would have capped such flows, was rescinded on 13 May 2025. First licences followed around 9 October 2025, initially for United States-operated data centres only, with G42 excluded until it was cleared on 20 November 2025. On 10-11 July 2026 the Emirates moved to Country Group A:5. The concession is narrower than it reads: G42 and Core42 hold time-limited Supplement 8 eligibility that expires on 6 April 2027 unless renewed or unless they become US companies, and Saudi Arabia was not moved to A:5 and remains licence-by-licence. Within those limits the flow is licence-free for G42, Core42, Amazon, Apple, xAI, Google, Meta, Microsoft, OpenAI and Oracle.
The Gulf can own its data, its keys and its model weights, and it increasingly does. It cannot yet own the fabrication of its compute. Sovereignty of the stack is real from the silicon upward and provisional at the silicon itself, and that single fact is the swing factor the Calls return to.
The Gulf stack has two centres of gravity. Saudi Arabia is the archetypal localiser at national scale; the United Arab Emirates is the region's compute hub and its most complex legal map. Between them they set the terms the smaller states adapt.
Saudi Arabia's Personal Data Protection Law, issued as Royal Decree M/19 in 2021, amended by M/148 in 2023 and fully enforced from 14 September 2024, is administered by SDAIA. It requires breach notification within 72 hours and carries fines to SAR 5m, doubling to SAR 10m for repeat offences, with sensitive-data misuse punishable by up to two years and SAR 3m. The 2024 Data Transfer Regulation permits cross-border movement under adequacy findings, standard contractual clauses or binding corporate rules: a mechanism this edition expects to become the de-facto Gulf standard (Call 2). Above the data law sits the CST's Cloud Computing Regulatory Framework, a four-level content classification under which the top two levels are localised, and a Cloud First Policy. The compute layer (HUMAIN, Center3, ALLaM) is examined on page 05.
The Emirates' federal Personal Data Protection Law, Federal Decree-Law 45/2021, came into force on 2 January 2022 under the UAE Data Office (Law 44/2021), whose functions passed to the Federal Authority for Artificial Intelligence and Data on 14 June 2026, but its executive regulations remained unpublished into mid-2026, so the compliance grace period had not been triggered and a de-facto 72-hour breach norm prevailed. Two financial free zones operate independently: the DIFC, under Data Protection Law 5/2020 with a 2025 amendment adding a private right of action, and the ADGM, under its 2021 regulations with fines to $28m and a dedicated Commissioner. Both are GDPR-aligned. The compute layer (G42,
Core42, Stargate UAE, Falcon, Jais) and the 500,000-chip partnership make the Emirates the region's AI centre, governed by Abu Dhabi's Digital Strategy 2025-27 and its 100% sovereign-cloud, AI-native-by-2027 target.
Saudi Arabia offers scale and a single, fully-enforced national regime; the Emirates offers the region's deepest compute and the most fragmented legal map. A multinational hosting across both is managing one localisation rule in Riyadh and three regulators in the Emirates: the practical shape of Gulf sovereignty in 2027.
The four smaller states span the full spectrum of Gulf sovereignty: the region's first data law, its most explicit localisation regime, its purest open-host guarantee, and its one hard enforcement cliff.
Qatar legislated first. Its Personal Data Privacy Protection Law 13/2016, in force from 13 February 2017, was the Gulf's earliest, administered by the NCSA and NCGAA with 72-hour breach notification and fines of QAR 1–5m. Above it sit the region's most developed classification stack: the National Data Classification Policy v3.0, running C0 Public to C4 Top Secret with confidentiality, integrity and availability impact scored 0-3; the NIA Standard v2.1 (2023), a 2022 Cloud Policy Framework, the Fanar large language model from QCRI, and a National Data Program launched in 2025.
Kuwait has no omnibus law, but CITRA's Data Privacy Protection Regulation (Res 42/2021, updated by Res 26/2024, in force 19 February 2024) sets a 72-hour breach norm and requires adequacy for cross-border transfers. Its Cloud Framework (Res 112, 2021) is the region's most explicit localiser: four tiers, with private-sensitive (L3) and highly-sensitive (L4) data required to remain in Kuwait, Defence and Interior exempt. With the announced Google (24 Jan 2023) and Microsoft (6 Mar 2025) regions not yet live, sovereignty is being delivered by telco and NVIDIA hardware: Ooredoo's first sovereign AI data centre (10 Nov 2025, NVIDIA H200) and KISR's ICARUS HPC/AI platform (12 Aug 2025), the basis of Call 9.
Oman sets the region's one hard date. Its PDPL (Royal Decree 6/2022, in force 13 February 2023) and executive regulations (Ministerial Decision 34/2024) became fully enforceable on 5 February 2026 and widened by Royal Decree 68/2026 from 7 September
2026, which adds extraterritorial reach, cuts the exemptions from ten to six and creates a right to human review of automated decisions after a final grace extension: the Gulf's only firm 2027-relevant cliff. The MTCIT administers it; sensitive cross-border transfers need Cyber Defence Centre approval, a DPO in Oman is preferred, and fines reach RO 500,000. Oman Data Park provides sovereign cloud and Omantel anchors a subsea-cable hub.
Bahrain is the open host. Its PDPL (Law 30/2018, in force 1 August 2019) is administered through the Ministry of Justice with fines of BHD 1,000–20,000, but its signature is the Data Embassy (Decree 56/2018) and a Cloud-First policy that was the first in MENA (2017). AWS opened its first Middle East region in Bahrain in 2019, and Beyon (the former Batelco) provides sovereign cloud.
The state of the sector · Volge Quarterly 2027
The evidence is in the specifications. Abu Dhabi's Digital Strategy 2025-27 does not merely prefer sovereign cloud; it commits AED 13bn and targets a government running entirely on it, AI-native, by 2027 [Abu Dhabi, 2025]. Saudi Arabia's Cloud First Policy is paired with a four-level classification framework that tells a buyer exactly which level may sit where [CST CCRF, 2018-19]. The move from mandate to architecture is what makes Call 1 (that classification-driven residency becomes a default clause in Gulf government RFPs) a high-confidence judgement rather than a hope.
The cost of the shift is rigidity. An architecture, once specified, is expensive to change: a sovereign region is a physical asset, a classification catalogue is an institutional commitment, and a key-management regime binds a buyer to an operator. The Gulf is, in 2027, locking in choices whose consequences run for a decade, which is precisely why the year rewards getting the classification layer right, because it is the one decision every later decision inherits.
Six through-lines that connect the region's build-out, its laws and its dependencies: from the shift out of mandate into architecture to the mechanics of keys, confidential computing and the air-gap.
The defining shift of 2027 is that sovereignty stops being declared and starts being built. The mandate era produced policy; the architecture era produces regions, operators and clauses.
A mandate is a statement that certain data belongs at home. An architecture is the set of systems that make the statement true: the in-country region, the vetted operator, the key-management scheme, the classification catalogue that tells an engineer where a given dataset may run. For most of the past decade the Gulf had mandates without architectures: Bahrain declared cloud-first in 2017, Kuwait set localisation tiers in 2021, but the regions to host the data and the operators to run them were incomplete. What changes now is that the architecture is arriving in volume, and a mandate met by an architecture becomes a procurement rule rather than an aspiration.
2027 is not the year sovereignty is announced: that happened years ago. It is the year sovereignty is specified, priced and signed. The through-line is the hardening of intent into infrastructure.
The Gulf is building sovereign compute on a supply chain it does not control. The paradox is not a flaw in the strategy; it is the strategy's defining condition, and managing it is the region's central 2027 task.
Sovereignty has layers, and they do not all resolve the same way. Data residency, national operation, key control and model weights can all be genuinely sovereign: held, run and owned in-region. Silicon cannot. Every advanced accelerator in the Gulf enters under a United States export licence, and the real-time enforcement frameworks that condition those licences place United States-defined controls inside the region's most sovereign facilities. The 500,000-chip UAE partnership is the clearest case: it is
simultaneously the largest enabler of Gulf sovereign AI and the largest single expression of its dependence [US-UAE partnership, May 2025].
The dependence is dynamic, not static. Between January 2025 and July 2026 the governing regime moved through four states: the Diffusion Rule cap, its rescission, licence-by-licence approval, and a licence-free Country Group A:5 designation. Each transition changed the region's effective compute ceiling without any change in Gulf policy or capital. That is what it means for the chip cap to be a throttle: the variable that most determines Gulf sovereign-AI capacity in 2027 sits in Washington, which is the substance of Call 5.
The strategic response is diversification within the constraint. AMD alongside NVIDIA; national models trained to run on whatever silicon is licensed; sovereign clouds engineered so that a change in chip supply does not compromise data control. None of it removes the dependence (fabrication remains outside the region), but it narrows the blast radius of a licensing reversal. The Gulf cannot make its supply chain sovereign in 2027; it can make its sovereignty of data and output robust to the supply chain's volatility.
Capital is abundant; the constraints are chips, power and talent. Of the three, chips are the only one held abroad, and so the dependence paradox, not the balance sheet, is the ceiling on the year.
If a single instrument explains Gulf sovereignty, it is data classification. The tier a dataset carries decides the hosting model it is permitted, and so the classification catalogue is the control plane of the entire regional stack.
A classification scheme sorts data by sensitivity and maps each tier to a permitted environment. Qatar's is the region's most explicit: C0 Public through C4 Top Secret, each with confidentiality, integrity and availability impact scored from 0 to 3 [NCSA v3.0, 2023]. Saudi Arabia's four content levels and Kuwait's four cloud tiers do the same work: the lower tiers may sit in a public hyperscaler region, the middle tiers require a sovereign region under national operation, and the top tier is confined to an air-gap. The scheme is
not paperwork; it is the switch that converts a sensitivity judgement into an architecture decision.
Because the switch is shared across regimes, it is where practical convergence is happening even as the statutes diverge. A localiser and an open host disagree about whether data may leave the country, but both agree that a Top Secret record and a public dataset belong in different environments, and both express that agreement through a tier-to-hosting map. For a multinational, this is a gift: classification is portable in a way that law is not, so a single internal data catalogue can be mapped onto six national schemes.
The 2027 consequence is procedural. As classification-driven residency becomes a default RFP clause (Call 1), the classification exercise moves upstream: from a compliance afterthought to the first decision in any Gulf cloud project. Get the tiering right and every downstream choice follows; get it wrong and the architecture is either over-secured and over-priced, or under-secured and non-compliant. The switch is where 2027 procurement begins.
Watch the classification catalogues, not the headlines. The tier-to-hosting map is the single artefact that determines where Gulf data lives, and the one that quietly harmonises six divergent regimes.
Sovereignty is not yet free. A sovereign region costs an estimated 15–30% more than public cloud, and in 2027 that premium persists: sovereignty remains a cost the Gulf pays for control, not a saving it earns.
The premium has clear sources. A sovereign region carries lower utilisation than a global one, because it cannot pool demand across the planet; it requires vetted local staff, whose supply is thin; it adds confidential-computing and key-management layers that consume overhead; and it forgoes the arbitrage of running a workload wherever compute is cheapest that hour. BCG puts the resulting mark-up at 15 to 30 per cent over public cloud, a directional figure rather than a consensus [BCG, 2025]. Against it sits one structural offset: power at $0.05–0.07 per kilowatt-hour, among the cheapest available to any operator, which makes the Gulf a cheap place to run the compute even where the sovereign layer is dear.
The premium is a buyer's calculation, not a builder's. For the state hosting the region, cheap power and sovereign-fund capital make the economics work. For the enterprise buying sovereign capacity, the 15–30% is a real line item that must be justified by the sensitivity of the workload, which is exactly why classification matters commercially as well as legally: it prevents a buyer paying the sovereign premium on data that never needed it.
The question the year does not resolve is whether the premium ever inverts. It could, if sovereign regions reach scale, if local talent deepens, and if the alternative (a compliance breach or a data-transfer block) is priced in. But 2027 is too early: the regions are still filling, the talent is still scarce, and the premium holds. Call 10 makes this explicit: sovereignty in 2027 is a cost of doing business in the Gulf, not yet a source of advantage.
Price the premium against the tier. The discipline of the year is to pay for sovereignty only where the classification demands it, and to treat the 15–30% as the current, not permanent, cost of Gulf data control.
The Gulf is not building one model of sovereign compute; it is building several. Saudi Arabia's state-scale programme, the Emirates' partnership-led hub and Kuwait's telco-delivered improvisation are three distinct answers to the same constraint.
The Saudi model is vertically integrated and state-scale. HUMAIN, chaired by the Crown Prince and funded through the Public Investment Fund, bundles chips, data centres, a national model and a cloud into a single $100bn programme, with NVIDIA, AMD, AWS and Oracle as suppliers rather than partners [PIF, 2025]. The premise is that sovereignty and scale are the same project, pursued through national ownership.
The Emirates' model is partnership-led. G42 took Microsoft's capital and a board seat, divested its China exposure, and built Stargate UAE with OpenAI, Oracle, NVIDIA, Cisco and SoftBank [2024-25]. The premise is that sovereignty is best secured by binding the region's compute to the United States technology stack: trading some independence for guaranteed access to
the frontier, and for the Country Group A:5 designation that access earned.
Kuwait's model is delivery by improvisation. With its announced hyperscaler regions not live, sovereignty is being carried by Ooredoo's NVIDIA-H200 sovereign data centre and KISR's ICARUS platform [2025]: telco and research infrastructure standing in for the cloud regions that have not arrived. It is the model of a state that has the mandate and the capital but not yet the hyperscaler steel, and it is the basis of Call 9.
There is no single Gulf compute model to copy. A state's model follows its endowment (Saudi scale, Emirati partnerships, Kuwaiti telco reach), but all three converge on the same dependency and the same classification switch. Whether a Saudi-approved model is recognised across other states is the test of Call 3.
Beneath residency sits a stack of technical controls that make sovereignty real for data in use, not merely data at rest. In 2027 the top of that stack (external key control and confidential computing) becomes a standard requirement for the most sensitive Gulf workloads.
Residency answers where data sits; it does not answer who can read it while it is being processed. The technical layers close that gap. Customer key control (bring-your-own-key, and its stronger form, hold-your-own-key, where the operator never possesses the key) ensures that a cloud provider cannot decrypt customer data even on its own hardware. Confidential computing goes further, isolating data in use inside hardware-encrypted enclaves using Intel TDX, AMD SEV-SNP or NVIDIA Confidential Computing, so that even a compromised host cannot observe the workload. For the top classification tier, the answer is starker still: an air-gap, physically disconnected from any external network.
Call 6 makes the 2027 judgement: hold-your-own-key, external key management and confidential computing become standard requirements for Gulf top-tier, Secret and L4 workloads. The direction is clear even if the pace is uncertain: as sensitive workloads move to cloud, the buyers who own their keys and isolate their
compute are the ones who can host at the top tier without an air-gap, and that capability becomes a procurement expectation rather than a differentiator.
These controls matter because they decouple sovereignty from geography. A workload protected by hold-your-own-key and confidential computing is, in a meaningful sense, sovereign even on foreign-operated hardware, which is what allows an open host like Bahrain to guarantee foreign data on its soil, and what allows the Gulf to accept United States-operated regions without conceding data access. The keys, not the coordinates, become the locus of control.
The air-gap is the answer of last resort; the keys and the enclave are the answer of scale. 2027 is the year the Gulf top tier standardises on external key control and confidential computing: sovereignty engineered into the silicon's trust boundary, not just the map.
The dependence paradox · Volge Quarterly 2027
Sources: the instruments named in the table. Several underlying texts are published in Arabic only, including Bahrain's Resolutions No. 42 and No. 44 of 2022. The resolution number of the Kuwait CITRA cloud framework, and the exact registration criteria for the Saudi Qualifying Category and Classes A, B and C, are not public. Where a label or a figure is contested in the sources, the conflict is stated in the body rather than resolved.
What the rules require, what has actually been built, what it costs, and how a workload is placed against the region's published classification tiers.
There is no Gulf localisation regime. There are six, and the distance between them is wide enough that a workload lawfully placed in one state may be unlawfully placed in the next.
The useful starting point for any buyer in the region is that the six states of the Gulf Cooperation Council do not share a data protection law, a classification scheme, a cloud licensing model or a cross-border transfer route. They share a direction of travel and very little else. The table below sets out, for each state, the principal instrument, what must remain inside the national border, the route by which data may lawfully leave it, and whether a provider must hold a licence or a registration in order to serve the market at all. It is to be noted that the differences are not cosmetic. Two of the six licence the provider as well as regulating the data. One of the six has no general data protection statute. One of the six has legislated to apply a foreign cloud consumer's own home law to data held in designated facilities on its territory.
| State | Principal instruments | What must stay in country | Cross-border route | Cloud licensing |
|---|---|---|---|---|
| Saudi Arabia | PDPL, Royal Decree M/19 of 2021, amended 2023; SDAIA enforcement commenced 14 September 2024; NCA CCC-2:2024 | No blanket localisation under PDPL. Localisation arises sector by sector: SAMA for financial institutions, NCA for critical national infrastructure | Regulation on Personal Data Transfer Outside the Kingdom, amended version dated 1 September 2024, with the Kingdom's first Standard Contractual Clauses issued alongside | CST Cloud Computing Services Provisioning Regulations, in force 10 October 2023, replacing CITC framework v3 (effective 3 December 2020). Registration via CST portal. A Qualifying Category plus Classes A, B and C |
| United Arab Emirates | Federal Decree-Law No. 45 of 2021, in force 2 January 2022; National Cloud Security Policy v2.0, UAE Cyber Security Council | No general statutory localisation. Hosted encryption keys and their backups must be generated, stored and managed within the UAE. At the sovereign tier, all data under exclusive UAE jurisdiction | Articles 22 and 23 turn on adequacy of the destination. No adequacy list has been published, so in practice standard contractual clauses, consent or Data Office approval | No separate cloud licence identified. The Cloud Security Policy binds all providers serving UAE consumers through three sovereignty tiers |
| Qatar | Law No. 13 of 2016 on Personal Data Privacy Protection; NCSA National Data Classification Policy V3.0, established May 2023 | Determined by classification tier. Five tiers, C0 to C4, for government entities, and a minimum of four tiers for non-government entities | Restricted unless the destination ensures adequate protection, the competent authority approves, or a statutory exception applies | No provider licence identified. MCIT Cloud First Policy, document P005 v1.0.0, published April 2024, applies to all government agencies |
| Bahrain | Personal Data Protection Law No. 30 of 2018, in force 1 August 2019; Legislative Decree No. 56 of 2018, the Cloud Law | No general localisation identified. Article 3 of the Cloud Law instead subjects data stored in designated Bahraini data centres by overseas cloud consumers to the law of the consumer's home state | Resolution No. 42 of 2022 permits direct transfer without PDPA permission to listed countries and territories. Resolution No. 44 of 2022 also governs outbound transfers | No provider licence identified. Cloud First Policy, April 2017 |
| State | Principal instruments | What must stay in country | Cross-border route | Cloud licensing |
|---|---|---|---|---|
| Oman | PDPL, Royal Decree No. 6/2022, published 13 February 2022; Executive Regulations by Ministerial Decision No. 34/2024, published 4 February 2024; fully enforceable from 5 February 2026 | Article 12 of the cloud bylaw prohibits transferring Levels 3 and 4 data, which include governmental and financial data, outside Oman without prior TRA approval | Article 23 of the PDPL and Article 37 of the Executive Regulations. Firms read the consent requirement differently. Anonymised transfers are treated as outside it | TRA Decision No. 1152/2/19/2024-20, September 2024, in force the day after Gazette publication. A licence is required to establish, operate or provide cloud or data centre services. Three-year licences |
| Kuwait | No single national data protection statute. CITRA Resolution No. 26 of 2024 on data privacy, superseding Resolution No. 42 of 2021, applying to CITRA-licensed telecommunications and IT service providers | Storage, classification and localisation requirements for government and citizen data under the CITRA cloud framework | No general statutory transfer route identified in the sources | CITRA Cloud Computing Regulatory Framework, version 2.4 cited, applying to licensed providers operating data centres in Kuwait |
The first thing the table shows is that localisation in the Gulf is rarely imposed by the privacy statute itself. The Saudi Personal Data Protection Law, issued as Royal Decree M/19 in 2021 and amended in 2023, imposes no blanket requirement that personal data be held in the Kingdom: the obligation arises sector by sector, through the Saudi Central Bank for financial institutions and through the National Cybersecurity Authority for critical national infrastructure. Federal Decree-Law No. 45 of 2021 in the UAE, in force since 2 January 2022, carries no general localisation mandate either, and the in-country obligation a buyer will actually encounter sits elsewhere, in the requirement under the National Cloud Security Policy v2.0 that hosted encryption keys and their backups be generated, stored and managed within the UAE. The consequence is that a compliance assessment built on the privacy law alone will understate the constraint in both of the region's two largest markets.
The third thing the table shows is that in two of the six states the provider is licensed and not merely supervised. Saudi Arabia requires registration under the CST Cloud Computing Services Provisioning Regulations, and Oman requires a three-year licence under the TRA bylaw to establish, operate or provide cloud or data centre services, with Article 28 permitting administrative penalties and the suspension of permits. Kuwait's framework reaches licensed providers operating data centres in the country rather than the
market at large. The commercial effect is that in those jurisdictions the field of lawful suppliers is narrower than the field of capable ones.
The second thing the table shows is that cross-border routes are open in principle and narrow in practice. Saudi Arabia published the amended Regulation on Personal Data Transfer Outside the Kingdom dated 1 September 2024 and issued its first Standard Contractual Clauses alongside it, although one source dates those clauses to August 2024 and the two accounts are unreconciled. Qatar permits transfer where the destination ensures adequate protection, where the competent authority approves, or where a statutory exception applies, with the regulator cited variously as the National Cyber Security Agency and as the MCIT Compliance and Data Protection Department. The UAE position is the most awkward of the six: Articles 22 and 23 turn on the adequacy of the destination, but no adequacy list has been published, which leaves standard contractual clauses, consent or Data Office approval as the working routes. Oman's position is unsettled in a different way, in that firms read the consent requirement under Article 23 of the PDPL and Article 37 of the Regulations differently, with anonymised transfers treated as falling outside it.
The fourth thing the table shows is an outlier that deserves more attention than it receives. Article 3 of Bahrain's Legislative Decree No. 56 of 2018, the Cloud Law, subjects data stored in designated Bahraini data centres by overseas cloud consumers to the law of the consumer's home state rather than to Bahraini law. This is the inverse of the sovereignty logic every other state in the table applies: instead of pulling foreign data under local jurisdiction, it pushes local hosting under foreign jurisdiction, by consent of the consumer. For a buyer whose principal concern is predictability of governing law rather than physical residency, that is a significantly different proposition, and it is the only such provision in the six.
The Bahraini provision operates on data centres designated by a Council of Ministers resolution. Whether that resolution has been issued could not be confirmed from the sources available, and the provision should be checked against the primary text and the designation instrument before it is relied upon.
The takeaway is that a single regional compliance position is not available and should not be constructed. What the table supports instead is a per-state, per-sector matrix, built from the named instruments rather than from vendor summaries, and refreshed on the
dates the instruments themselves carry. Where the sources conflict, as they do on the Saudi clauses and on the Kuwaiti notification period, the conflict belongs in the matrix rather than being resolved by preference.
The fifth thing the table shows is that the stated penalties are high and the enforcement record is empty. Saudi Arabia provides for fines of up to SAR 5m per violation, doubled on repetition, up to SAR 3m and two years' imprisonment for unlawful disclosure of sensitive data, and up to SAR 1m for unlawful transfer abroad. Qatar's penalties are reported at QAR 1m to QAR 5m with no custodial element. Bahrain's are reported at up to BHD 20,000 per breach plus criminal sanctions for unauthorised cross-border transfer. Oman's are reported from OMR 500 to OMR 500,000 for unlawful cross-border transfer. Against those ceilings, no named published fine or enforcement decision under any of the six laws could be located. That absence is itself a finding, and it cuts both ways: it tells a buyer that the probability of a public penalty is currently unmeasurable, and it tells the same buyer that no body of practice yet exists to clarify how the written rules will be read.
Penalty figures above are statutory or reported ceilings drawn from secondary commentary, not amounts imposed. Kuwait's breach notification period is reported as 24 hours by one source and 72 hours by another, and the two accounts are unreconciled. CITRA is also reported to have repealed its 2021 Data Classification Policy alongside Resolution No. 26 of 2024, which could not be confirmed.
In the Kingdom the control level applied to a workload is decided by the data's classification rather than by the privacy statute, and the date by which a bidder must be ready is decided by procurement rather than by the regulator.
The Saudi regime is best read as two mechanisms working at different speeds. The slower one is classification: the National Data Management Office scheme sorts data into four levels, and the National Cybersecurity Authority's Cloud Cybersecurity Controls then attach a control level to each, so that the hosting model follows from the label rather than from the contract. The faster one is procurement: from 1 April 2027 local content carries weight in the financial evaluation of IT service contracts for government tenders of SAR 10m or more, which is a harder date than anything in the privacy framework. A buyer who prepares only for the Personal Data Protection Law will be compliant and uncompetitive.
Transfers out of the Kingdom are governed by the Regulation on Personal Data Transfer Outside the
Kingdom, in the amended version dated 1 September 2024, alongside which the Kingdom's first Standard Contractual Clauses were issued. One source dates the clauses to August 2024. The two accounts are unreconciled and a buyer drafting a transfer instrument should take the date from the published text rather than from commentary. The enforcement ceilings are substantial: up to SAR 5m per violation, doubled on repetition; up to SAR 3m and two years' imprisonment for unlawful disclosure of sensitive data; and up to SAR 1m for unlawful transfer abroad. No named enforcement decision under the law has been verified.
The Personal Data Protection Law was issued as Royal Decree M/19 in 2021 and amended in 2023, with enforcement by the Saudi Data and Artificial Intelligence Authority commencing on 14 September 2024. It is to be noted that the law imposes no blanket localisation. The obligation to keep data in the Kingdom is sectoral: the Saudi Central Bank imposes it on financial institutions, and the National Cybersecurity Authority imposes it on critical national infrastructure. The practical reading is that the statute governs how personal data is handled, while the question of where a workload may run is answered by the sector regulator and by the classification scheme.
The National Data Management Office operates a national data classification scheme of four levels keyed to the severity of harm that disclosure would cause. The labels are contested. Saudipedia gives Top Secret, Secret, Restricted and Public. The policy text and the vendor summary published by Fortra give Highly Confidential, Confidential, Restricted and Public. Both accounts agree on four levels and on the harm-severity basis; they disagree on nomenclature, and the conflict is not resolved in the public sources. The significance for a buyer is modest in architecture and considerable in paperwork, because tender responses, data inventories and contractual schedules all have to name a level, and naming it from the wrong source invites a finding.
The control side is clearer. The National Cybersecurity Authority's Cloud Cybersecurity Controls CCC-1:2020 have been superseded by CCC-2:2024, which cover government bodies and private owners and operators of critical national infrastructure. The controls run to four levels keyed to data classification, with Level 1 applied to top secret material and Level 2 to secret material. This is the two-axis model in its simplest form: the classification axis sets sensitivity, and the controls axis sets the obligations that follow. The two together, rather than the statute, determine whether a workload can sit in a public region at all.
Providers require registration. The Communications, Space and Technology Commission's Cloud Computing Services Provisioning Regulations came into force on 10 October 2023, replacing the CITC framework version 3 that had been effective from 3 December 2020. Registration runs through the CST portal, and the tiers are described as a Qualifying Category plus Classes A, B and C. The exact registration criteria for each are not public. A buyer cannot therefore verify from published material which class a given provider ought to hold for a given workload, and the position is that the class must be confirmed with the provider and, where the workload warrants it, against the regulator.
From 1 April 2027, local content carries weight in the financial evaluation of IT service contracts for tenders of SAR 10m or more. This is the single most consequential date in the Saudi calendar for a technology buyer, and it is not a privacy date. It changes the arithmetic of a bid rather than the lawfulness of a deployment. Two commercial dates sit immediately before it. The AWS Saudi Arabia region, announced in March 2024 with three Availability Zones and an announced investment of USD 5.3bn, was still not live as at early September 2026, when AWS reiterated a December 2026 launch. The Microsoft Azure Saudi Arabia East region in the Eastern Province, also with three Availability Zones, is targeted for the fourth quarter of 2026 and is likewise not live. It is to be noted that the USD 5.3bn figure is announced and has not been disclosed as committed or as spent.
A third arrangement is weaker than its coverage suggested. The Public Investment Fund, SITE and Microsoft signed a non-binding memorandum of understanding on 20 November 2025 to explore the
delivery of Microsoft sovereign cloud services in the Kingdom, subject to regulatory approvals. It is not a joint venture and no value has been published. Separately, stc Group operates as a sovereign cloud provider using Oracle Alloy, although the operator-control and key-custody terms are not public, which is precisely the information a buyer with classified workloads needs.
The Communications, Space and Technology Commission published a draft Global AI Hub Law on 14 April 2025, with consultation closing on 14 May 2025. The draft creates Private, Extended and Virtual Hubs: in substance, data embassies operating under foreign law on Saudi territory. If enacted it would be the most significant structural change in the regional picture, because it would offer a lawful route to foreign governing law inside the Kingdom rather than outside it. It remains a draft and no commencement date has been published. It should be tracked, not planned around.
The takeaway for the Kingdom is sequencing. Classification precedes control selection, control selection precedes region selection, and all three precede the 1 April 2027 procurement change, which is the only firm regulatory date on the horizon and which binds bidders rather than operators. Where the public record is silent, as it is on CST class criteria, on the exact NDMO labels and on the key-custody terms of the sovereign cloud arrangements, the silence is to be closed by written confirmation from the counterparty rather than by assumption.
The UAE has produced the only sovereignty taxonomy in the Gulf that a buyer can act on directly. The tier, and not the federal privacy statute, is what decides where a workload may sit.
The position in the United Arab Emirates is unusual: the privacy statute is the weaker instrument and the cloud security policy is the stronger one. Federal Decree-Law No. 45 of 2021 has been in force since 2 January 2022, but the core Executive Regulations appear still to be unissued as at 2026 and no adequacy list has been published, which leaves the transfer provisions difficult to operate. The National Cloud Security Policy v2.0, issued by the UAE Cyber Security Council and reported on 15 October 2026 as an update to the first release of May 2023, does the opposite: it sets out three sovereignty tiers with named conditions, and it is
the most directly actionable sovereignty taxonomy published anywhere in the region.
The policy applies on the supply side to all cloud service providers serving UAE consumers, and on the demand side to federal ministries and authorities together with non-government critical information infrastructure entities. Where an Emirate runs its own critical information infrastructure programme, that Emirate's lead authority assesses and enforces, which means the applicable supervisor can differ between two entities in the same sector depending on where they sit. Free zones remain separate: the DIFC operates under Data Protection Law No. 5 of 2020 and the ADGM under its Data Protection Regulations 2021. A buyer with entities inside and outside a free zone is therefore dealing with more than one regime in a single country.
| Tier | When it applies | What it requires |
|---|---|---|
| Tier 1, Baseline secure cloud | All providers, irrespective of data classification | Foundational controls |
| Tier 2, Partial sovereign cloud | Data classified confidential or restricted | All consumer data stored strictly within UAE geographical boundaries and legal jurisdiction. On a valid foreign disclosure order, the provider must make reasonable efforts to redirect the request to the consumer and must inform the consumer and the UAE authorities |
| Tier 3, Sovereign cloud | Data classified above confidential or restricted | All data under exclusive UAE jurisdiction. Consumer data and metadata not disclosed to any authority other than those explicitly approved by the UAE. All foreign access requests denied by default. Expect UAE-based key management, UAE-resident redundancy, UAE-licensed telecommunications routing and UAE-cleared personnel as eligibility conditions |
Three rules run across all three tiers and are easy to miss because they are not tier-specific. Data location must be known at all stages. Hosted encryption keys and their backups must be generated, stored and managed within the UAE. And the consumer, not the provider, must assess whether hold-your-own-key is required. Where tiers overlap, the most stringent control applies, which removes the temptation to argue a workload down into a cheaper tier by reference to one of its less sensitive components.
The tier descriptions above rest on a law firm summary dated 15 October 2026. The primary text of the National Cloud Security Policy v2.0 was not retrieved and the conditions should be checked against it before a tiering decision is treated as settled.
Tier 1 asks for diligence and little more, and it applies regardless of classification, so there is no workload in scope of the policy that escapes it. Tier 2 asks for residency and for a disclosure protocol: the requirement that the provider make reasonable efforts to redirect a foreign order to the consumer, and inform both the consumer and the UAE authorities, is a procedural obligation that a buyer can and should see evidenced in the provider's own documented process rather than asserted in a sales response. Tier 3 asks for something closer to an operating model: exclusive jurisdiction over data and metadata, denial of foreign access requests by default, and a set of eligibility conditions covering key management, redundancy, routing and personnel clearance. The distance between Tier 2 and Tier 3 is not a feature flag. It is a different supply chain.
The transfer position under the federal statute is the weak link. Articles 22 and 23 turn on the adequacy of the destination, but with no adequacy list published, adequacy cannot be established by reference to a list. In practice transfers proceed on standard contractual clauses, consent or Data Office approval. The status of the core Executive Regulations could not be confirmed, and a date of full compliance for the federal law that circulates in secondary commentary could not be verified against any primary instrument and is therefore not treated here as a date at all.
The UAE has the deepest built capacity in the region. Microsoft operates the live UAE North and UAE Central regions. Microsoft's announced commitment to the UAE runs to USD 15.2bn through 2029, of which USD 7.3bn is stated as invested by the end of 2025 and USD 7.9bn is planned for 2026 to 2029; more than USD 10bn of the total is earmarked for cloud and artificial intelligence data centres and over USD 3bn for local operations and research. This builds on a USD 1.5bn equity investment in G42 that carries a board seat. On 5 November 2025 Microsoft and G42 announced a 200 MW data centre expansion to be delivered by Khazna Data Centers, with operations expected before the end of 2026; site locations and
capital expenditure were not disclosed. The e& and AWS UAE Sovereign Launchpad is framed on local data residency, but its terms are not public.
One point of control deserves naming because it is frequently assumed and not found. No published golden-share instrument or formal operator-control instrument between a GCC state entity and a hyperscaler could be located. The nearest published control features in the UAE are Microsoft's board seat at G42 and the in-country key custody rule in the Cloud Security Policy. A buyer relying on state control as a sovereignty assurance is relying on something that has not been published.
A Federal Register notice dated 10 July 2026 is reported to have moved the UAE from Country Groups D:3 and D:4 to A:5, clearing G42, Core42 and MGX for licence-free advanced chip purchases. This rests on secondary reporting; the primary notice text was not retrieved. Key custody and operator-access terms for Core42 and for the e& and AWS arrangement are likewise not public.
The takeaway is that in the UAE the tiering decision is the architecture decision, and it is made before a provider is selected rather than after. The three cross-tier rules, and in particular in-country key custody and the hold-your-own-key assessment, are the items most likely to be overlooked in a Tier 1 or Tier 2 deployment, because they do not feel like sovereignty requirements and they are not negotiable.
The four smaller GCC markets are routinely treated as a single residual category. They are not. One applies the consumer's home law to locally hosted data, one licenses every cloud operator and fences two classification levels inside its border, one works through classification tiers, and one has no general data protection statute at all.
The habit of grouping Qatar, Bahrain, Oman and Kuwait as the rest of the Gulf is the single most expensive simplification available to a regional buyer, because the four differ from one another more than either of the two large markets differs from its neighbour. Bahrain's Cloud Law points jurisdiction outwards. Oman's cloud bylaw licenses the operator and fences two levels of data inside the country. Qatar's position turns on a five-tier classification scheme and a government cloud-first policy. Kuwait has no general data protection statute and regulates through the telecommunications and information technology licence instead. A deployment template built for any one of them will fail in at least two of the others.
The operative instrument for placement is the National Cyber Security Agency's National Data Classification Policy V3.0, established in May 2023, which sets five tiers, C0 to C4, for government entities and a minimum
of four tiers for non-government entities. Alongside it, the MCIT Cloud First Policy, document P005 v1.0.0, published in April 2024, applies to all government agencies. The built position is reasonably mature for the size of the market: Microsoft's Qatar Central region launched in August 2022 with the local partner MEEZA, and Google Cloud's Doha region opened in May 2023 with three zones, having been committed to the Government of Qatar in March 2020.
Qatar operates under Law No. 13 of 2016 on Personal Data Privacy Protection, under which cross-border transfer is restricted unless the destination ensures adequate protection, the competent authority approves, or a statutory exception applies. The regulator is cited variously as the National Cyber Security Agency and as the MCIT Compliance and Data Protection Department, and the sources do not reconcile the two. Penalties are reported in the range of QAR 1m to QAR 5m with no custodial element, and no enforcement action was found.
One secondary summary states that C4 data is exempt from cloud migration. This is unverified and should be checked against the primary classification policy before it is relied upon in a placement decision.
Bahrain's Personal Data Protection Law No. 30 of 2018 has been in force since 1 August 2019, supervised by the Personal Data Protection Authority. Resolution No. 42 of 2022 permits direct transfer without the authority's permission to listed countries and territories, and Resolution No. 44 of 2022 also governs outbound transfers; the official texts are available in Arabic only. Penalties are reported at up to BHD 20,000 per breach plus criminal sanctions for unauthorised cross-border transfer, and a Cloud First Policy has been in place since April 2017.
The distinctive instrument is Legislative Decree No. 56 of 2018, the Cloud Law. Article 3 subjects data stored in designated Bahraini data centres by overseas cloud consumers to the law of the consumer's home state. Every other regime in the Gulf seeks to extend local jurisdiction over data held locally. Bahrain's provision does the reverse, and does so deliberately, as an inducement to foreign consumers who want physical proximity without a change of governing law. The qualifying data centres are designated by a Council of Ministers resolution, and whether that resolution has been issued could not be confirmed, which is a significant gap in an otherwise attractive proposition. Bahrain's built position is also the oldest in the region: AWS opened three facilities in July 2019, the first hyperscaler region in the Gulf, and Bahrain's Information and eGovernment Authority reports 1,357 government information technology operations moved to AWS, on a page last updated in March 2025.
Oman has moved furthest, most recently. The Personal Data Protection Law, Royal Decree No. 6/2022, was published on 13 February 2022, its Executive Regulations followed by Ministerial Decision No. 34/2024 published on 4 February 2024, and the regime became fully enforceable from 5 February 2026. Transfers run under Article 23 of the law and Article 37 of the Regulations, with firms reading the consent requirement differently and anonymised transfers treated as outside it. Penalties are reported from OMR 500 to OMR 500,000 for unlawful cross-border transfer.
The cloud bylaw is the harder instrument. TRA Decision No. 1152/2/19/2024-20, issued in September 2024 and in force the day after Gazette publication, requires a licence to establish, operate or provide cloud or data centre services, granted for three years. Article 12 prohibits transferring Levels 3 and 4 data, which explicitly include governmental and financial data, outside Oman without prior TRA approval. Article 28 provides for administrative penalties and permits may be suspended. The full level definitions and the licence classes were not retrieved, so the boundary of Levels 3 and 4 beyond the governmental and financial categories cannot be stated from public material. Oman also has no hyperscaler region domiciled in the Sultanate as at May 2026: provision is through Omani operators and the national shared artificial-intelligence platform Mu'een, with the Oman Data Park Muscat facility listed at 10 MW. The combination of a fenced
classification level and no domestic hyperscaler region is the tightest constraint in the four.
Kuwait has no single national data protection statute. CITRA Resolution No. 26 of 2024 on data privacy supersedes Resolution No. 42 of 2021 and applies to CITRA-licensed telecommunications and information technology service providers rather than to controllers generally. Breach notification is reported as 24 hours by one source and 72 hours by another, and the two are unreconciled. CITRA is also reported to have repealed the 2021 Data Classification Policy at the same time, which could not be confirmed. The CITRA Cloud Computing Regulatory Framework, version 2.4 cited, applies to licensed providers operating data centres in Kuwait and carries storage, classification and localisation requirements for government and citizen data; its resolution number is not public. No enforcement action was found. On the built side, the Google region is not live, with reported requests for 60 MW across three sites or a single Sulaibiya site at 30 MW, both marked planned, and a Microsoft Azure artificial-intelligence region listed at 50 MW, planned. The Ooredoo and NVIDIA Kuwait Sovereign AI Lab, listed at 10 MW, is described as operational.
Megawatt figures for Kuwait, Oman and Bahrain in this chapter derive from an index site and are mostly marked planned or estimated. They are indicative and are not published operator figures.
The takeaway is that the four smaller markets each require their own placement rule, and that the binding constraint differs in each: governing law in Bahrain, operator licensing and fenced levels in Oman, classification tiering in Qatar, and the provider's licence status in Kuwait. Where the sources are unreconciled, as on the Qatari regulator, the Kuwaiti notification period and the Bahraini designation resolution, the primary texts are the only reliable authority, and two of them are published in Arabic only.
The gap between the Gulf's announced cloud capacity and its live cloud capacity is the most important single fact in this edition. Both of the Kingdom's hyperscaler regions remain unlaunched, and several of the arrangements marketed as sovereign have no published terms at all.
The region's built position is strong in the United Arab Emirates, Qatar and Bahrain, and still prospective in Saudi Arabia, Oman and Kuwait. That distinction matters more than the headline investment figures, because a region that is not live cannot host a workload and a memorandum of understanding
cannot be architected against. The table below separates what is operating from what has been announced, and records, where it exists, the announced value or capacity. It is to be noted that in almost every row the announced value is announced only: it has not been disclosed as committed or as spent.
| Arrangement | Status | Date | Announced value or capacity |
|---|---|---|---|
| AWS Bahrain, three facilities | Live. The first hyperscaler region in the region | Opened July 2019 | Not published. Bahrain's iGA reports 1,357 government IT operations moved to AWS, page last updated March 2025 |
| Microsoft Azure UAE North and UAE Central | Live | Not stated in the sources | Not published |
| Microsoft Qatar Central, with local partner MEEZA | Live | Launched August 2022 | Not published |
| Google Cloud Doha, three zones | Live | Opened May 2023. Committed to the Government of Qatar March 2020 | Not published |
| Google Cloud Dammam | Live | Opened 2023 | Not published |
| Oracle Jeddah and Riyadh | Live | Not stated in the sources | Not published |
| AWS Saudi Arabia, three Availability Zones | Not live. As at early September 2026 AWS reiterated a December 2026 launch | Announced March 2024 | USD 5.3bn announced, not disclosed as committed or spent |
| Microsoft Azure Saudi Arabia East, Eastern Province, three Availability Zones | Not live | Targeted Q4 2026 | Not published |
| Microsoft UAE investment programme | Part invested | Through 2029 | USD 15.2bn announced; USD 7.3bn stated as invested by end-2025; USD 7.9bn planned 2026 to 2029; more than USD 10bn earmarked for cloud and AI data centres; over USD 3bn for local operations and research; builds on a USD 1.5bn equity investment in G42 carrying a board seat |
| Microsoft and G42 UAE expansion, delivered by Khazna Data Centers | Announced. Operations expected before end-2026 | Announced 5 November 2025 | 200 MW. Site locations and capex not disclosed |
| Arrangement | Status | Date | Announced value or capacity |
|---|---|---|---|
| PIF, SITE and Microsoft sovereign cloud in Saudi Arabia | Non-binding memorandum of understanding to explore delivery, subject to regulatory approvals. Not a joint venture | Signed 20 November 2025 | No value published |
| stc Group on Oracle Alloy | Operating as a sovereign cloud provider | Not stated in the sources | Operator-control and key-custody terms not public |
| e& and AWS UAE Sovereign Launchpad | Framed on local data residency | Not stated in the sources | Terms not public |
| Ooredoo and NVIDIA Kuwait Sovereign AI Lab | Operational | Not stated in the sources | Listed at 10 MW |
| Batelco (Beyon) and Qareeb Data Centers, Bahrain edge AI-ready data centre | Launched | January 2026 | Listed at 10 MW |
| Oman Data Park Muscat AI facility | Live. No hyperscaler region domiciled in the Sultanate as at May 2026; provision via Omani operators and the national shared-AI platform Mu'een | As at May 2026 | Listed at 10 MW |
| Google Cloud Kuwait | Not live. Planned | Not published | Reported requests for 60 MW across three sites, or a single Sulaibiya site at 30 MW. Both marked planned |
| Microsoft Azure AI region, Kuwait | Planned | Not published | Listed at 50 MW, planned |
Megawatt figures for Kuwait, Oman and Bahrain derive from an index site and are mostly marked planned or estimated. They should be treated as indicative and not as published operator figures. Where the sources do not state a launch date for a live region, none is given above rather than inferred.
What the table shows first is that the region's oldest hyperscaler footprint is the smallest market's. AWS opened three facilities in Bahrain in July 2019, and Bahrain has the only published migration volume in the region: 1,357 government information technology operations moved, on a page last updated in March 2025. No comparable published figure exists for any other GCC state. That is not evidence that migration has not happened elsewhere; it is evidence that it has not been published, and the two should not be confused.
What the table shows third is that the word sovereign is doing more work in the market than the published record supports. Three of the arrangements most often described as sovereign, the stc Group
deployment on Oracle Alloy, Core42 in the UAE, and the e& and AWS Sovereign Launchpad, have no public key-custody or operator-access terms. Those terms are the substance of sovereignty: who holds the keys, who can be compelled, and what the operator is contractually able to see. In their absence the label is a marketing description rather than a verifiable control, and the only way to convert it into one is a written answer from the counterparty on key custody, operator access and response to foreign legal process.
What the table shows second is that Saudi Arabia, the largest market and the most active policy-maker, has no hyperscaler region live from the two largest providers. The AWS region was announced in March 2024 and as at early September 2026 the December 2026 launch was being reiterated rather than delivered. The Azure Saudi Arabia East region is targeted for the fourth quarter of 2026. A buyer planning a 2027 Saudi deployment is therefore planning against two unlaunched regions and a procurement rule that begins to bite on 1 April 2027. Google Cloud Dammam and the Oracle regions in Jeddah and Riyadh are live, which narrows the practical field rather than removing it.
What the table shows fourth is the distinction between announced, committed and spent, and how rarely the public record permits it. Microsoft's UAE programme is the exception that proves the rule: of USD 15.2bn announced through 2029, USD 7.3bn is stated as invested by the end of 2025 and USD 7.9bn is planned for 2026 to 2029. That is a disclosure structure no other arrangement in the table matches. The AWS Saudi Arabia figure of USD 5.3bn is announced and has not been disclosed as committed or spent. The Microsoft and G42 200 MW expansion announced on 5 November 2025 carries no disclosed site locations or capital expenditure. The PIF, SITE and Microsoft memorandum of 20 November 2025 is explicitly non-binding, is not a joint venture, is subject to regulatory approvals, and carries no published value.
The takeaway is that a capacity plan should be built on the live column and nothing else, with announced capacity treated as optionality rather than supply. Where a provider's sovereignty claim rests on an arrangement whose terms are not public, the claim is unverified until the terms are obtained in writing, and the absence of any published golden-share or formal operator-control instrument anywhere in the Gulf means state involvement cannot be relied upon to supply the assurance in its place.
The Gulf's sovereign artificial-intelligence programmes are measured almost entirely in announced megawatts. No operating megawatt figure and no energised GPU count has been published for the region's two flagship clusters, and the compute they depend on sits behind foreign export decisions.
The single most useful discipline a buyer can apply to the Gulf's sovereign artificial-intelligence build-out is to ask what is energised. The announced numbers are very large and are becoming larger. HUMAIN in Saudi Arabia targets approximately 3 GW by 2030 and more than 6 GW by 2034. Stargate UAE is a 1 GW cluster inside a 5 GW campus in Abu Dhabi. Against those figures, no published operating megawatt figure and no energised GPU count exists for HUMAIN, and no official confirmation of commissioning for Stargate UAE could be located. The capacity is announced. Whether it is running is not public.
HUMAIN and NVIDIA announced a partnership in May 2025 projecting up to 500 MW and several hundred thousand graphics processing units over five years. That was superseded in specificity, though not in ambition, by an expansion announced on 22 October 2026 covering up to 84 MW in two phases. Phase one runs from the fourth quarter of 2026 at 14 MW of Blackwell HGX B300, approximately 6,000 GPUs. Phase two runs from mid-2027 at 70 MW of Vera Rubin, approximately 21,000 GPUs. Nebius is named as anchor customer for up to 50 MW at Oxagon. Riyadh and Dammam sites were reported at up to 100 MW each initially, with launch targeted for 2026.
It is to be noted that these figures do not reconcile into a single capacity number and should not be added. The 500 MW of May 2025 is a five-year projection. The 84 MW of October 2026 is a two-phase programme. The 3 GW by 2030 and more than 6 GW by 2034 are targets announced only. The 100 MW per
site for Riyadh and Dammam is a report of an initial figure. Each is a separate statement on a different basis, and the only honest aggregate is that none of them is an operating figure.
Separately, AWS and HUMAIN announced an AI Zone with more than USD 5bn announced and up to 50 MW available by 2028, using AWS Trainium and NVIDIA infrastructure. The USD 5bn is announced, not disclosed as committed or spent, and the 50 MW is an availability target for 2028 rather than a current figure.
Stargate UAE is a 1 GW cluster being built by Khazna inside the 5 GW UAE and US AI Campus in Abu Dhabi. The first phase is 200 MW, with up to 35,000 NVIDIA GB300. The target was completion in the third quarter of 2026. That target date has passed and no official confirmation of commissioning could be located. The correct statement of the position is therefore neither that the cluster is running nor that it is late, but that its status is not public. For a buyer evaluating whether a sovereign artificial-intelligence platform in the UAE can carry a production workload in 2027, that is the finding, and it is not one a vendor roadmap will supply.
Two decisions in Washington govern whether the silicon arrives. On 19 November 2025 US Commerce authorised advanced chip exports to G42 in the UAE and HUMAIN in Saudi Arabia, each up to the equivalent of 35,000 NVIDIA GB300, conditioned on security and reporting requirements monitored by the Bureau of Industry and Security. The licence numbers, the full list of conditions and the delivery schedules are not public. On 10 July 2026 the UAE was moved to Country Group A:5, with G42, Core42 and MGX cleared for licence-free purchases. Saudi Arabia is not covered by that change and remains on the November 2025 licence basis.
The 10 July 2026 Country Group reassignment rests on secondary reporting dated to a Federal Register notice of that date; the primary notice text was not retrieved and should be checked. The conditions attaching to the 19 November 2025 authorisations are not public and the figures above state only what was reported.
The asymmetry created by those two dates is the most significant structural fact in the regional compute picture. Since 10 July 2026 the UAE's named entities buy without a licence, while Saudi Arabia's flagship programme buys under a licence whose conditions are not published and whose delivery schedule is not published. Two national programmes with comparable ambition therefore sit on different procurement footings, and a buyer committing to a Saudi sovereign artificial-intelligence platform is accepting a supply dependency whose terms it cannot read.
The takeaway is that capacity planning against sovereign artificial-intelligence platforms in the Gulf is currently planning against announcements. The question that distinguishes a usable platform from a press release is a short one and it is answerable by the operator rather than by the market: what is energised today, what is contracted with a delivery date, and under which export authorisation does the hardware arrive. Until the first of those three is published, the announced megawatt figures should carry no weight in a workload decision.
No vendor has published a sovereign-cloud price delta for any Gulf market. The only measurable sovereign premium in the world is European, and the only firm Gulf cost figures are construction costs, which are the lowest of any region.
The honest answer to what sovereignty costs in the Gulf is that no one has published it. No GCC sovereign-cloud price delta is published by any vendor. No published tender or contract value for GCC government cloud could be located: not from Etimad, not from the UAE, not from Qatar, and the Saudi Deem service appears only as a government cloud category in a World Bank report with no contract detail attached. What exists instead is a European proxy that can be measured, a construction cost index that can be compared, and a pipeline figure that is announced rather than committed. Those three are useful. They are not a price.
The AWS European Sovereign Cloud is the only sovereign offering in the world whose premium can be measured, because AWS publishes its list prices through its own calculator and a buyer can therefore compute the gap for a specific workload. The measurements do not agree. tecRacer measured roughly 15 per cent above eu-central-1 in Frankfurt across common services in January 2026. Devoteam puts the figure at 10 to 15 per cent. Other commentary cites 20 to 30 per cent as an industry estimate. One comparison found three services slightly cheaper in the sovereign environment than outside it. The spread between the low and the high of those accounts is wide enough that none of them can be adopted as the number, and the range should be carried as a range.
Microsoft's sovereign add-on pricing is less clear still: the sources conflict between no extra list cost and a range of 15 to 20 per cent, and the position is that it is not reliably public. It is to be noted that transposing any of these figures to the Gulf would be an extrapolation rather than a finding, and that the research supporting this edition contains no basis for it. The European premium tells a Gulf buyer that a sovereign premium exists and is likely to be measured in low tens of per cent somewhere. It does not tell that buyer what a Gulf sovereign tier will be priced at.
Construction cost is the one dimension on which the Gulf position is firmly quantified, and it favours the region. Turner and Townsend's tenth Data Centre Construction Cost Index places the Middle East at USD 9.93 per watt, the cheapest region surveyed.
| Region or country | Data centre construction cost, USD per watt |
|---|---|
| Middle East | 9.93 |
| North America | 12.06 |
| Asia | 12.10 |
| Europe | 12.76 |
| Saudi Arabia | 10.80 |
| Qatar | 9.80 |
| United Arab Emirates | 9.20 |
Source: Turner and Townsend, tenth Data Centre Construction Cost Index. Regional and country figures are drawn from the same index and are not additive. The same source forecasts Middle East construction cost inflation at 5.1 per cent in 2027, the second-highest of any region.
Two observations follow. The first is that the Gulf's cost advantage in construction is real but narrow in absolute terms: USD 9.93 per watt against USD 12.76 in Europe is a difference of a little under a quarter, and the forecast of 5.1 per cent Middle East construction cost inflation in 2027, the second-highest of any region, erodes it. The second is that the direction of travel globally is against the buyer. JLL's global shell-and-core estimate stands at approximately USD 11.3m per MW in 2026 against USD 7.7m per MW in 2020. That estimate is global and not GCC-specific, and should not be read as a Gulf figure, but it indicates the trend that any sovereign capacity commitment made in 2027 is being made into.
GlobalData and MEED track GCC data centre investment above USD 93bn across more than 174 active and planned projects. This is an announced pipeline. It is not committed and it is not spent, and the distinction is not pedantic: a pipeline figure of that size, published without a committed subset, tells a buyer about intent and nothing about delivered capacity. Arizton projects Middle East data centre market revenue of USD 7.61bn in 2025 rising to USD 20.74bn by 2031, with cumulative investment of USD 33.79bn for 2025 to 2030. Those are analyst projections rather than reported
figures, and they are not reconciled with the USD 93bn pipeline because the two measure different things.
It is also to be noted what is absent. No GCC-specific public cloud spending forecast from IDC, Gartner or Synergy was available for this edition. No published tender or contract value for government cloud in any of the six states was located. The combined effect is that a Gulf buyer cannot benchmark a sovereign cloud price against a comparable public transaction, because no comparable public transaction has been published.
The takeaway is that pricing a sovereign deployment in the Gulf is currently an exercise in first-principles costing rather than benchmarking. What is measurable is the construction cost per watt, which is the lowest of any region and rising at the second-fastest rate; what is measurable by proxy is a European sovereign premium somewhere in a disputed range from 10 to 30 per cent; and what is not available at any price is a published Gulf delta or a comparable contract value. Any business case that presents a Gulf sovereign premium as a single figure has invented it, and the absence of the number is a stronger negotiating fact than any estimate of it.
Four of the six Gulf states have published a classification or tiering scheme, and in each case it is the scheme rather than the privacy statute that decides where a workload may run. Placement is therefore a classification exercise first and a procurement exercise second.
The practical mechanism by which sovereignty binds a workload in the Gulf is classification. The United Arab Emirates publishes three sovereignty tiers and attaches named conditions to each. Qatar publishes five classification tiers for government entities and a minimum of four for others. Saudi Arabia operates a two-axis model in which the National Data Management Office sets sensitivity and the National Cybersecurity Authority sets the corresponding control level. Oman fences two of its levels inside the national border. In every one of those four cases the label attached to the data determines the hosting model, and the sequence that follows from this is fixed: classify, then select the control level, then select the region, then select the provider. Reversing that sequence is how workloads end up in the wrong place.
The UAE tier descriptions rest on a law firm summary dated 15 October 2026 rather than on the primary policy text. The exact NDMO classification labels are contested between sources. Oman's full level definitions and licence classes, and Qatar's claimed C4 cloud-migration exemption, were not retrieved. Each should be checked against the primary instrument.
| Jurisdiction | Scheme | Levels | What the top levels require |
|---|---|---|---|
| United Arab Emirates | National Cloud Security Policy v2.0, UAE Cyber Security Council | Three sovereignty tiers | Tier 2 applies where data is classified confidential or restricted: all consumer data stored strictly within UAE geographical boundaries and legal jurisdiction. Tier 3 applies above that: exclusive UAE jurisdiction, data and metadata not disclosed to any authority other than those explicitly approved by the UAE, all foreign access requests denied by default |
| Qatar | NCSA National Data Classification Policy V3.0, established May 2023 | Five tiers, C0 to C4, for government entities; a minimum of four tiers for non-government entities | Not fully retrieved from public material. One secondary summary states C4 is exempt from cloud migration; this is unverified |
| Saudi Arabia | NDMO national data classification plus NCA Cloud Cybersecurity Controls CCC-2:2024 | Four classification levels and four corresponding cloud control levels | Control Level 1 applies to top secret data and Control Level 2 to secret data. CCC-2:2024 covers government bodies and private owners and operators of critical national infrastructure |
| Oman | TRA cloud and data centre bylaw, Decision No. 1152/2/19/2024-20, September 2024 | Levels referenced to 4; full definitions not retrieved | Article 12 prohibits transferring Levels 3 and 4 data, explicitly including governmental and financial data, outside Oman without prior TRA approval |
The UAE scheme is the most directly actionable because it states conditions rather than principles. Tier 1, baseline secure cloud, requires foundational controls of all providers irrespective of data classification, which means no in-scope workload is exempt. Tier 2, partial sovereign cloud, applies where data is classified confidential or restricted and requires that all consumer data be stored strictly within UAE geographical boundaries and legal jurisdiction, with a disclosure protocol: on a valid foreign disclosure order the provider must make reasonable efforts to redirect the request to the consumer and must inform both the consumer and the UAE authorities. Tier 3, sovereign cloud, applies above confidential or restricted and places all
data under exclusive UAE jurisdiction, with consumer data and metadata not disclosed to any authority other than those explicitly approved by the UAE and all foreign access requests denied by default. The eligibility conditions to expect at Tier 3 are UAE-based key management, UAE-resident redundancy, UAE-licensed telecommunications routing and UAE-cleared personnel.
Two rules in the UAE scheme have disproportionate practical effect. The first is that where tiers overlap, the most stringent control applies, which means a composite workload inherits the obligations of its most sensitive component and cannot be averaged down. The second is that across all tiers, data location must be known at all stages, hosted encryption keys and their backups must be generated, stored and managed in the UAE, and the consumer must assess whether hold-your-own-key is required. That last obligation sits with the buyer and not with the provider, which makes it the condition most frequently left unperformed.
The Saudi model is used differently because it separates the two questions. The classification level answers how sensitive the data is; the CCC-2:2024 control level answers what must therefore be done. Level 1 attaches to top secret data and Level 2 to secret data, and the scope of CCC-2:2024 covers government bodies and private owners and operators of critical national infrastructure, which means a private enterprise outside that definition is not within the controls at all and is governed instead by its sector regulator and by the Personal Data Protection Law. Establishing whether an entity is in scope is therefore the first question in the Kingdom, and it precedes classification.
Qatar's scheme is used as a gate. Five tiers for government entities and a minimum of four for non-government entities, under the National Data Classification Policy V3.0 established in May 2023, sit alongside the MCIT Cloud First Policy, document P005 v1.0.0 of April 2024, which applies to all government agencies. The combination means a Qatari government workload is pushed towards cloud by policy and constrained by tier, and the claimed exemption of C4 from cloud migration, if it is confirmed, is the point at which the two instruments meet. It is unverified and should not be assumed in either direction.
Oman's scheme is used as a boundary rather than a gradient. The operative provision is Article 12 of the cloud bylaw: Levels 3 and 4, which explicitly include governmental and financial data, may not leave Oman without prior TRA approval. Because the full level definitions were not retrieved, the practical approach is to treat governmental and financial data as fenced and to establish the status of anything adjacent to those categories with the regulator before designing a cross-border flow. The constraint is sharpened by the absence of any hyperscaler region domiciled in the Sultanate as at May 2026.
The takeaway is a single ordering, applied per jurisdiction and per entity. Establish whether the entity is in scope of the national cloud instrument. Classify the data under that jurisdiction's own scheme and not under a group standard. Read the control level or tier that the classification compels. Only then test which live regions and licensed providers can satisfy it. Where the classification labels or level definitions are contested or unretrieved, as they are in three of the four schemes above, the classification is provisional until confirmed against the primary text, and a provisional classification should not be used to justify the cheaper tier.
Only three dates in the Gulf's next three years bind anybody. The rest are vendor targets, and the distinction between the two is the most useful line a buyer can draw across a planning horizon.
A calendar of Gulf digital sovereignty looks crowded and is not. Of the dated events between 2026 and 2028, three impose obligations on identifiable parties: Oman's Personal Data Protection Law becoming fully enforceable, the UAE National Cloud Security Policy v2.0 binding providers and specified consumers, and the Saudi local content weighting taking effect in the financial evaluation of government IT tenders. The remainder are commercial targets set by vendors, which may move without anybody being in breach. The table separates the two, because a plan that treats a vendor target as a deadline will be built on the wrong foundation.
| Date | Event | Binding on | Nature |
|---|---|---|---|
| 5 February 2026 | Oman PDPL fully enforceable | All controllers and processors in Oman | Regulatory |
| In force 2026 | Oman TRA cloud and data centre bylaw. Level 3 and 4 data cannot leave Oman without TRA approval | Providers and data centre operators in Oman | Regulatory |
| Date | Event | Binding on | Nature |
|---|---|---|---|
| From 15 October 2026 reporting | UAE National Cloud Security Policy v2.0 | All providers serving the UAE; federal ministries and authorities; non-government critical information infrastructure; Emirate government entities where no Emirate CII programme exists | Regulatory |
| Q4 2026 target | Microsoft Azure Saudi Arabia East region | Nobody. A vendor target | Commercial |
| December 2026 target | AWS Saudi Arabia region | Nobody. A vendor target | Commercial |
| 1 April 2027 | Saudi local content weighting in the financial evaluation of IT service contracts | Bidders for Saudi government IT tenders of SAR 10m or more | Regulatory |
| Mid-2027 | HUMAIN phase two, 70 MW of Vera Rubin | Nobody. A vendor target | Commercial |
| 2028 | AWS and HUMAIN AI Zone, up to 50 MW available | Nobody. A vendor target | Commercial |
| Not dated | Saudi Global AI Hub Law | Nobody. A draft, no commencement date published | Draft |
One further compliance deadline for the UAE federal data protection law circulates in secondary commentary. It could not be verified against any primary instrument and is therefore excluded from the calendar above rather than carried with a caveat.
The regulatory entries have a common character: each of them changes what a named class of party must do, and each of them is traceable to a published instrument. Oman's position is the heaviest because two instruments land close together. The Personal Data Protection Law, Royal Decree No. 6/2022 published on 13 February 2022, with Executive Regulations by Ministerial Decision No. 34/2024 published on 4 February 2024, became fully enforceable from 5 February 2026 and binds all controllers and processors in the Sultanate. The TRA cloud and data centre bylaw, Decision No. 1152/2/19/2024-20 of September 2024, took effect the day after Gazette publication and binds providers and data centre operators, with Levels 3 and 4 data prohibited from leaving without prior TRA approval. A buyer operating in Oman is therefore subject to a new privacy regime and a new provider licensing regime within the same period.
The Saudi entry is the one with the clearest commercial consequence. From 1 April 2027 local content carries weight in the financial evaluation of IT service contracts for tenders of SAR 10m or more, binding bidders rather than operators. It is a procurement rule and not a privacy rule, and it is the only date in the
table that changes who wins work rather than what work may lawfully be done.
The UAE entry is the broadest in scope. From the reporting of 15 October 2026, the National Cloud Security Policy v2.0 binds all providers serving the UAE, federal ministries and authorities, non-government critical information infrastructure entities, and Emirate government entities where no Emirate critical information infrastructure programme exists. The last of those four is the one most often missed, because it makes the applicable supervisor depend on whether the relevant Emirate runs its own programme.
The calendar contains no commencement date for the Saudi Global AI Hub Law, because none has been published. The draft was published by the Communications, Space and Technology Commission on 14 April 2025 with consultation closing on 14 May 2025, and it would create Private, Extended and Virtual Hubs, which is to say data embassies operating under foreign law. It remains a draft. It is the largest single uncertainty in the Gulf's three-year picture and it has no date.
The calendar also contains no enforcement dates, because no named published fine or enforcement decision under any of the six GCC data protection laws could be located. It contains no commissioning date for Stargate UAE, whose third-quarter 2026 target has passed with no official confirmation located. It contains no date on which the status of the UAE federal Executive Regulations will be resolved, because their status could not be confirmed. And it contains no energised capacity milestone for any sovereign artificial-intelligence programme in the region, because energised as opposed to announced capacity is not published for HUMAIN or for Stargate UAE.
The takeaway is that a three-year sovereignty plan for the Gulf has three fixed points and a great deal of vendor-dependent optionality around them. The three regulatory dates can be planned against and should be. The commercial targets should be carried with a stated fallback, since two of them concern Saudi regions that were still unlaunched at the latest reported position and one concerns a 2028 availability figure. The undated items, and above all the draft hub law, warrant monitoring of the primary instrument rather than a place in a plan.
Every sovereign artificial-intelligence platform in the Gulf runs on foreign silicon moving under foreign export licences whose conditions are not published. Sovereignty over data is achievable in the region. Sovereignty over compute is not, and a buyer should price the difference rather than ignore it.
The uncomfortable fact underneath the Gulf's sovereignty programme is that its most advanced layer is its least sovereign. Data residency can be engineered: it is a matter of region selection, key custody and routing, and the UAE National Cloud Security Policy v2.0 shows that it can be specified in writing and tested. Compute cannot be engineered in the same way, because the hardware on which every announced sovereign artificial-intelligence platform in the region depends is designed and manufactured outside the region and moves under export authorisations granted outside the region. The control a buyer has over the first is substantial. The control it has over the second is, on published evidence, none.
On 19 November 2025 US Commerce authorised advanced chip exports to G42 in the UAE and HUMAIN in Saudi Arabia, each up to the equivalent of 35,000 NVIDIA GB300, conditioned on security and reporting requirements monitored by the Bureau of Industry and Security. The licence numbers are not public. The full condition list is not public. The delivery schedules are not public. What is public is the ceiling and the fact of conditionality. On 10 July 2026 the UAE was reported to have moved from Country Groups D:3 and D:4 to A:5, clearing G42, Core42 and MGX for licence-free purchases; Saudi Arabia was not covered and remains on the November 2025 licence basis.
Three consequences follow for a buyer. The first is that the compute supply for the Kingdom's flagship programme is governed by conditions that cannot be
read, which means the risk cannot be assessed, only acknowledged. The second is that the two largest Gulf markets now sit on different footings, with the UAE's named entities purchasing without a licence and Saudi Arabia's purchasing under one. The third is that both footings are revocable by a foreign administrative decision, and the two dates in this paragraph are themselves the evidence that such decisions change: the position in July 2026 was not the position in November 2025.
The 10 July 2026 Country Group reassignment rests on secondary reporting; the primary Federal Register text was not retrieved. The conditions attaching to the 19 November 2025 authorisations are not public and no claim about their content is made here.
The asymmetry between those two lists is the honest statement of the Gulf's sovereignty position in 2027. The first list is specific, written and testable. The second is a set of dependencies whose terms are
withheld from the party bearing the risk. It is to be noted that this is not a criticism of the Gulf programmes: no jurisdiction outside the small number that manufacture advanced accelerators is in a different position, and the export control decisions of November 2025 and July 2026 placed the UAE and Saudi Arabia among the most favourably treated non-manufacturing jurisdictions in the world. The point is narrower. A sovereignty claim that covers the data layer should not be read as covering the compute layer, and the published record does not support extending it.
The takeaway is that the two layers should be assessed and documented separately. For the data layer, the controls are specifiable and the right course is to specify them: location at all stages, key custody, hold-your-own-key where the assessment requires it, and the provider's documented response to foreign legal process. For the compute layer, the only defensible treatment is to record the dependency explicitly, to note that the licence conditions and delivery schedules are not public, and to hold a fallback for the case in which the authorisation basis changes. A plan that assumes the compute dependency away is assuming away the only part of the stack whose governing terms the buyer has never seen.
India has built the opposite of the Gulf model: a phased national statute with no general localisation mandate, and subsidised common compute at a published price. The contrast shows that the Gulf has chosen sovereignty by infrastructure where India has chosen sovereignty by rulemaking.
India is included here for one reason: it is the nearest large jurisdiction that has pursued digital sovereignty by a different route, and the difference is instructive rather than competitive. Where the Gulf states have legislated residency and classification and then built regions and clusters to satisfy them, India has notified a single national statute with a phased commencement, has declined so far to notify any specific localisation mandate, and has put subsidised compute into the market at a price that is at least partly published. The contrast is between sovereignty expressed as infrastructure and sovereignty expressed as rulemaking, and the Gulf buyer's interest in it is in seeing which of the two produces a position a buyer can plan against.
The Digital Personal Data Protection Rules 2025 were notified on 13 November 2025 as G.S.R. 846(E), although the Press Information Bureau release carries the date 14 November 2025 and the two dates are unreconciled in the sources. Commencement is phased. The Data Protection Board provisions took effect immediately. The consent manager rules follow at twelve months, in November 2026. The duties of principal data fiduciaries follow at eighteen months, which Rule 1(2) puts at 13 or 14 May 2027, the ambiguity tracking the notification date conflict.
Designations of Significant Data Fiduciaries under Section 10 are still pending. The additional duties that attach on designation include a resident data protection officer, an independent auditor and an annual impact assessment. On localisation, the position is that fiduciaries may be subject to government directions on restricted categories including local storage, but no specific localisation mandate has been notified. That is the central structural difference from the Gulf: the power to require localisation exists and has not been exercised, whereas in Oman and Kuwait the requirement is already written into the cloud instrument.
The IndiaAI Mission reports 38,231 graphics processing units onboarded to common compute, per a Lok Sabha reply dated 25 March 2026, with a further 20,000 announced under AI Mission 2.0 in February 2026, completion not confirmed. The price is where the comparison becomes genuinely useful. IndiaAI common compute is offered at roughly Rs 65 per GPU-hour according to the same 25 March 2026 Lok Sabha reply, while other sources cite Rs 115.85 to Rs 150. Those accounts conflict and are not reconciled.
Even allowing for that conflict, the contrast with the Gulf is stark on a point of kind rather than degree. India has a published figure for onboarded compute, stated to a parliament on a date, and at least one published unit price. For HUMAIN and for Stargate UAE, no published operating megawatt figure and no energised GPU count exists, and no GCC vendor has published a sovereign-cloud price delta at all. A buyer in the Gulf comparing options can therefore establish what Indian common compute costs per GPU-hour within a disputed range, and cannot establish what Gulf sovereign capacity costs within any range.
The Indian GPU-hour figures are unreconciled across sources and the 38,231 and 20,000 GPU figures rest on a parliamentary reply and an announcement respectively, with completion of the second not confirmed. The notification date conflict of 13 against 14 November 2025 propagates into the Rule 1(2) commencement date and is left unresolved here.
Three things follow for a Gulf reader. The first concerns transparency as a procurement asset. India's phased commencement dates, its pending designations and its compute volumes are all stated publicly, with dates and with an identifiable source, and the conflicts in the record are visible precisely because so much of it is published. In the Gulf the most significant figures, including energised capacity, sovereign price deltas, government cloud contract values and the exact classification labels in two schemes, are not published at all, so there is nothing for sources to conflict about. A buyer should not mistake the absence of contradiction for the presence of certainty.
The second concerns the shape of the obligation. India's statute assigns duties and defers them, with the
heaviest set, the principal data fiduciary duties, arriving at eighteen months, and with the designation of significant fiduciaries still to come. The Gulf's instruments assign duties by classification and apply them on commencement, as Oman's 5 February 2026 enforceability and the UAE Cloud Security Policy's immediate scope both demonstrate. The Indian model gives longer notice of a less specified obligation. The Gulf model gives shorter notice of a more specified one. For planning purposes the Gulf model is easier to engineer against and harder to defer.
The third concerns localisation itself. India's choice to hold the localisation power in reserve, exercisable by direction on restricted categories, means an Indian deployment carries a contingent liability rather than a fixed constraint. The Gulf's choice to write residency and classification into instruments means a Gulf deployment carries a fixed constraint and no contingent liability of that kind. Neither is better in the abstract. The fixed constraint is more expensive to satisfy and cheaper to plan around; the contingent one is cheaper today and cannot be sized.
The takeaway, for the Gulf and not for India, is that the region's weakest point is not its rules but its disclosure. The rules are specific, dated and increasingly actionable, and the UAE tiering scheme is the clearest sovereignty taxonomy published anywhere covered by this edition. What is missing is the operating data: energised capacity, sovereign price deltas, contract values, enforcement decisions and operator-access terms. Where India's record can be checked and found to conflict, the Gulf's frequently cannot be checked at all, and the appropriate response in a workload decision is to treat an unpublished figure as unknown rather than as favourable.
The instrument is Circular C 14/2021, effective 15 July 2021, binding all banks operating in the UAE on a solo and group-wide basis. One exception is written in. Article 6.2 permits branches of foreign banks, subject to approval, to comply “by retaining a copy of the Master System of Record, updated on at least a daily basis, within the UAE”. It concedes the primary location for a foreign branch; it does not concede the in-country copy, and it does not operate without approval. Article 6.3 adds that customer Confidential Data “must not be shared outside the UAE without Central Bank approval and obtaining prior written consent from the customer”. Above the location rule sits an approval gate: Article 8.1 requires prior non-objection before outsourcing any material activity, and Article 3 requires a register of every arrangement, recording whether Confidential Data is involved. That register is the
artefact a buyer will be asked for. The transition is closed, Article 13.3 having required all agreements, including pre-existing ones, to comply “by no later than 31 December 2023”.
What the rules require of a banking, health, energy, government, telecommunications or logistics workload, and the sequence of questions by which any workload is placed.
Financial services is the one Gulf sector in which a regulator has reduced sovereignty to a sentence about geography. The UAE writes that sentence hardest, Saudi Arabia writes it as a default with an approval route out, and Bahrain has deleted the sentence it once had.
Banking is where Gulf sovereignty rules are most legible, because one regulator named a place rather than described a control. Article 6.1 of the Central Bank of the UAE Outsourcing Regulation for Banks requires that “the Master System of Record, which includes all Confidential Data, is continuously maintained and stored within the UAE”. That is not a risk expectation or a maturity level but a flat obligation, and its object is defined widely: the Master System of Record is “the collection of all data, including Confidential Data, required to conduct all core activities of a Bank”. A buyer wanting to know what sovereignty costs here should read that definition against their own application inventory, because most core systems fall inside it.
The Saudi position is widely described as absolute residency. On the primary text it is not. Section 3.4.3 of the Saudi Central Bank Cyber Security Framework, version 1.0 of May 2017, provides that “in principle only cloud services should be used that are located in Saudi Arabia, or when cloud services are to be used outside Saudi Arabia that the Member Organization should obtain explicit approval from SAMA”. The qualifier and the express alternative both matter. The same clause requires approval “prior to using cloud services or signing the contract with the cloud provider”, an
unusually early gate: approval precedes signature, not deployment.
The sharper Saudi gate is not the location clause. Paragraph 42 of the Rules on Outsourcing requires written non-objection for any outsourcing to a provider located overseas, and limb (c) requires the “Rationale for outsourcing (including why it cannot be done within KSA)”. The applicant must argue a negative. That is a different species of obligation from a prohibition, and it explains the published pattern: Saudi institutions on private cloud, or buying public cloud through a mandated in-Kingdom reseller, rather than seeking approvals they would have to justify.
The Framework uses “should” rather than “must” and is enforced through maturity assessment, the published timetable requiring full compliance by the end of 2019 at minimum maturity level three. The contrast with the UAE's flat “must” is significant.
Bahrain runs the other way, deliberately. Rulebook module OM-2.1.7 provides that “Prior CBB approval is required on any outsourcing to a third-party outside Bahrain (excluding cloud data services)”, with post-notification within five working days of signature sufficing for a “third-party cloud data services provider inside or outside Bahrain”. Cloud is carved out of offshore approval and offshore hosting is permitted in terms. The archived predecessor, in force to 31 December 2019, said the opposite: OM-3.9.2 required that service providers handling customer information “must be licensed by the CBB and located in Bahrain”. A jurisdiction that removes its own residency rule and leaves the record visible is making an argument, not merely deregulating.
| State | Instrument | Location obligation on a banking workload | Approval |
|---|---|---|---|
| UAE | Outsourcing Regulation for Banks, C 14/2021, effective 15 July 2021 | Master System of Record and all Confidential Data must be maintained and stored in the UAE. Foreign branches may instead hold a daily-updated copy in country, with approval | Prior non-objection for material outsourcing. Separate approval plus customer consent to share Confidential Data abroad |
| Saudi Arabia | Cyber Security Framework v1.0, May 2017, s.3.4.3; Rules on Outsourcing, Chapter V | In-Kingdom by default, “in principle”, with explicit approval as the route offshore | Approval before using cloud or signing the contract. Written non-objection for all overseas outsourcing, with a case for why the work cannot be done in the Kingdom |
| Bahrain | CBB Rulebook OM-2.1.7, added January 2020, chapter replaced July 2022, amended April 2023 and May 2026 | None. Offshore hosting of banking data is expressly permitted | Cloud data services carved out of prior approval. Post-notification within five working days of signature |
| State | Instrument | Location obligation on a banking workload | Approval |
|---|---|---|---|
| Qatar | Technology Risks Regulation for Banks; a 2024 cloud instrument is reported | Not established. No QCB instrument could be read | Not established |
| Kuwait | Cybersecurity Baselines for the Kuwaiti Banking Sector, v1.0, January 2020 | Not established. The cloud security section was not retrievable | Not established. No cloud-specific CBK circular was located |
| Oman | Circular BM 1185, Rules for Availing Cloud Services by Licensed Institutions, 8 May 2022 | Not established. The text was not retrieved | Not established |
The last three rows record a research limit, not a regulatory vacuum. Kuwait's baselines carry sections titled “Third Party Outsourcing” and “Cloud Security” at pages 42 and 43 of 55, and the retrieved file truncated at page 37. Oman's BM 1185 is corroborated by name and date, but not one operative word was retrieved. Unestablished is not the same as permissive. The UAE and Bahrain rulebooks were retrieved in full.
Named moves are thinner than the coverage suggests. A Bahraini bank announced on its own newsroom on 9 February 2021 that it had migrated its international remittance system to in-Kingdom cloud regions, with no figure published. First Abu Dhabi Bank's data centre migration, announced 6 March 2024, is framed around sovereign controls to “help achieve data sovereignty and meet its UAE regulatory obligations”, and Mashreq's October 2024 selection of a cloud-at-customer database was justified partly by the need to “comply with local data residency regulations”. Those two, and the region's one completed core banking migration, were all announced by vendors or integrators rather than by the banks.
A widely quoted figure, that about 92 per cent of a UAE group's core banking systems are in cloud, comes from an interview page carrying no date, and should not reach a board paper undated.
What a buyer will test is enforcement, and the record does not support it. Across the six states no published penalty on a bank by a Gulf financial regulator is expressly grounded in outsourcing, cloud, residency or technology governance. The UAE enforcement listing runs to roughly sixty releases between November 2019 and December 2025 and not one cites those grounds. Saudi architecture explains part of this: the Instructions for Publishing Banking Penalties of 17 December 2017 require banks to disclose punitive decisions by category, and the categories are supervisory instructions, customer protection, due diligence and cash-machine performance. There is no technology or outsourcing category, so a residency breach would be invisible.
The takeaway is uncomfortable for both sides. The rules are the region's most prescriptive and the ceilings are rising: the UAE maximum on a licensed financial institution moved to not more than AED 1,000,000,000 under Article 168 of Federal Decree-Law No. 6 of 2025, in force 16 September 2025, from AED 200,000,000. Saudi Arabia's most technology-relevant statutory figure is the banking-secrecy limb of the Banking Control Law, a fine not exceeding SAR 20,000. In a sector with a flat location mandate, a rising ceiling and no published case, the risk is not the fine. It is the power, held by both regulators, to order an outsourcing arrangement terminated.
Only the United Arab Emirates has a health-specific data localisation rule. Elsewhere in the Gulf health data is reached as a sensitive category under the general data law, and what is required is a permit or a transfer condition rather than a location.
In healthcare a single instrument carries almost the whole regional burden, and buyers routinely attribute its effect to states with no equivalent. Article 13 of UAE Federal Law No. 2 of 2019 on the Use of Information and Communication Technology in Health Fields provides that “the health information and data related to the health services provided in the State may not be stored, processed, generated or transferred outside the State, unless in the cases defined by virtue of a decision issued by the Health Authority in coordination with the Ministry”. Three features deserve attention before any architecture is drawn. It covers four verbs, so a computation that never moves the record still engages it. Its trigger is the place of service, not the patient's nationality. And the gateway is a decision of a health authority, defined to include any local governmental health authority, so an emirate-level body can open it.
The law issued on 6 February 2019, published in Official Gazette 647, in force three months after publication under Article 31. Article 2 applies it “including the Free Zones”, closing the route most often assumed open. The penalty is specific: Article 24 provides that a violation of Article 13 “shall be punished by a fine of not less than AED (500.000) five hundred thousand and not more than AED (700.000) seven hundred thousand”. Article 25 adds a disciplinary track against institutions running to AED 1,000,000 and cancellation of authorisation to use the federal central system. Article 20(1)(a) requires health data kept “no less than (25) twenty-five years” from the last procedure.
The exemptions sit in a Ministry of Health and Prevention resolution, number 51 of 2021, issued 28 April 2021 and effective 16 May 2021 on publication in Gazette 702. It is not Cabinet Resolution No. 32 of 2020, which is the implementing regulation and carries no exemption list. The categories cover patients treated abroad, overseas laboratory samples, scientific research, insurance and claims management, cooperating authorities, consumer health devices, pharmacovigilance, remote medical services, and transfers the health authorities or the patient request, each conditioned typically on written consent and encryption. The commercially significant point is that several of those lawful routes still require a copy of the health information to be kept inside the UAE.
Two reputable published tabulations of this resolution give eight categories and ten respectively, and neither is the gazette. That conflict is unreconciled and should be resolved against the instrument before an architecture depends on a particular category. Secondary summaries giving the Article 13 range as up to AED 2,000,000 are wrong against the ministry's own text, which was retrieved in full.
No published rule requires the health data of Saudi patients to be hosted inside the Kingdom. The Ministry of Health's Saudi Health Information Exchange Policies, document IS0303 version 1.0, was retrieved in full and searched: it contains no hosting, localisation, residency, cloud or data-centre-location provision of any kind. That is a reliable negative on the primary instrument most likely to carry one. What it does contain is the opposite duty, that “Data SHALL NOT be deleted at any time from the Saudi Health Information Exchange”.
Health data is reached instead as sensitive data under the Personal Data Protection Law, Royal Decree M/19 of 16 September 2021 as amended in March 2023, in force 14 September 2023. On location the regulator's own guidance works an example in which data of people living in the Kingdom sits on a cloud solution “hosted in the USA”, and concludes that “such storage of personal data will be regulated by the PDPL” even though the data is held abroad. A regulator publishing that example is not operating a localisation regime.
Draft amendments that would require controllers to store personal data within the Kingdom went to consultation from 6 October to 5 November 2026, with entry into force stated as sixty days after gazette publication if adopted. That rests on law-firm reporting, the draft text itself not having been retrieved, and is not settled.
Qatar points the other way. Article 15 of Law No. 13 of 2016 provides that “the Controller shall be forbidden from taking any decision or measure that may limit the Cross-Border Data Flow”, with that flow defined as access, use or storage “without restrictions of the State's borders”. What Qatar requires is permission, not place: Article 16 classes health data as of a special nature, processable “only after obtaining the permission from the Competent Department”. Vendor material asserting a Qatari health residency requirement grounded in this law should be rejected, because the law contains none.
Oman also requires a permit. Article 5 of Royal Decree 6/2022 prohibits processing health data “except after obtaining a permit” from the ministry, which has 45 days to decide, silence counting as rejection. Article 23 on cross-border transfer is permissive. Royal Decree 68/2026, in force from 7 September 2026, extended the law to processing “inside or outside the Sultanate”. Bahrain treats health data as sensitive and restricts export by destination adequacy under Article 12 of Law No. 30 of 2018, with criminal exposure: unlawful transfer carries imprisonment up to one year and a fine of BD 1,000 to BD 20,000. Kuwait has no general data protection law and no health-specific hosting instrument was located, which is itself the finding.
A regional provider faces a contradiction no architecture resolves. The UAE requires health data kept for not less than twenty-five years. Saudi Arabia's exchange policy says its data shall never be deleted. Oman's new Article 15 requires that the controller and processor “shall erase personal data immediately upon the end of the purpose of its processing”. These cannot all be satisfied by one retention policy, so the regional estate must be partitioned by jurisdiction for reasons unconnected to hosting location.
The takeaway is that healthcare has the region's sharpest statutory language and its least visible compliance. No enforcement action, prosecution or published decision was found against any Gulf entity for a health data hosting or residency breach, under any of the five penalty regimes above. The obligations exist and are unlitigated in public, and the absence of cases may reflect non-publication rather than non-enforcement. For a buyer the operative risk is not the fine. It is that the UAE rule bites on processing and generation as well as storage, so an analytics or model-training workload touching Emirati clinical data engages Article 13 even if the record never leaves the country.
No Gulf state has an energy-sector hosting instrument. Operators are caught by general critical-infrastructure frameworks, no state publishes who is designated, and the in-Kingdom cloud clause the market still cites was removed from the Saudi controls in 2024.
The position in energy is that the sector has no rules of its own. Every hosting obligation traced here reaches an oil company, a refinery or a water and electricity utility through a general cybersecurity or cloud framework, by virtue of being critical national infrastructure, a government entity, an affiliate of one, or a holder of government data. That absence is the finding, and it is not for want of power. Abu Dhabi's Department of Energy, established under Law No. 11 of 2018, publishes its own governing framework and its own responsibilities, which include “Regulating all aspects of the Energy Sector through developing policies, codes, standards, rules, resolutions”. Neither “cyber” nor “data” nor “information security” appears among them. The power exists; no code has been made under it.
The most consequential thing a buyer can learn from this sector is that the Saudi in-Kingdom cloud hosting requirement, still the most cited hosting clause in the Gulf, has been deleted. The National Cybersecurity Authority's Cloud Cybersecurity Controls are now at edition CCC-2:2024, and the Authority states that they “have been updated to reflect changes related to data localization requirements”. Annex D records two deletions in terms. Subcontrol 2-3-P-1-10 formerly read “Provide cloud computing services from within the Kingdom, including systems used for storage, processing, and disaster recovery centers”. Subcontrol
2-3-P-1-11 covered “monitoring, and support”. Both are gone.
The stated rationale is a transfer of competence, not a relaxation: “Controls related to data localization have been transferred from the document to the National Data Management Office (NDMO) at the Saudi Data and Artificial Intelligence Authority”, and entities “must refer to the National Data Management Office regarding data localization before taking any action”. The obligation has moved rather than vanished. Any advice asserting that the Saudi cloud controls mandate in-Kingdom hosting by classification level is out of date against the published text. The successor rules sit with a body whose own site rejects automated retrieval; they were not obtained and are not characterised here.
Two corrections follow. The four levels are data classification levels, not provider tiers: level 1 applies to data classified as top secret through to level 4 public, with the highest level governing a mixed dataset. And what survives on location concerns people, not data. Control 1-4-P-1-1 requires that “Positions of cybersecurity functions in CSP's data centers within the Kingdom must be filled with qualified and suitable Saudi nationals”, and the Essential Cybersecurity Controls, now at ECC-2:2024, require managed cybersecurity operations centres using remote access to sit entirely inside the Kingdom. Sovereignty expressed as staffing and operations location is a different commercial problem from data residency.
Scope turns on designation, and designation is almost everywhere invisible. The Saudi controls bind “all private sector entities owning, operating or hosting Critical National Infrastructures (CNIs)”, and no Saudi register or sector list was located: the perimeter is self-identifying. Oman holds the power expressly, Royal Decree 64/2020 requiring the Cyber Defence Centre to prepare a classification of cyber infrastructure and identify the relevant sectors and entities, and publishes no output. Bahrain's strategy for 2025 to 2028 organises itself around protecting critical national infrastructure and names no sector at all. Dubai is the half-exception: its Electronic Security Centre publishes ten critical sectors, among them “Oil & Gas” and “Electricity & Water”, labels them “Examples of Critical Sectors”, and does not publish the roster of designated organisations. Qatar's position could not be established, its cyber agency's portal returning an empty shell.
Two states do reach a hosting result, and neither does it through energy law. Kuwait's Cloud Computing Regulatory Framework, version 2.4, carries the only express private-sector residency prohibition found in the region. Clause 4.2.1.1 requires subscribers to ensure nothing is hosted or stored outside Kuwait “for any purpose, or in any format, whether temporarily or permanently” where the data falls in the third or fourth classification tiers, or is personal data held by government bodies, private companies or service providers. Oman reaches a similar result by consent: Article 4(5) of the Cyber Defence Centre system requires approval of the Centre before storing any sensitive data, or processing it, outside the Sultanate, a duty confined neither to personal data nor to government bodies.
The published deployment record points towards the hyperscaler stack, bought through a national intermediary. The Emirati national oil company's artificial intelligence joint venture announced on 10 March 2025 a contract of USD 340 million to deploy its agentic platform across the upstream value chain, stating that it “will integrate best-in-class technologies, including Azure cloud stack, agentic AI platform functionality, Open Subsurface Data Universe framework (OSDU), and OpenAI models”. On the Saudi side a non-binding memorandum of 12 February 2026 lists among its workstreams “Digital Sovereignty and Data Residency: Exploring the development of a roadmap for deploying solutions on the Microsoft cloud, enhanced with sovereign controls”, including “meeting national data residency requirements”. No figure is attached to it.
The structural move is consolidation. On 1 November 2026 the Saudi national oil company closed its acquisition of a 49 per cent stake in the sovereign-fund artificial intelligence vehicle, contributing approximately SAR 9.9 billion in cash in 2026 against a stated plan to deploy three gigawatts of data centre capacity by 2030. Two days later the exclusive in-Kingdom reseller through which Saudi customers must buy one hyperscaler's services was folded into that same vehicle.
The takeaway is that in this sector the binding constraint is not a residency clause but a designation the buyer cannot see and a counterparty list that is narrowing. No published penalty or enforcement action touching Gulf energy-sector hosting or data handling was located, and the Saudi authority's express powers
to identify cybersecurity violations date only from Royal Decree M/117 of 22 December 2024, with no outcome yet published. Six of the nine relevant regulators could not be reached, so the structural point is safe to state and the absolute claim is not. Diligence in energy runs to the designation question first, the intermediary second, and the instrument last.
Gulf governments are the largest buyers of sovereign cloud and the least regulated users of it. The UAE expressly exempts government data from its only data protection law, no federal cloud or hosting instrument exists, and the binding Dubai rules are not published.
The striking fact about public-sector workloads is that the state has largely declined to regulate itself by residency rule. Article 2(2) of UAE Federal Decree-Law No. 45 of 2021 provides that the law shall not apply to, among other things, “Government Data” and “Governmental entities which control or process Personal Data”. Government data is excluded as a category and government entities are separately excluded as controllers and processors, so the cross-border provisions of the federal personal data regime do not reach a government entity's hosting decisions at all. The boundary is drawn at corporate form: Article 1 defines an establishment to include “companies wholly owned by the federal or the local government, or in which they are shareholders”. State-owned companies are in scope; government entities are not.
The law issued on 20 September 2021 and comes into force on 2 January 2022. Article 28 required the executive regulations within six months of promulgation, and Article 29 starts the compliance clock for controllers and processors only from the date those regulations are issued. No primary evidence that they have been issued was found, and four primary indications are consistent with non-issuance, among them official guidance updated 4 December 2025 still describing the federal data regulator's functions in the future tense, and a regulator with no reachable website. The position is unconfirmed either way rather than positively established.
The federal government cloud is a platform, not a policy. The complete publications catalogue of the UAE's digital government authority was enumerated across its own guidelines, strategies, policies, reports, laws and standards filters, running to twenty titles, and it contains no cloud policy, cloud-first policy, cloud adoption standard, data classification standard or hosting instrument. That is a strong primary negative rather than an absence of looking, because the catalogue includes its own laws and standards categories. What exists instead is a federal network described as a “Sovereign Cloud”, with forty-six entities stated to
benefit from its cloud infrastructure, eleven central services and a published availability target of 99.9 per cent, and a service catalogue at version 1.1 of January 2025 offering eight cloud services priced “Free” to federal government entities.
Dubai legislates data and does not legislate location. Law No. 26 of 2015 Regulating Data Dissemination and Exchange in the Emirate of Dubai came into force on publication, applies to local and federal government entities and to persons producing or exchanging data relating to the emirate “including Special Development Zones and free zones, such as the Dubai International Financial Centre”, and classifies data into “Open Data” and “Shared Data” under a Dubai Data Manual adopted in coordination with the electronic security centre. All twenty-one articles were read. The law contains no in-country hosting requirement, no residency clause, no localisation provision and no reference to cloud computing of any kind. Its taxonomy governs who may receive data, not where it may sit.
Where a penalty power over government data exists, it is delegated and unpublished. Article 16 provides that a violator “will be punished by the penalties and measures prescribed by a resolution of the Chairman of the Executive Council”, and that resolution was not found. A residency rule could in principle live in the Dubai Data Manual, which was not retrievable. With the Information Security Regulation available only on email request, the honest finding for Dubai is that the position cannot be determined from public sources.
Abu Dhabi has gone furthest in commitment and no distance at all in instrument. Its Department of Government Enablement, established under Law No. (14) of 2023, announced a Government Digital Strategy 2025 to 2027 deploying “AED13 billion” and aiming “to achieve 100% adoption of sovereign cloud computing for government operations”. That is the strongest sovereign-cloud commitment of any UAE jurisdiction and it is a strategy target in a press release. The department's complete published legal inventory contains three laws and two policies, and no information
security standard, data classification policy, cloud policy or hosting requirement.
Saudi Arabia binds its government differently, through cybersecurity controls rather than data law. The Cloud Cybersecurity Controls apply to “any government agency in the Kingdom of Saudi Arabia inside or outside the Kingdom including ministries, authorities, establishments and other entities and their companies and sub-entities” as a cloud service tenant, and require “Continuous compliance with all laws, regulations, instructions, decisions, regulatory frameworks and controls, and mandates regarding cybersecurity in the Kingdom”. The sanction is non-compliance assessed by the regulator, not a fine in the instrument. A reported Cloud First Policy of around October 2021 could not be located or dated and is not relied on.
Kuwait is the clearest and the most misdescribed. Its Data Classification Policy, version 2.3, sets four tiers from “Public Data” to “Highly Sensitive Data” and imposes no obligation that any tier be stored inside Kuwait. What the tiers trigger is encryption in transmission, cataloguing and reporting: the regulator is to obtain “The documents that show the locations of stored data according to classification tiers”. Storage location is reported and audited, not prohibited. The onshore element sits in a separate instrument conditioning a provider's licence on hosting the two most sensitive tiers in Kuwait.
Bahrain's reported Cloud First Policy of April 2017 could not be confirmed, and a widely repeated claim that seventy per cent of the systems of seventy-two government agencies migrated to one hyperscaler within three years could not be verified as to figure, speaker, date or venue. It should not be used. Oman's ministry publishes the titles “Cloud and Hosting Services Standard” and “Cloud Governance Framework” in its own library index; their content and status were not obtained.
The takeaway is that government is the only sector where the regulator and the regulated are the same party, and the Gulf has resolved that by building rather than binding. No published penalty, enforcement action or named adverse finding against any Gulf government entity for a hosting or public-sector data breach was located. The discipline is policy, framework compliance, audit, reporting and platform licensing. For a commercial buyer the practical consequence is the reverse of what the rhetoric suggests: the enforceable chokepoint is upstream, in provider authorisation and platform registration, and it binds the supplier rather than the ministry.
The same framework also prohibits directly, which makes Kuwait the only Gulf state with both levers in one instrument. Clause 4.2.1.1 requires subscribers to ensure nothing is hosted or stored outside Kuwait “for any purpose, or in any format, whether temporarily or permanently” where the data falls in the third or fourth
tier, or is personal data of individuals held by government bodies, private companies or service providers. Clause 4.2.1.3 requires providers to disclose to the regulator the locations of their in-country data centres and the countries of any data centres used to process or transfer Kuwaiti subscribers' content.
Telecoms is distinctive because the sector regulator licenses the cloud provider as well as regulating the operator's data. Kuwait's framework is the clearest residency instrument in the Gulf, Saudi Arabia runs class licensing, and the UAE's position cannot be established from published sources.
Telecommunications inverts the logic of every other sector in this part. Elsewhere the obligation attaches to the regulated entity and asks where it may put its data. In telecoms it attaches upstream, to whoever wishes to sell hosting at all, and the lever is a licence rather than a prohibition. Kuwait states this most plainly. Clause 2.4 of its Cloud Computing Regulatory Framework, version 2.4, provides that the Authority licenses cloud service providers hosting third-tier and fourth-tier data who have data centres inside Kuwait, and clause 2.1 confines contracting for public-sector cloud services to those licensed by the Authority with a data centre and infrastructure inside the country. The residency outcome is achieved by controlling the seller.
One provision goes beyond location and deserves attention. Clause 4.3.3 prohibits a provider complying with a foreign jurisdiction's laws from handing subscriber content to any third party including “a local or foreign government, or public authorities”, with disclosure permitted only on official request from Kuwaiti authorities under Kuwaiti law. That is a blocking provision, not a residency provision, and it decides whether an in-country region actually answers the sovereignty question. The framework enters into force thirty days after gazette publication; that date was not retrieved, so the commencement date cannot be stated.
Saudi Arabia runs a class licensing regime for cloud, administered by the Communications, Space and Technology Commission rather than by the cybersecurity authority. The clearest published account comes from a hyperscaler's own technical documentation, which records that the Commission “has granted a Class C license (covering Qualifying, Class A, B, and C
licenses)” to its Dammam region, and that “the qualification is based on an assessment by the National Cybersecurity Agency (NCA) against the NCA's Essential Cybersecurity Controls (ECC) and the Cloud Cybersecurity Controls (CCC)”. The architecture is a telecoms licence whose conditions are set by cybersecurity controls, which is why the deleted localisation subcontrols discussed earlier matter here too.
That account is primary to the provider and secondary as to the framework. The Commission's own licensing register is client-rendered and returned no results when retrieved on 10 November 2026, and the Cloud Computing Regulatory Framework itself was not obtained. The class structure is reported at the strength stated and the operative wording of each class is not established. Vendor claims about Gulf cloud licence classes are among the least reliable material in this field and should be checked against the register before they are relied on.
The sharper Saudi gate is commercial rather than regulatory. The same documentation records that access to services from the Dammam region “is available only to KSA-based customers that have signed up to purchase cloud services through CNTXT”, and that customers with a Saudi billing address “are required to purchase Google Cloud services for the Dammam me-central2 region or any other Google Cloud region” through that reseller. A Saudi buyer's route to that provider is a single intermediary, for every region and not merely the in-Kingdom one. On 3 November 2026 that reseller was folded into the sovereign fund's artificial intelligence vehicle. Residency achieved by exclusive distribution is a different risk from residency achieved by statute, and it concentrates rather than diversifies.
Qatar's telecoms regulator publishes a Cloud Policy Framework dated 7 June 2022, describing it in its own words as defining “a comprehensive set of policy and regulatory recommendations” and establishing “a cloud-friendly environment for foreign and domestic investments in data centres and cloud services”. On that description it is a policy framework, not a licensing regime, and no cloud service provider licence or registration category appears in the regulator's published licensing navigation. The document sits behind a challenge gate and any residency clause in it is unverified. The absence of a licence category is an absence-of-evidence reading from a navigation tree, not a confirmed negative.
The United Arab Emirates is the significant gap, and it should be reported as one rather than filled. The national information portal records that the telecommunications regulator “developed the ‘UAE Information Assurance Regulation’” and “will designate the critical entities as per the UAE CIIP Policy to implement the IA Regulation”, applying its requirements “to the use, processing, storage and transmission of information or data”. That is a scope clause, not a location clause: it says the controls apply to storage, not where storage must be. The regulation, its date and any residency provision could not be obtained, the regulator domain returning empty bodies. A widely repeated claim that UAE policy requires internet-of-things data to be held in country is unverified.
What the sector sells is more revealing than what it is told. Gulf operators have become the hosts of record for sovereignty-constrained workloads. A Saudi operator subsidiary hosts one hyperscaler's Riyadh region, launched in October 2024 as part of a stated USD 1.5 billion investment. A separate Saudi cloud company, launched on 5 June 2022, is held by that operator at 55 per cent alongside a hyperscaler at 10 per cent, with capital of SAR 894 million. In the UAE an operator joint venture supplies the data centre capacity behind one hyperscaler's expansion, and in February 2024 another operator announced work with a hyperscaler to establish sovereign cloud capabilities “to meet the data residency and security requirements imposed on government entities and regulated industries in the UAE”.
Enforcement cannot be characterised. The only published telecoms penalty located is a sanction of 3.5 million Qatari riyals on Ooredoo Qatar by the regulator's Financial Sanctions Committee for “Violating Instructions Issued by the CRA”. The decision was not opened, so whether it touches data, hosting or cloud is unknown, and it must not be called a data penalty. The Saudi, Emirati and Kuwaiti enforcement registers were not reached, so the negative finding other sectors support cannot be stated here.
The takeaway is that telecoms is where Gulf sovereignty is becoming a product rather than a rule. In four of the six states the operator is simultaneously the regulated entity, the licensed host, the sovereign-cloud vehicle and, increasingly, the only route to a hyperscaler. A buyer evaluating a telecoms-hosted sovereign offer should therefore ask the licence question rather than the residency question: which class of licence the provider holds, what it was assessed
against, whether the purchase route is exclusive, and whether the contract carries a blocking provision of the Kuwaiti kind. Residency without a blocking provision answers where the data sits and not who can compel it.
No Gulf aviation regulator and no Gulf port authority has published a data hosting or residency instrument. The international layer legislates where data must be sent, never where it must reside. The absence of a sectoral middle layer is the finding.
This is the sector in which the honest answer is that there is no sector-specific obligation, and it is established rather than assumed. Across every Gulf aviation instrument read in full and every register read in full, there is no aviation-sector or port-sector data hosting or residency provision of any kind. Qatar's complete published aviation-laws taxonomy contains no information-security, cybersecurity or data category, and its Civil Aviation Security Regulations 2019 were retrieved in full and searched: the terms “cyber”, “personal data”, “passenger data” and “Advance Passenger” do not appear. Oman's complete list of civil aviation regulations contains no title mentioning cyber or information security. The Saudi cloud controls contain no occurrence of “aviation”, “airport” or “air transport”.
The distinction a buyer most needs here is between a duty to send data and a duty to hold it somewhere, and on the retrieved instruments that distinction holds without exception. Annex 9 to the Chicago Convention obliges contracting states to create a passenger data single window, to establish an advance passenger information system, to support it “by appropriate legal authority” and to confine data elements to the standard message format. Its only locational language is a transmission destination: the explanatory note records that “this information is electronically transmitted to the border control agencies in the destination or departure country”. Nothing in it addresses where the data then rests.
United Nations Security Council Resolution 2396 of 2017, adopted under Chapter VII, is to the same effect. Paragraph 11 decides that “Member States shall require airlines operating in their territories to provide API to the appropriate national authorities”, and paragraph 12 that states “shall develop the capability to collect, process and analyse” passenger name record data. These are collection, provision, analysis and sharing duties; territory appears only as the trigger identifying which flights are caught. The civil aviation cyber resolution chain runs four deep from 2016 to the current text adopted in 2025, and the 2019 resolution was retrieved in full: it contains no reference whatever to localisation, residency or the territorial location of systems or data.
The maritime position is the same and is more often misdescribed. The International Maritime Organization resolution on maritime cyber risk management in safety management systems, adopted on 16 June 2017, was retrieved in full. It is four operative paragraphs. Its verb is “ENCOURAGES” and its addressee is administrations, meaning flag states, not shipowners: it encourages them “to ensure that cyber risks are appropriately addressed in safety management systems no later than the first annual verification of the company's Document of Compliance after 1 January 2021”. It works through the safety management code and flag-state verification. It contains no reference to data location, residency, hosting, cloud, servers, jurisdiction, cross-border transfer or storage. Vendor material routinely presents it as a mandatory cyber rule binding operators directly, and that is wrong on the text.
The accompanying guidelines, of 5 July 2017 and as revised on 7 June 2022, were both retrieved in full. They state in terms that “these Guidelines are recommendatory”, set out five functional elements and list eight classes of vulnerable shipboard system. Neither version mentions hosting, residency, data location, cloud or server location. The mandatory port facility security code, in force since 1 July 2004, predates the question: the word “cyber” does not appear in the organisation's own authoritative description of it.
One premise commonly stated in this field could not be verified and should not be repeated: that a mandatory maritime single window took effect on 1 January 2024. What is verified is that a mandatory electronic-exchange duty on public authorities has applied since 8 April 2019, and that the single window itself is framed as a recommended practice. The organisation's own live page, retrieved on 10 November 2026, still describes it that way and mentions no 2024 mandatory single window. Either that page is stale or the instrument is not where it is usually placed. The conflict is unreconciled.
Three honest gaps. The Emirati civil aviation registers are populated by script and returned no records, so no finding either way is available, and the Saudi
aviation authority's site could not be read. Advance passenger information rules for all six states could not be obtained, so whether any Gulf regime requires such data to be held in country rather than merely pushed is unknown. Customs single windows, the most promising untested place for a real trade-data hosting obligation, could not be reached at all, and no enforcement search was possible.
Where a Gulf regulator does publish something, it governs only itself. The Saudi Transport General Authority, the maritime regulator, publishes a Cybersecurity General Policy last updated 16 December 2024. Every one of its nine clauses has the Authority itself as the subject, and none imposes any duty on a port, a terminal operator, a shipping company or a licensee. It contains no data location, hosting, residency or cloud provision. Its operative pointer is clause five, that the Authority “complies with the regulatory, legal, and legislative requirements issued by the National Cybersecurity Authority”: the sector regulator defers upward. Clause six adopts an international information security standard by choice, the same voluntary standard the maritime guidelines point to.
Port authorities publish less still. Abu Dhabi Maritime's full index of marine notices and circulars was retrieved and searched for cyber, data, hosting, privacy and single window, and no circular title mentions any of them. The corporate sites of the Emirati and Omani port groups were retrieved in full and searched: the only data-related artefact in each is a corporate privacy policy. The Kuwaiti ports authority publishes nothing of the kind at all, though its site carries a notice that it is under development. The Arab regional aviation body maintains a standing cybersecurity working group whose published-records page returns “No records”.
The takeaway is that this sector's data governance is borrowed in both directions, upward from the national cyber and data regime and sideways from voluntary international standards, with nothing in between. For a buyer that is not reassurance. A sector with no sectoral instrument has no sectoral safe harbour either, so the hosting question is decided entirely by whichever general rule catches the entity: a critical-infrastructure designation it cannot see, a general data protection law, or the terms of a border or customs system it does not control. The absence of a rule is why diligence here is harder, not easier.
Rows record what was established from retrieved instruments. Qatar's banking cloud position, Kuwait's and Oman's banking cloud rules, and the Emirati telecoms assurance regulation could not be read and are marked as not established rather than as absent. Entries reading “nothing” mean that the instrument named was retrieved and contains no location requirement, not that no obligation could ever arise.
Across the six sectors only two carry an instrument that names a place, and only one of those is a flat mandate. Everywhere else the binding question is not where the data sits but who must approve, who is designated, and who is permitted to sell the hosting.
What the comparison shows is that sector matters less than the form the obligation takes. Of the six sectors examined, only financial services and healthcare are governed by an instrument that names a territory, and of those only the Emirati banking and health rules are flat mandates. In the other four sectors the operative constraint is procedural rather than geographic: an approval that must precede signature, a designation the entity cannot see, or a licence condition that binds the provider rather than the buyer. A buyer who organises diligence around the question “must this data stay in country” will therefore get the wrong answer in two thirds of cases, because in those cases the rule does not speak about country at all.
| Sector | Regulator and instrument | What must stay in country | Prior approval |
|---|---|---|---|
| Banking and financial services | CBUAE, Outsourcing Regulation for Banks C 14/2021, effective 15 July 2021. SAMA, Cyber Security Framework v1.0 of May 2017, s.3.4.3, and Rules on Outsourcing. CBB, Rulebook OM-2.1.7 | UAE: the Master System of Record and all Confidential Data. Saudi Arabia: cloud services, in principle, unless approved otherwise. Bahrain: nothing | Yes in the UAE and Saudi Arabia. In Saudi Arabia before the contract is signed. Bahrain carves cloud out and takes post-notification within five working days |
| Healthcare | MOHAP and the health authorities, UAE Federal Law No. 2 of 2019, Art. 13, with MOHAP Resolution 51 of 2021. Elsewhere the general data law only | UAE: health data related to services provided in the state, and a copy under several exemptions. Saudi Arabia, Qatar, Bahrain, Oman, Kuwait: nothing | No general approval to host. Qatar and Oman require a permit to process health data at all. UAE export needs an exemption and usually written patient consent |
| Sector | Regulator and instrument | What must stay in country | Prior approval |
|---|---|---|---|
| Energy, oil and gas, utilities | No sector regulator instrument found. NCA, ECC-2:2024 and CCC-2:2024. CITRA, Cloud Computing Regulatory Framework v2.4. Oman, Royal Decree 64/2020 | Saudi Arabia: no cloud hosting rule since the 2024 deletion; managed security operations centres only. Kuwait: tiers three and four, and personal data. Oman: nothing, but see approval | Kuwait licenses the provider. Oman requires Cyber Defence Centre approval before any sensitive data is stored or processed abroad. Dubai requires CSP certification to sell to government |
| Government and public services | TDRA and the UAE Data Office, Federal Decree-Law No. 45 of 2021, Art. 2(2). DESC, ISR under Resolution 13 of 2012. NCA, CCC-2:2024. CITRA, Data Classification Policy v2.3 | Federal UAE: nothing, government data being excluded from the data law. Dubai: cannot be determined, the rules not being published. Kuwait: nothing under the classification policy | No approval regime found. The discipline is framework compliance, audit, reporting and platform registration, and it binds the provider more than the entity |
| Telecommunications | CST, class licensing for cloud. CITRA, Cloud Computing Regulatory Framework v2.4. CRA, Cloud Policy Framework of 7 June 2022. TDRA, UAE Information Assurance Regulation | Kuwait: tiers three and four, and personal data, with no temporary exception. Saudi Arabia: not established from the framework itself. UAE and Qatar: not established | Yes, as a licence rather than a consent. Saudi Arabia grants class licences assessed against the NCA controls. Kuwait licenses providers holding the top two tiers. Qatar publishes recommendations, not a licence |
| Aviation, logistics and ports | No Gulf aviation regulator or port authority instrument found. ICAO Annex 9, UNSCR 2396, IMO Resolution MSC.428(98), MSC-FAL.1/Circ.3/Rev.2 | Nothing. No instrument retrieved imposes any storage or hosting location requirement at either the international or the Gulf layer | None on hosting. The duties are to transmit passenger data, to manage cyber risk in safety management systems, and to exchange port information electronically |
Three things escape the grid, and each has decided a real architecture. The first is designation. In four of the six sectors scope turns on whether the entity is critical national infrastructure, and no Gulf state publishes its
roster. Dubai publishes ten sectors and not the organisations; Saudi Arabia defines the perimeter functionally and leaves the entity to identify itself; Oman holds the power on the face of its decree and publishes no output. An obligation whose addressee is secret cannot be planned against from the instrument alone.
The second is that location and control are different questions. Kuwait is the only state examined whose cloud instrument contains a blocking provision, prohibiting a provider from surrendering subscriber content to “a local or foreign government, or public authorities” when complying with foreign law. Residency without such a clause answers where the data sits and not who can compel its production, which is usually the question the board was actually asking.
The third is that the obligation increasingly travels with the purchase route rather than the rule. A Saudi buyer of one hyperscaler's services must buy through a single exclusive in-Kingdom reseller for every region, and that reseller was absorbed into a sovereign fund vehicle on 3 November 2026. Dubai makes certification mandatory for any provider selling cloud to its government and certifies no hyperscaler, while certifying the state utility's own subsidiary. Sovereignty delivered by exclusive distribution concentrates counterparty risk at the moment it resolves jurisdictional risk.
The questions below are ordered so that each answer narrows the next. It is to be noted that the order matters: taking them in any other sequence produces the familiar error of asking a hosting question of a regime that does not regulate hosting.
Two cautions on using this sequence. It resolves the published position and not the supervisory one: in several states the operative rules are not public documents, Dubai's information security regulation and cloud provider standard being available only on email request, and the Saudi localisation rules having been transferred in 2024 to a body whose publications could not be retrieved. And it resolves obligations, not practice. Across all six sectors no published penalty expressly grounded in hosting, cloud or data residency was located anywhere in the Gulf's regulated perimeter.
The takeaway for the edition is that the Gulf's sovereignty regime is not converging on residency. It is converging on authorisation. Two of the six sectors tell a buyer where to put the data; four tell them who must say yes first, or who is allowed to sell them the capacity. That is a more tractable problem than localisation and a less comfortable one, because an approval can be withheld, a designation can be made without notice, and an exclusive distribution route can be consolidated into a state vehicle in a single week, as happened in Saudi Arabia at the start of this November. The instrument is the easy part of the diligence. The counterparty is the hard part.
A Call is a dated, falsifiable prediction with a stated confidence, written to be graded in the 2028 Scorecard. Ten Calls for GCC digital sovereignty in 2027: the heart of this edition.
Ten dated, falsifiable predictions for GCC digital sovereignty. Confidence is stated so that a contrarian medium-confidence Call and a high-confidence base case are read (and scored) differently.
Each Call draws on the sector evidence in Part I and on Volge's Rules of the Game regulatory tracker. No Scorecard appears in Vol. I: the 2028 edition grades these against the record.
Confidence: High · Resolves end-2027
By end-2027, classification-driven data residency (sovereignty-by-design) is a default clause in GCC government cloud requests for proposal, specifying a tier-to-hosting map rather than a blanket localisation rule.
The architecture is in place across all six regimes, each already routes hosting through a classification tier, and the shift from mandate to architecture makes the clause the path of least resistance for a procurement officer. The mechanism exists; 2027 is when it becomes boilerplate.
Saudi Arabia's SDAIA adequacy list and standard contractual clauses become the de-facto reference mechanism for GCC cross-border transfers; no unified GCC-wide transfer instrument emerges in 2027.
Confidence: Medium–High · Resolves end-2027
The 2024 Data Transfer Regulation gives Saudi Arabia the region's most complete transfer framework. SDAIA has since consulted, from 6 October to 5 November 2026, on draft amendments that would require personal data to be kept in the Kingdom, so the Call is graded on the transfer regime's influence separately from that pending localisation duty. The smaller states lack the capacity to build their own; a single Gulf instrument would require political convergence that is not on the 2027 agenda.
A Saudi-approved sovereign hosting model (the Center3, HUMAIN or SDAIA template) is recognised or replicated across at least two other GCC states by end-2027.
Confidence: Medium · Resolves end-2027
Saudi scale makes its model the natural reference, and smaller states short of hyperscaler steel have reason to adopt rather than build. The confidence is medium because national pride and the Emirates' rival partnership model both cut against straightforward replication.
The build is funded, staffed and under construction by Khazna, but the first phase was not live as at mid-October 2026, Khazna now indicates a Q4 2026 go-live, and September 2026 reporting suggests re-planning from one site into several. Confidence is cut accordingly.
Delivery risk is real but the narrative dominance is near-certain given the scale and the backers.
Stargate UAE's first 200 MW cluster goes live in 2026 and the sovereign-cloud and sovereign-AI narrative it anchors dominates GCC procurement conversations through 2027.
Confidence: Medium · Resolves end-2027
United States export-licence posture remains the single biggest swing factor for GCC sovereign-AI capacity through 2027, whether through further easing or a China-diversion clawback.
Confidence: High · Resolves end-2027
The 2025-26 sequence (Diffusion Rule, rescission, licence-by-licence, Country Group A:5) shows the region's effective compute ceiling moving on Washington's decisions. The live variables for 2027 are the 6 April 2027 expiry of Supplement 8 eligibility for G42 and Core42, and Saudi Arabia's exclusion from Country Group A:5. This also shows the ceiling moving on Washington's decisions alone. No Gulf variable has comparable leverage over capacity.
Hold-your-own-key or external key management, together with confidential computing, become standard requirements for GCC top-tier (Secret and L4) workloads by end-2027.
Confidence: Medium · Resolves end-2027
The controls are mature and already specified in the most advanced tenders, and they are what allow top-tier workloads to move to cloud without an air-gap. Confidence is medium because standardisation across six regimes at differing maturity is uneven.
Oman's fully-enforced PDPL (effective 5 February 2026) triggers visible 2027 compliance and localisation demand, and at least one other GCC state tightens its cross-border transfer rules over the year.
Confidence: Medium · Resolves end-2027
Enforcement began on schedule, and Royal Decree 68/2026 then widened the law from 7 September 2026. The demand driver is now extraterritorial reach rather than the February cliff. A first fully-enforced small-state regime typically pulls neighbours toward tighter rules. The second clause (another state tightening) is the source of the medium rating.
At least one national Arabic large language model (ALLaM, Falcon, Jais or Fanar) is adopted into a named, non-pilot GCC government production workflow with a public reference, by end-2027. The Ministry of Commerce's use of HUMAIN's AI cloud from 1 September 2026 does not by itself satisfy this threshold.
Confidence: Medium–High · Resolves end-2027
ALLaM already powers HUMAIN Chat, Microsoft and HUMAIN announced on 26 August 2026 an intent to offer it through Microsoft Foundry, and the UAE Cabinet set an objective in April 2026 to move half of federal services to agentic AI within two years, so the threshold has been raised to keep the Call falsifiable; the sovereignty logic of a home-hosted model makes production adoption a matter of when, not whether. Confidence short of high reflects the gap between pilot and production.
Kuwait's announced Google Cloud and Microsoft Azure regions remain committed-but-not-live into 2027, keeping the country's sovereignty telco- and NVIDIA-delivered through Ooredoo and KISR.
Confidence: Medium · Resolves end-2027
Both regions were announced (2023 and 2025) without live dates, and Kuwait's grid and data-centre base is thin (AGBI, Feb 2026). The telco-delivered stopgap (Ooredoo's H200 site, KISR's ICARUS) is already the working sovereignty layer.
The sovereign premium (the 15–30% mark-up of sovereign over public cloud) persists through 2027; sovereignty remains a cost of Gulf data control, not yet a saving.
Confidence: Medium · Resolves end-2027
The ten Calls are the firm's considered judgements as of July 2026, each written to be graded against the public record in the 2028 Scorecard.
The Call grades persistence rather than magnitude, the band having been restated to BCG's 15 to 30 per cent. The premium's sources, lower utilisation, scarce vetted staff and added key and confidential-computing layers, do not resolve in a single year while the regions are still filling. Cheap power offsets the build cost but not the buyer's premium.
A Call is not a forecast in the ordinary sense. It is a dated, falsifiable proposition with a stated confidence, constructed so that a reader can grade it a year later without recourse to the author.
Three disciplines govern every Call. First, falsifiability: each is written so that a clear real-world outcome would mark it right or wrong: 'a national Arabic LLM enters a government production workflow' can be checked; 'AI adoption grows' cannot. Second, dating: every Call resolves at a stated point, here end-2027, so that it cannot be quietly deferred. Third, calibrated confidence: a Call is rated High, Medium-High, Medium or Low, and the ratings are meant to be read as probabilities: a well-calibrated set of Medium Calls should be wrong about as often as it is right.
Each Call is built from the sourced evidence in Part I (the market sizing, the regulatory crosswalk, the compute commitments) and from Volge's Rules of the Game regulatory tracker. Confidence reflects two things: the strength of the mechanism (is there a specific, dated cause?) and the number of independent ways the Call could fail. Call 5, on United States export posture, is High because the mechanism is proven and singular; Call 3, on model replication, is Medium because national pride, rival models and procurement inertia each offer a separate route to failure.
The confidence ratings are also a hedge against the author. A High Call is a claim the firm expects to be judged harshly on if it fails; a Medium Call signals genuine uncertainty and should not be scored as a miss with the same weight. The 2028 Scorecard will grade the full set and, crucially, assess calibration (whether the Highs came in more often than the Mediums) because a forecaster who is right half the time on Medium Calls is better calibrated than one who is right nine times in ten but rated them all High.
Volge Quarterly is governed by Volge's Morpheus standard: conclusion first, every figure sourced and year-stamped, every Call dated and falsifiable. The Calls are reasoned judgements, not guarantees; and they are written to be marked.
National models of compute · Volge Quarterly 2027
The year read from four angles (technology, markets, regulation and delivery) and resolved into bull, base and bear cases for GCC digital sovereignty in 2027.
Read as technology, 2027 is the year the sovereign stack's upper layers mature while its foundation stays imported. Read as a market, it is the year sovereign capacity is bought at a premium the buyer can no longer avoid.
The technology story is one of a stack maturing from the top down. Model layer, key-management layer and confidential-computing layer are all reaching production readiness in-region: national LLMs move toward government workflows, hold-your-own-key schemes enter the most sensitive tenders, and hardware enclaves (Intel TDX, AMD SEV-SNP, NVIDIA Confidential Computing) make top-tier cloud hosting viable without an air-gap. The one layer that does not mature in-region is the silicon, which is fabricated, licensed and rate-limited abroad. The technical frontier of 2027 is therefore the trust boundary: how much sovereignty can be engineered above a supply chain the region does not own.
The market story is one of capacity bought faster than it can be supplied. The Saudi cloud market grows from $15.4bn in 2025 toward $27.9bn by 2030, and the
UAE's from $12.8bn toward $56.3bn by 2031 [Mordor, 2025]; Middle East live data-centre capacity rises from about 1.2 GW at end-2025 to 2.2 GW in 2026 and 3.2 GW in 2027, with the 14 GW announced likely to deliver closer to 5 GW by 2030 [Knight Frank, Data Centre Atlas 2026]; MEA colocation grows from $3.3bn toward $12.9bn by 2031 [Arizton, 2025]. Against that demand, the constraints bind: vetted staff are scarce, power is abundant but grid connections are not everywhere ready, and chips are capped. The sovereign premium of 15–30% is the price of buying into that supply-constrained market, and Call 10 holds that it persists.
Technology matures above the silicon; the market pays a premium for capacity it cannot yet supply at scale. Both point to the same 2027 discipline: classify precisely, so that neither the engineering nor the spend exceeds what the workload's tier requires.
Read as regulation, 2027 is the year enforcement catches up with legislation. Read as delivery, it is the year the gap between an announced region and a live one decides which states have sovereignty in fact.
The regulatory story is enforcement, not new law. The statutes largely exist (Qatar since 2017, Bahrain since 2019, the Emirates and Saudi Arabia since 2021-22, Kuwait's regulation since 2024), and the pivotal 2027 event is Oman's PDPL becoming fully enforceable on 5 February 2026 and widened by Royal Decree 68/2026 from 7 September 2026, which adds extraterritorial reach, cuts the exemptions from ten to six and creates a right to human review of automated decisions, the region's one hard cliff. Around it, breach-notification norms converge on 72 hours, Saudi Arabia's transfer regime becomes the de-facto reference (Call 2), and the Emirates' still-unpublished federal executive regulations remain the region's largest single piece of unfinished legislative business. No unified Gulf instrument arrives; convergence runs through classification, not statute.
The delivery story is the gap between announcement and availability. Kuwait's Google and Microsoft regions, announced in 2023 and 2025, remain not-live, so its sovereignty is carried by Ooredoo and KISR (Call 9);
the Emirates' Stargate first phase is a 2026 delivery test whose success sets the regional narrative (Call 4). Delivery is where the year separates states that have specified sovereignty from states that have built it; and the separator is not capital, which is abundant, but hyperscaler steel, vetted staff and grid connections, which are not.
Regulation is converging through enforcement and classification; delivery is diverging through the announce-to-live gap. The states that close that gap in 2027 are the ones whose sovereignty is more than a statute.
The year resolves along one dominant axis: United States export posture and the pace of sovereign-cloud delivery. Three scenarios bound the range.
Country Group A:5 holds and eases further; Stargate UAE's first 200 MW lands on schedule and a second phase is committed; classification-driven residency becomes standard RFP language across all six states; a national Arabic LLM reaches production in government. Sovereign capacity scales, the premium begins to compress at the margin, and the Gulf consolidates as the third pole of sovereign AI after the United States and China. Calls 1, 4, 5 and 8 all resolve firmly right.
The central case. Export posture stays broadly permissive but remains the dominant variable; Stargate's first phase lands with the expected slippage; Saudi Arabia's transfer regime becomes the de-facto reference; Oman's enforcement bites; Kuwait's regions stay committed-not-live; the 15–30% premium persists. Sovereignty is specified and largely delivered by the two poles, improvised by the smaller states, and the classification switch becomes the shared operating layer. Most Calls resolve as rated.
A China-diversion clawback re-imposes licensing or re-tiers the Emirates; chip flows slow; Stargate's first phase slips into 2027; grid and talent constraints bind harder than expected. Sovereign-AI capacity stalls against a ceiling set abroad, the premium widens as scarcity deepens, and the dependence paradox
moves from a managed condition to an active constraint. Call 5 dominates the year, and Calls 4 and 8 resolve late or wrong.
By domain · Volge Quarterly 2027
What the year turns on, the signals worth tracking as it runs, and the reference annex: sources, standards and glossary behind every figure in this edition.
The signals that will seed next year's Calls. Each is a specific, observable event whose occurrence would move the 2028 read: watched here so that the 2028 edition can grade both the Calls and the conditions that shaped them.
| Signal | Why it matters | What to watch |
|---|---|---|
| US export re-tiering | Sets the sovereign-AI ceiling (Call 5) | Any change to UAE Country Group A:5 status; new licence conditions |
| Stargate UAE phase 2 | Confirms or slows the flagship build (Call 4) | First 200 MW live date; second-phase commitment |
| UAE federal Exec Regs | Closes the region's largest legislative gap | Publication of PDPL executive regulations and grace-period trigger |
| Oman enforcement actions | Tests the 5 Feb 2026 cliff (Call 7) | First fines or transfer refusals under the enforced PDPL |
| Kuwait region go-live | Would falsify Call 9 | Google or Microsoft Kuwait region reaching general availability |
| National LLM in production | Resolves Call 8 | ALLaM, Falcon, Jais or Fanar in a named government workflow |
| Sovereign premium data | Tests Call 10 | Published or tendered sovereign-vs-public price differentials |
| Cross-border precedent | Tests Call 2 | SDAIA adequacy findings; any move toward a GCC-wide instrument |
The watchlist is not a set of Calls; it is the field of signals from which the 2028 Calls will be drawn and against which this year's will be graded.
Each signal is specific and observable, so that a year from now its occurrence or non-occurrence is a matter of record, not interpretation: the same standard the Calls are held to.
The full regulatory crosswalk behind this edition, by country, on the axes a hosting decision turns on. Every figure elsewhere in the edition traces to this table or to the numbered sources that follow.
| State | Law & date | Breach | Cross-border rule | Max penalty |
|---|---|---|---|---|
| Saudi Arabia | PDPL M/19 (2021) / M/148 (2023); enforced 14 Sep 2024 | 72h | Data Transfer Regulation 2024: adequacy / SCCs / BCRs. Draft amendments consulted 6 Oct to 5 Nov 2026 would add an in-Kingdom storage duty | SAR 5m (10m repeat); +2yr & SAR 3m for sensitive misuse |
| UAE (federal) | PDPL Decree-Law 45/2021; in force 2 Jan 2022; Exec Regs unpublished | 72h (de-facto) | Adequacy-based; free zones separate | Set by unpublished Regs |
| UAE (DIFC) | DP Law 5/2020; 2025 amendment | 72h | GDPR-aligned adequacy / SCCs | Private right of action (2025) |
| UAE (ADGM) | DP Regs 2021 | 72h | GDPR-aligned; Commissioner | To $28m |
| Qatar | PDPPL Law 13/2016; in force 13 Feb 2017 | 72h | Adequacy; QFC DP Regs 2021 separate | QAR 1–5m (QFC to $1.5m) |
| Kuwait | CITRA DPPR Res 42/2021 → 26/2024; in force 19 Feb 2024 | 72h | Adequacy required; L3/L4 in-country (Res 112) | Regulatory (CITRA) |
| Bahrain | PDPL Law 30/2018; in force 1 Aug 2019 | Notify | Open host; Data Embassy Decree 56/2018 | BHD 1,000–20,000; to 1yr |
| Oman | PDPL RD 6/2022; Exec Regs MD 34/2024; enforced 5 Feb 2026 | 72h | Sensitive transfers need CDC approval | To RO 500,000 (Art 23) |
Sources: national statutes and regulator publications; Volge Rules of the Game tracker. Current to July 2026. The UAE appears three times because the federal law and the DIFC and ADGM free-zone regimes are distinct.
The decisive columns are the cross-border rule and the classification note within it: they are what a tier-to-hosting map is built on, and what separates the
localisers (Saudi Arabia, Kuwait, Oman) from the open hosts (Bahrain, Qatar).
Numbered sources for the figures and claims in this edition. Where a live URL is available it is given; statutory instruments are cited by decree or law number and date.
All figures were refreshed to November 2026 and are year-stamped at point of use. Statutory citations are to the instrument as amended. This edition is not investment advice.
The working vocabulary of Gulf digital sovereignty, defined as used in this edition.
Figures are reconciled across independent sources; where sources disagree on scope, a range is given rather than a false point estimate. All data was current to July 2026 and is year-stamped at point of use. The Calls are the firm's reasoned judgements, not guarantees, and are written to be graded against the public record.
Volge Quarterly is Volge's signature series, published every quarter for enterprise customers. Each edition takes one subject, sets out where it currently stands on the published record, and makes a set of dated, falsifiable judgements on where it is going, written to give a buyer something firmer than a forecast: every figure carries its source and its date, every judgement carries the date on which it resolves, and the following year's edition grades each one hit, missed or partial.
Where a figure is not published, the edition says so rather than estimating it. Where published sources disagree, both are given and the disagreement is stated rather than averaged away. A number that does not survive checking is corrected here rather than repeated.
The edition carries no recommendation of any supplier and no offer of service. Where a vendor, platform or deployment is named it is because it is part of the published record of the subject.
This edition may be shared freely. It may not be altered or republished without permission, and figures carry their stated basis and date. Names and marks appearing in it are the property of their respective owners.