>
The Foundation Gap : GCC Data Engineering Outlook
Volge
Quarterly
Q4 2026 · The Data Engineering issue
AI FAQs
101 for the board
Green Premium
The hydrogen bet
Trust Fabric
Trust as plumbing

Foundation
Gap

Pipelines, quality, lineage and ownership: the engineering that decides which AI programmes compound.

volge.com/quarterlyComplimentary editionISSN 2977-4404 · 03

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.

Plumbing before intelligence. The Gulf is buying artificial intelligence faster than it is building the data foundations beneath it; 2027 is the year data engineering becomes the binding constraint.

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.

Contents

  1. 02The year data engineering becomes the binding constraint
  2. 02Plumbing before intelligence
  3. 03The state of the sector
  4. 03The region in numbers
  5. 03The national programmes
  6. 04The national programmes: emirate and Qatar
  7. 05How it is built today
  8. 05The maturity model
  9. 06Where it falls short
  10. 06The forces shaping the year
  11. 06Sovereignty by design
  12. 07The talent bottleneck
  13. 07From projects to platforms
  14. 08Building for AI
  15. 08Cost discipline
  16. 09The buyer's ground
  17. 09What the rules actually require of your data
  18. 11Abu Dhabi, read clause by clause
  19. 13Saudi Arabia: the apparatus, and the number nobody publishes
  20. 14Oman and Kuwait: the two regimes a buyer can read
  21. 16The ninety-five per cent that is not what you think
  22. 18Where the Gulf actually stands
  23. 19What it costs, and the Gulf premium nobody quotes
  24. 21A maturity model a buyer can use without a licence
  25. 23The dated calendar, and what is not on it
  26. 25The Calls
  27. 25The Calls at a glance
  28. 26Calls one to three
  29. 26Calls four to six
  30. 26Calls seven and eight
  31. 27Calls nine and ten
  32. 27How the Calls are made
  33. 27The year by domain
  34. 27Technology and markets
  35. 28Regulation and delivery
  36. 28Three ways the year runs
  37. 29The watchlist
  38. 29What to watch, and the record
  39. 29The data appendix
  40. 30Sources
  41. 31Glossary
Volge Quarterly is Volge's signature series, published every quarter for enterprise customers.
Volge · Panache Editions · The GCC Data Engineering edition
Data EngineeringVolge Quarterly01 •

The year data engineering becomes the binding constraint

2027 is the year the Gulf's data foundations, not its models or its capital, decide what artificial intelligence delivers.

The Gulf is buy­ing in­tel­li­gence faster than it is build­ing the plumb­ing be­neath it. The order of the two mat­ters, be­cause the plumb­ing is what the in­tel­li­gence runs on. Com­pute has been se­cured at na­tional scale, cap­ital is ef­fect­ively un­con­strained, and the polit­ical will to lead in ar­ti­fi­cial in­tel­li­gence is stated at the highest level of gov­ern­ment. What has not kept pace is the un­glam­or­ous layer between the raw data and the model: the in­ges­tion, the clean­ing, the cata­loguing, the gov­ernance and the in­teg­ra­tion that turn scattered re­cords into a found­a­tion an al­gorithm can stand on. In 2027 that layer be­comes the bind­ing con­straint, and the edi­tion's single thesis is that the con­straint is now made of data en­gin­eer­ing rather than of chips, money or am­bi­tion.

The evid­ence for the ceil­ing is by now hard to dis­pute. A pre­lim­in­ary MIT Pro­ject NANDA study found that roughly 95 per cent of or­gan­isa­tions in its evid­ence base, 52 in­ter­views and 153 sur­vey re­sponses, had re­cor­ded no meas­ur­able profit-and-loss im­pact from gen­er­at­ive AI [MIT Pro­ject NANDA, July 2025]. The find­ing is about or­gan­isa­tions rather than about in­di­vidual pi­lots, and it is the ceil­ing on meas­ur­able re­turn, an out­come at­trib­uted to weak data found­a­tions rather than to model qual­ity [MIT NANDA, "State of AI in Busi­ness 2025", via For­bes, 15 Oc­to­ber 2025]. The find­ing is not that the mod­els are in­ad­equate; it is that they are being asked to reason over data that has not been en­gin­eered to be reasoned over. Gart­ner has reached a com­pat­ible con­clu­sion, ob­serving that half of gen­er­at­ive-AI pro­jects fail and that more than 40 per cent of agen­tic-AI pro­jects are ex­pec­ted to be can­celled be­fore the end of 2027 [Gart­ner, 26 Janu­ary 2026; 25 June 2025]. The pat­tern is con­sist­ent across the re­gion's lar­ger eco­nom­ies, where cap­ital ex­pendit­ure has been ob­served to race ahead of data and gov­ernance ma­tur­ity [McKinsey, "The state of AI in GCC coun­tries", 13 Novem­ber 2025].

The foundation, not the model, is the ceiling, and in the Gulf the foundation is being built last.

This is not a pess­im­istic read. The mar­ket is large and its tra­ject­ory is steep: the GCC cloud mar­ket was worth in ex­cess of thirty bil­lion US dol­lars in 2025, with Saudi Ar­a­bia at 15.39 bil­lion and the United Arab Emir­ates at

12.84 bil­lion, the lat­ter grow­ing at a com­pound an­nual rate near 28 per cent to 2031 [Mordor In­tel­li­gence, 2025]. The re­gional data-centre mar­ket has been sized at 2.39 bil­lion US dol­lars in 2025, pro­jec­ted to reach 21.67 bil­lion by 2034 [IMARC, 2025]. The buil­dout is real, it is well fun­ded, and it is ac­cel­er­at­ing. What is ar­gued here is nar­rower and more use­ful: that the re­turn on all of it turns, in 2027, on a layer that has been treated as an af­ter­thought.

The quieter story is where ad­vant­age now moves. It moves away from the en­tit­ies that have bought the most com­pute and to­wards those that have en­gin­eered the clean­est, best-gov­erned and most in­ter­op­er­able data; and it moves from the sys­tem in­teg­rator, who builds and de­parts, to the in-house plat­form team, who owns and com­pounds. The Gulf has proven it can buy any­thing it chooses. Whether it can build the found­a­tion is the ques­tion of the year, and it is answered one data plat­form at a time.

Plumbing before intelligence

The sequence is the argument: foundations first, or the intelligence layer has nothing sound to stand on.

Three facts, taken to­gether, define the year. The first is that power and land are cheap and abund­ant: in­dus­trial elec­tri­city in Riy­adh and Abu Dhabi runs at roughly 0.05 to 0.07 US dol­lars per kilo­watt-hour, some five to seven times below the United King­dom [Ox­ford Eco­nom­ics, 22 Janu­ary 2026]. The second is that cap­ital is deep bey­ond or­din­ary con­straint: the Pub­lic In­vest­ment Fund re­ports as­sets of 900 bil­lion US dol­lars and above, about SAR 3.4 tril­lion at end-2025 [PIF, 2025 an­nual re­port]; third-party rank­ings put it near 1.15 tril­lion by June 2026, sit­ting along­side Mubadala at roughly 330 bil­lion and ded­ic­ated vehicles such as MGX and the HUMAIN pro­gramme [sov­er­eign dis­clos­ures, 2025–26]. The third is that the data found­a­tion is thin: the cost of poor data qual­ity has been es­tim­ated at about 12.9 mil­lion US dol­lars per or­gan­isa­tion each year [Gart­ner, via IBM, 23 Janu­ary 2026].

Data EngineeringVolge Quarterly02 •

The con­sequence is pre­cise. Where com­pute and cap­ital are abund­ant and the found­a­tion is scarce, the scarce input sets the re­turn. It is to be noted that this is the re­verse of the po­s­i­tion most Gulf en­tit­ies be­lieve them­selves to be in; the per­ceived con­straint has been ac­cess to chips and to mod­els, and the ac­tual con­straint has quietly be­come the abil­ity to feed them. PwC has pro­jec­ted that ar­ti­fi­cial in­tel­li­gence will add roughly 320 bil­lion US dol­lars to the Middle East by 2030, with Saudi Ar­a­bia at about 135 bil­lion and the UAE at about 96 bil­lion [PwC, 2023–24]; the real­isa­tion of that fig­ure is now a data-en­gin­eer­ing ques­tion rather than a com­pute one.

The one-line read

The Gulf has se­cured the com­pute, the power and the cap­ital for ar­ti­fi­cial in­tel­li­gence. In 2027 the re­turn on all three is de­cided by the data-en­gin­eer­ing layer be­neath them, which has been built last and re­mains the re­gion's bind­ing con­straint.

What fol­lows reads that layer end to end. Part I es­tab­lishes the state of the sec­tor; Part II isol­ates the five forces; Part III makes the Calls; Part IV reads the year by do­main and runs the scen­arios; Part V is the toolkit and the annex. Through­out, the re­gister is de­lib­er­ately plain and every fig­ure is sourced and year-stamped, so that an op­er­ator, an in­vestor or a board may see where the re­gion stands and where it is going.

The ed­it­ors, Volge · Pan­ache Edi­tions, July 2026.

Two points qual­ify the totals. The ca­pa­city fig­ure is a live-load meas­ure and not a meas­ure of what that ca­pa­city is made to do: Middle East live IT ca­pa­city was 761 mega­watts in 2024, 1,115 in 2025 and is fore­cast at 1,669 in 2026, with the UAE hold­ing 35.2 per cent of the GCC and Khazna alone car­ry­ing a pipeline above 300 mega­watts [Knight Frank / DC Byte; IMARC / Busi­ness Wire, Janu­ary 2026]. The coloca­tion seg­ment across the Middle East and Africa has been sized at 3.33 bil­lion US dol­lars in 2025, rising to 12.88 bil­lion by 2031 at a com­pound rate of 25.32 per cent, at a build cost of six to twelve mil­lion US dol­lars per mega­watt [Ar­izton, 2025]. The mega­watts are being poured; the ques­tion the edi­tion presses is what frac­tion sits on data that has been en­gin­eered to be used.

Re­con­ciled from Mordor In­tel­li­gence and IMARC (cloud and data-centre siz­ing, 2025), Knight Frank / DC Byte (live ca­pa­city, 2024–26), sov­er­eign dis­clos­ures (PIF, 2025–26), PwC (AI value, 2023–24) and Ox­ford Eco­nom­ics (power, Janu­ary 2026). Ranges re­flect genu­ine source dis­agree­ment on scope. Cur­rent to July 2026.

The state of the sector

A cloud market above thirty billion US dollars in 2025, a data-centre buildout compounding near 27 per cent, and five national data programmes racing to govern it, set against a data-engineering layer that has not kept pace.

The region in numbers

The Gulf's data infrastructure is large, well funded and compounding fast; the figures below are the reconciled basis for the edition.

The scale of the buil­dout is not in dis­pute; the dis­agree­ment among sources is one of scope and defin­i­tion rather than of dir­ec­tion. The GCC cloud mar­ket stood above thirty bil­lion US dol­lars in 2025 on a na­tional build-up, led by Saudi Ar­a­bia at 15.39 bil­lion, fore­cast to 27.93 bil­lion by 2030 at a com­pound an­nual rate of 12.66 per cent, and the United Arab Emir­ates at 12.84 bil­lion, fore­cast to 56.26 bil­lion by 2031 at 27.93 per cent [Mordor In­tel­li­gence, 2025]. The re­gional data-centre mar­ket has been sized at 2.39 bil­lion US dol­lars in 2025, pro­jec­ted to 21.67 bil­lion by 2034 at a com­pound rate of 26.78 per cent [IMARC, 2025].

  • GCC cloud mar­ket, 2025: $30B+, KSA $15.39B and UAE $12.84B on a na­tional build-up [Mordor, 2025]
  • GCC data-centre mar­ket, 2025: $2.39B, To $21.67B by 2034, a 26.78% CAGR [IMARC, 2025]
  • ME live IT ca­pa­city, 2026f: 1,669 MW, About 1.2 GW at end-2025, 2.2 GW in 2026 and 3.2 GW in 2027 [Knight Frank, 2026]
  • PIF as­sets, June 2026: ~SAR 3.4T, From ~$925B in May 2025; cap­ital is ef­fect­ively un­con­strained
  • AI to the Middle East by 2030: ~$320B, KSA ~$135B, UAE ~$96B [PwC, 2023–24]
  • Per kWh, Gulf in­dus­trial power: $0.05–0.07, Five to seven times below the UK [Ox­ford Eco­nom­ics, 2026]

The national programmes

Every Gulf state has stood up a national data authority; the maturity of what sits beneath the authority varies widely.

Saudi Ar­a­bia has gone fur­thest in in­sti­tu­tional terms. The Saudi Data and Ar­ti­fi­cial In­tel­li­gence Au­thor­ity was es­tab­lished by Royal Order A/471 on 30 Au­gust 2019, re­ports to the Prime Min­is­ter, and is over­seen by a board chaired by the Crown Prince, with Dr Ab­dul­lah Alghamdi as pres­id­ent; it houses the Na­tional In­form­a­tion Centre, the Na­tional Data Man­age­ment Of­fice and the Na­tional Centre for AI [SDAIA, 2019]. The Na­tional

Data EngineeringVolge Quarterly03 •

Strategy for Data and AI was ap­proved on 17 July 2020, tar­get­ing more than 20,000 data and AI spe­cial­ists and around 75 bil­lion Saudi riy­als of in­vest­ment, with global lead­er­ship stated as the 2030 goal [NSDAI, 2020]. The Na­tional Data Man­age­ment Of­fice frame­work is un­usu­ally con­crete: fif­teen do­mains, sev­enty-seven con­trols and 191 spe­cific­a­tions, ar­ranged as a hier­archy from do­main to con­trol to spe­cific­a­tion, with a data-clas­si­fic­a­tion re­gister man­dated under its thir­teenth do­main [NDMO Stand­ards, 2022].

The Saudi data estate

The scale of in­teg­ra­tion already achieved is sig­ni­fic­ant. The Na­tional Data Bank and as­so­ci­ated lake in­teg­rate more than 360 gov­ern­ment sys­tems; the Na­tional Data Index cov­ers more than 370 sys­tems across in ex­cess of 130 Gov­ern­ment Data Of­fices; and the open-data portal data.gov.sa is the pub­lic face of the es­tate [SDAIA, 2025]. Ad­op­tion of the na­tional di­gital-iden­tity rails is high, with Na­fath at roughly 21.9 mil­lion act­ive users and Tawakkalna at around 33 mil­lion [SDAIA, 2025]. Data is clas­si­fied across four levels (Top Secret, Secret, Con­fid­en­tial or Re­stric­ted, and Pub­lic), which is the mech­an­ism through which res­id­ency and ac­cess are sub­sequently gov­erned.

The federal and emirate layer in the UAE

The United Arab Emir­ates gov­erns fed­eral data through policy rather than through a single stat­ute. The Smart Data Strategy and its ac­com­pa­ny­ing Smart Data Frame­work (Part 1 on prin­ciples and stand­ards, Part 2 on im­ple­ment­a­tion, draw­ing on the European In­ter­op­er­ab­il­ity Frame­work, ISO and W3C) set the fed­eral stand­ard, with bay­anat.ae op­er­at­ing as the fed­eral open-data portal under the Fed­eral Com­pet­it­ive­ness and Stat­ist­ics Centre [FCSC, 2025]. It is to be noted that this portal is dis­tinct from the G42-linked com­mer­cial en­tity of a sim­ilar name. The fed­eral Data Index meas­ures en­tity ma­tur­ity across reg­u­lat­ory (30 per cent), qual­ity and pro­tec­tion (30 per cent) and ac­cess­ib­il­ity and flow (40 per cent) di­men­sions. The Per­sonal Data Pro­tec­tion Law, Fed­eral De­cree-Law 45 of 2021, has been in force since 2 Janu­ary 2022, though its ex­ec­ut­ive reg­u­la­tions re­mained un­pub­lished in mid-2026, a ma­ter­ial gap for any en­tity en­gin­eer­ing for com­pli­ance.

The national programmes: emirate and Qatar

Below the federal layer, Dubai and Abu Dhabi run the region's most advanced sub-national data platforms, and Qatar has moved quickly to catch up.

Dubai treats data as an owned asset in law. Dubai Data Law No. 26 of 2015 es­tab­lished that Dubai Data is owned by gov­ern­ment and re­quired all such data to be clas­si­fied as Open or Shared; the gov­ern­ing body has since be­come the Dubai Data and Stat­ist­ics Es­tab­lish­ment under Law 24 of 2023, sit­ting within the Dubai Di­gital Au­thor­ity cre­ated by Law 22 of 2023 [Gov­ern­ment of Dubai, 2015–23]. The cent­ral plat­form, Dubai Pulse, de­liv­ers plat­form, data and in­fra­struc­ture as a ser­vice, built with the op­er­ator du. An up­dated Dubai Data Manual was launched on 1 July 2026, adding man­dat­ory stand­ards, self-as­sess­ment check­lists, an AI-read­i­ness scope and a Know­ledge Hub on Data.Dubai [DDSE, 1 July 2026]. The Dubai Uni­ver­sal Blue­print for AI, launched on 29 April 2024 with 22 Chief AI Of­ficers, sup­ports the D33 tar­get of 100 bil­lion dirhams a year from di­gital.

Abu Dhabi

Abu Dhabi has con­sol­id­ated its data func­tion within the De­part­ment of Gov­ern­ment En­able­ment, es­tab­lished in 2023, which ab­sorbed the former Abu Dhabi Di­gital Au­thor­ity. The emir­ate's Gov­ern­ment Di­gital Strategy 2025–2027 com­mits 13 bil­lion dirhams, tar­gets full sov­er­eign-cloud ad­op­tion for gov­ern­ment, and states the am­bi­tion of the world's first fully AI-nat­ive gov­ern­ment by 2027, with a pro­jec­ted con­tri­bu­tion above 24 bil­lion dirhams to GDP and more than 5,000 jobs [DGE, 2025]. In a con­crete step, the De­part­ment and e& an­nounced on 17 Oc­to­ber 2025 an Abu Dhabi Uni­fied Gov­ern­ment Data Centre to power more than forty gov­ern­ment en­tit­ies, pre­cisely the kind of con­sol­id­a­tion that the plat­form thesis of this edi­tion an­ti­cip­ates.

Qatar

Qatar has moved from a stand­ing start with speed. The Na­tional Plan­ning Coun­cil was es­tab­lished by Amiri De­cision 13 of 2024; the Na­tional Data and Stat­ist­ics Strategy was launched on 4 May 2025; and the Na­tional Data Pro­gram, launched in Janu­ary 2025, provides a cent­ral na­tional data re­pos­it­ory, a Na­tional Data Policy and Stand­ards, a Qatar Data Trust Index and the QInsights plat­form of 119 in­dic­at­ors [NPC, 2024–25]. Taken to­gether, the five pro­grammes show a re­gion that has largely solved the ques­tion of who gov­erns data and has not yet solved the ques­tion of

Data EngineeringVolge Quarterly04 •

whether the data be­neath the gov­ernance is en­gin­eered to be used.

How it is built today

The physical foundation is a lattice of hyperscaler regions and sovereign stacks, most of it stood up since 2019 and much of it arriving in 2026.

The es­tate on which Gulf data plat­forms are built is young. The first hy­per­scaler re­gion in the Middle East, AWS Bahrain, reached gen­eral avail­ab­il­ity only in July 2019; the densest wave of ca­pa­city ar­rives in 2026, when both the Mi­crosoft Saudi Ar­a­bia East re­gion and the AWS Saudi re­gion are due. The table sets out the prin­cipal hy­per­scaler foot­print with dates, which is the single most use­ful ref­er­ence for where a com­pli­ant, in-re­gion data plat­form can ac­tu­ally be built today.

ProviderRegionStatus & dateCommitment
Microsoft AzureUAE North / CentralLive 2019UAE spend >$15.2B by end-2029; G42 deal $1.5B (Apr 2024)
Microsoft AzureQatar CentralGA Aug 2022Cloud for Sovereignty GA Dec 2023
Microsoft AzureSaudi Arabia East (Riyadh)Customer availability targeted Nov 2026 (3 AZs)Construction complete on 3 sites, Dec 2024
AWSBahrain (me-south-1)GA Jul 2019 (first ME, 3 AZs)–
AWSUAE (me-central-1)GA Aug 2022~$5B by 2036
AWSKSA (me-central-2)Announced Mar 2024, launch 2026$5.3B
Google CloudDoha, QatarLive 2023–
Google CloudDammam, KSALive 2024JV with Center3 / stc; no UAE region
OracleUAE & KSA (Jeddah, Riyadh)Jeddah live ~2020$14B KSA over 10 years (May 2025)

Sources: pro­vider an­nounce­ments and gen­eral-avail­ab­il­ity no­tices, 2019–2026; Mi­crosoft–G42 (April 2024); Or­acle KSA com­mit­ment (May 2025). Cur­rent to July 2026.

Be­neath the hy­per­scalers sits a sov­er­eign layer that is where much gov­ern­ment data will in prac­tice reside. In Saudi Ar­a­bia this in­cludes Center3, backed by stc and tar­get­ing one gigawatt by 2030 on a ten-bil­lion-dol­lar plan of which roughly three bil­lion has been spent, along­side Alat, backed by the Pub­lic In­vest­ment Fund, and SDAIA's own ca­pa­city [com­pany dis­clos­ures, 2025]. In the UAE it in­cludes Core42 and G42's sov­er­eign cloud run­ning on Azure, e& en­ter­prise, du, Khazna and Moro Hub; in Qatar, Oore­doo and Meeza; in Bahrain, Beyon; and in Oman, Oman Data Park. The es­tate is there­fore hy­brid by de­fault, and it is this hy­brid­ity (data spread across hy­per­scaler re­gions and sov­er­eign stacks, under dif­fer­ing clas­si­fic­a­tion re­gimes) that makes the en­gin­eer­ing of a single, gov­erned data plat­form hard.

The maturity model

Read country by country, the region divides into a leading tier that has institutionalised data governance and a following tier still assembling it.

Ma­tur­ity in data en­gin­eer­ing is not the same as am­bi­tion in ar­ti­fi­cial in­tel­li­gence, and the two are fre­quently con­fused. The as­sess­ment below rates each state on the depth of its na­tional data pro­gramme, the in­teg­ra­tion achieved across gov­ern­ment sys­tems, and the read­i­ness of its found­a­tion to sup­port pro­duc­tion AI, as dis­tinct from the scale of com­pute it has pro­cured. It is to be noted that the or­der­ing tracks in­sti­tu­tional depth rather than head­line spend­ing.

CountryGoverning bodyFoundation depthAI-readiness of data
Saudi ArabiaSDAIA (NIC, NDMO, NCAI)High: 360+ systems integrated; 15 domains, 77 controls, 191 specificationsAdvancing: strong governance, engineering talent the constraint
UAE (federal)FCSC / Smart Data FrameworkMedium-High: policy-led; PDPL exec regs unpublished mid-2026Advancing: measured by Data Index; uneven across entities
DubaiDDSE / Dubai Digital AuthorityHigh: data-as-asset since 2015; Pulse platform; Manual updated Jul 2026Advancing: AI-readiness now in scope of the Data Manual
Abu DhabiDGE (absorbed ADDA, 2023)Medium-High: unified data centre for 40+ entities (Oct 2025)Ambitious: "AI-native government" targeted for 2027
QatarNational Planning CouncilMedium: National Data Program stood up Jan 2025Early: repository and standards new; QInsights live
BahrainiGA / BeyonMedium: first-mover on cloud (AWS 2019), smaller estateEarly: narrower programme, limited disclosure
OmanMTCIT / Oman Data ParkLow-Medium: nascent national programmeEarly: foundation-building stage

Volge as­sess­ment, July 2026, draw­ing on the na­tional-pro­gramme sources cited in sec­tion 02. Rat­ings are re­l­at­ive within the re­gion and re­flect data-found­a­tion depth, not com­pute pro­cure­ment or AI am­bi­tion.

Data EngineeringVolge Quarterly05 •
Ambition in artificial intelligence has been mistaken for maturity in data engineering across the region; they are not the same measure.

The prac­tical im­plic­a­tion is that the lead­ing tier (Saudi Ar­a­bia, Dubai and Abu Dhabi) has the gov­ernance scaf­fold­ing largely in place and now faces an ex­e­cu­tion prob­lem in en­gin­eer­ing and tal­ent, while the fol­low­ing tier still faces an in­sti­tu­tional one. For both, the con­straint of 2027 sits at the level of en­gin­eer­ing ca­pa­city rather than of policy.

Where it falls short

The foundation gap is measurable, and the clearest measure is the rate at which generative-AI initiatives fail for want of sound data.

The single most im­port­ant fig­ure in this edi­tion is that roughly 95 per cent of or­gan­isa­tions in MIT Pro­ject NANDA's evid­ence base re­cor­ded no meas­ur­able profit-and-loss im­pact from gen­er­at­ive AI [July 2025], an out­come at­trib­uted to weak data found­a­tions rather than to model qual­ity [MIT NANDA, "State of AI in Busi­ness 2025", via For­bes, 15 Oc­to­ber 2025]. The fig­ure is not a com­ment on the soph­ist­ic­a­tion of the mod­els being de­ployed; it is a com­ment on the state of the data being fed to them. Where the data is frag­men­ted, poorly cata­logued, in­con­sist­ently clas­si­fied and un­gov­erned, no model, how­ever cap­able, can pro­duce a re­li­able res­ult, and the pilot stalls be­fore it reaches pro­duc­tion.

  • GenAI pi­lots re­turn­ing noth­ing: ~95%, At­trib­uted to data found­a­tions, not mod­els [MIT NANDA, Oct 2025]
  • GenAI pro­jects that fail: ~50%, [Gart­ner, 26 Janu­ary 2026]
  • Agen­tic-AI pro­jects can­celled by end-2027: >40%, [Gart­ner, 25 June 2025]
  • An­nual cost of poor data qual­ity, per org: $12.9M, [Gart­ner, via IBM, 23 Janu­ary 2026]
  • Pro­jec­ted cloud waste, 2025: $44.5B, Spend out­pa­cing value [Har­ness, 2025]
  • Ma­chine iden­tit­ies per human: 109:1, The scale of the un­gov­erned es­tate [Palo Alto Net­works, 2026]

MIT NANDA / For­bes (15 Oc­to­ber 2025); Gart­ner (26 Janu­ary 2026; 7 April 2026; 25 June 2025); McKinsey (13 Novem­ber 2025); Gart­ner via IBM (23 Janu­ary 2026); Har­ness (2025); Palo Alto Net­works, 2026 Iden­tity Se­cur­ity Land­scape. Two de­vel­op­ments postdate the re­search and bear on every Saudi and UAE entry: SDAIA con­sul­ted from 6 Oc­to­ber to 5 Novem­ber 2026 on draft amend­ments that would re­quire per­sonal data to be kept in the King­dom, and the UAE es­tab­lished a Fed­eral Au­thor­ity for Ar­ti­fi­cial In­tel­li­gence and Data on 14

June 2026, which ab­sorbs the Data Of­fice and is the ex­pec­ted is­suer of the PDPL ex­ec­ut­ive reg­u­la­tions. Cur­rent to Novem­ber 2026.

The cor­rob­or­at­ing evid­ence is con­sist­ent. Gart­ner has sep­ar­ately ob­served that AI in­fra­struc­ture-and-op­er­a­tions pro­jects stall be­fore reach­ing re­turn on in­vest­ment, and that more than 40 per cent of agen­tic-AI pro­jects will be can­celled by the end of 2027 [Gart­ner, 7 April 2026; 25 June 2025]. McKinsey, ex­amin­ing the Gulf spe­cific­ally, has found cap­ital ex­pendit­ure ra­cing ahead of data and gov­ernance ma­tur­ity [McKinsey, 13 Novem­ber 2025]. The gaps be­neath these fig­ures are well un­der­stood by prac­ti­tion­ers: a short­age of data-en­gin­eer­ing tal­ent; over-re­li­ance on sys­tem in­teg­rat­ors; gov­ernance treated as an af­ter­thought; tool sprawl; mi­gra­tion and tech­nical debt; and, above all, data qual­ity in­ad­equate for ground­ing an AI model. Each is an en­gin­eer­ing prob­lem, and each is why the found­a­tion, not the model, is the ceil­ing.

The forces shaping the year

Five through-lines cut across the region. Each is a force that will move data engineering in 2027, and together they explain the Calls that follow.

Sovereignty by design

Classification-driven data residency is moving from a compliance overlay to a first principle of architecture.

The Gulf can buy any model on the market. It cannot buy, at the speed it wishes, the engineered data foundation those models require to be worth anything.

Sov­er­eignty in the Gulf is ceas­ing to be a policy stated after a sys­tem is built and is be­com­ing a con­straint de­signed into it from the first line. The mech­an­ism is data clas­si­fic­a­tion: Saudi Ar­a­bia's four-level scheme (Top Secret, Secret, Con­fid­en­tial or Re­stric­ted, and Pub­lic) and the par­al­lel schemes in the UAE, Dubai and Qatar de­term­ine not merely who may see a re­cord but where, phys­ic­ally and jur­is­dic­tion­ally, that re­cord may be pro­cessed and stored [SDAIA; DDSE; FCSC, 2015–26]. When res­id­ency is a func­tion of clas­si­fic­a­tion, and clas­si­fic­a­tion is man­dated at the level of the in­di­vidual data ele­ment, sov­er­eignty be­comes an en­gin­eer­ing prop­erty of the plat­form rather than a legal wrap­per around it.

The prac­tical con­sequence for data en­gin­eer­ing is sig­ni­fic­ant. A plat­form built for clas­si­fic­a­tion-driven res­id­ency must carry the clas­si­fic­a­tion with the data through every pipeline, en­force it at query time, and

Data EngineeringVolge Quarterly06 •

prove it under audit, which de­mands cata­logue, lin­eage and policy en­gines of a ma­tur­ity that most Gulf es­tates do not yet pos­sess. The Dubai Data Manual's ad­di­tion of AI-read­i­ness scope and self-as­sess­ment check­lists on 1 July 2026 is an ac­know­ledge­ment that these con­trols must now be demon­strable rather than as­ser­ted [DDSE, 1 July 2026].

The evid­ence that vendors are en­gin­eer­ing for this is dir­ect. Mi­crosoft's Cloud for Sov­er­eignty reached gen­eral avail­ab­il­ity in Decem­ber 2023, and its 1.5-bil­lion-dol­lar ar­range­ment with G42 in April 2024 was ex­pli­citly framed around sov­er­eign con­trols; Or­acle of­fers Alloy, Ded­ic­ated Re­gion and Sov­er­eign Cloud pre­cisely to place a com­pli­ant bound­ary around reg­u­lated data [pro­vider dis­clos­ures, 2023–25]. Abu Dhabi has gone fur­ther in policy, tar­get­ing 100 per cent sov­er­eign-cloud ad­op­tion for gov­ern­ment within its 2025–2027 strategy [DGE, 2025]. Sov­er­eignty is thus not a niche re­quire­ment but the de­fault pos­ture of the re­gion's largest data owner, its gov­ern­ment.

When residency is a function of classification, sovereignty stops being a legal wrapper and becomes an engineering property of the platform.

It is to be noted that this force is, on bal­ance, con­struct­ive. Clas­si­fic­a­tion-driven design forces the cata­loguing, lin­eage and gov­ernance work that a sound data found­a­tion re­quires in any case, and which is too often de­ferred. The en­tit­ies that treat sov­er­eignty as an ar­chi­tec­tural dis­cip­line rather than a doc­u­ment­a­tion ex­er­cise will, as a by-product, build the cleaner and bet­ter-gov­erned found­a­tion that pro­duc­tion AI de­pends upon.

The talent bottleneck

The scarcest input in the Gulf data estate is not compute or capital but the engineers who can build and run the platforms.

The bind­ing con­straint on Gulf data en­gin­eer­ing in 2027 is human, not phys­ical. Cap­ital is ef­fect­ively un­con­strained and com­pute is being pro­cured at na­tional scale, but the sup­ply of data en­gin­eers cap­able of design­ing pipelines, gov­ern­ing cata­logues and op­er­at­ing plat­forms is thin re­l­at­ive to de­mand, and it can­not be ex­pan­ded at the speed cap­ital can be de­ployed. Saudi Ar­a­bia's Na­tional Strategy for Data and AI it­self tar­gets the train­ing of more than 20,000 data and AI spe­cial­ists by 2030, a fig­ure that is both an

am­bi­tion and an ad­mis­sion of the size of the present short­fall [NSDAI, 2020].

The gap is struc­tural rather than cyc­lical. Loc­al­isa­tion pro­grammes across the Gulf are pro­du­cing gradu­ates, but the spe­cific dis­cip­line of data en­gin­eer­ing (as op­posed to data sci­ence or gen­eral soft­ware de­vel­op­ment) has a long ap­pren­tice­ship, and the re­gional pool of senior prac­ti­tion­ers who have built and run pro­duc­tion plat­forms at scale re­mains small. The con­sequence has been a heavy re­li­ance on ex­pat­ri­ate tal­ent and on sys­tem in­teg­rat­ors to sup­ply cap­ab­il­ity that the en­tity does not hold in-house, which de­fers rather than re­solves the short­age.

The scale of the es­tate that this thin tal­ent base must gov­ern is eas­ily un­der­es­tim­ated. There are now roughly 109 ma­chine iden­tit­ies for every human in a mod­ern en­ter­prise [Palo Alto Net­works, 2025], each gen­er­at­ing data that must be in­ges­ted, se­cured and gov­erned; and the cost of fail­ing to do so has been es­tim­ated at about 12.9 mil­lion US dol­lars per or­gan­isa­tion each year in poor data qual­ity alone [Gart­ner, via IBM, 23 Janu­ary 2026]. The mis­match between the size of the es­tate and the size of the work­force able to en­gin­eer it is the mech­an­ism by which the tal­ent gap be­comes a found­a­tion gap.

Capital can buy compute in a quarter and a platform in a year; it cannot manufacture a senior data engineer in either.

This is the force least amen­able to a fast solu­tion, and it is there­fore the one held with the highest con­fid­ence in the Calls that fol­low. Cap­ital can buy com­pute in a quarter and a data plat­form in a year; it can­not man­u­fac­ture a senior data en­gin­eer in either. Through 2027 the de­mand for data-en­gin­eer­ing tal­ent is ex­pec­ted to con­tinue out­strip­ping loc­ally avail­able sup­ply des­pite loc­al­isa­tion, and the en­tit­ies that suc­ceed will be those that build dur­able in-house teams rather than rent­ing cap­ab­il­ity by the pro­ject.

From projects to platforms

The operating model is shifting from the one-off, integrator-led build towards the durable, in-house data platform.

The dom­in­ant way Gulf data cap­ab­il­ity has been de­livered (as a dis­crete pro­ject, scoped, built and handed over by a sys­tem in­teg­rator) is reach­ing the limit of its use­ful­ness. A pro­ject ends; a plat­form com­pounds. As en­tit­ies dis­cover that the value of their data lies in its con­tinu­ous en­gin­eer­ing rather than in a single

Data EngineeringVolge Quarterly07 •

de­livered sys­tem, the op­er­at­ing model is be­gin­ning to shift to­wards owned plat­forms staffed by in-house teams, with the in­teg­rator moved from builder to part­ner. It is to be noted that this shift is early and un­even, but its dir­ec­tion is clear.

The struc­tural driver is that AI-grade data found­a­tions are not a state to be reached and left, but a cap­ab­il­ity to be op­er­ated. Cata­logues drift, clas­si­fic­a­tions change, sources mul­tiply, and mod­els re­quire con­tinu­ously cur­ated, high-qual­ity data to re­main re­li­able. An es­tate that has 109 ma­chine iden­tit­ies per human and in­teg­rates hun­dreds of source sys­tems (Saudi Ar­a­bia's Na­tional Data Bank alone spans more than 360) can­not be main­tained by a vendor en­gage­ment that con­cludes on han­dover [Palo Alto Net­works, 2026; SDAIA, 2025].

The con­sol­id­a­tion now vis­ible in the phys­ical layer pre­fig­ures the same move in the lo­gical one. The Abu Dhabi Uni­fied Gov­ern­ment Data Centre, an­nounced by the De­part­ment of Gov­ern­ment En­able­ment and e& on 17 Oc­to­ber 2025 to power more than forty gov­ern­ment en­tit­ies, is a plat­form play rather than a pro­ject: shared in­fra­struc­ture, op­er­ated cent­rally, serving many con­sumers [DGE, 2025]. The same logic (build once, op­er­ate con­tinu­ously, serve many) is what dis­tin­guishes a data plat­form from a data pro­ject, and it is the logic the re­gion is slowly ad­opt­ing.

A project ends; a platform compounds. The value of Gulf data lies in its continuous engineering, not in a system delivered once.

The bar­rier is the tal­ent bot­tle­neck of the pre­ced­ing force. A plat­form op­er­at­ing model re­quires a stand­ing team, and stand­ing teams are pre­cisely what the re­gion is short of. The Calls there­fore treat this shift as real but gradual, ex­pect­ing it to ap­pear first at a small num­ber of flag­ship en­tit­ies in 2027 rather than across the es­tate.

Building for AI

The artificial-intelligence layer the region is buying will only perform on a data foundation engineered to feed it.

The cent­ral irony of the Gulf's po­s­i­tion is that the more it in­vests in ar­ti­fi­cial in­tel­li­gence, the more ex­posed it be­comes to the weak­ness of its data found­a­tion. A model is only as good as the data it is groun­ded in, and the fail­ure rate of gen­er­at­ive-AI ini­ti­at­ives (roughly 95 per cent of or­gan­isa­tions re­cord­ing no meas­ur­able profit-and-loss im­pact, at­trib­uted to

data found­a­tions rather than to model qual­ity) is the clearest pos­sible state­ment that the con­straint has moved down­stream of the model [MIT NANDA, via For­bes, 15 Oc­to­ber 2025]. Build­ing for AI, cor­rectly un­der­stood, is there­fore an in­struc­tion to build the data found­a­tion first.

The cor­rob­or­a­tion is broad and re­cent. Gart­ner has found that about half of gen­er­at­ive-AI pro­jects fail, that AI in­fra­struc­ture-and-op­er­a­tions pro­jects stall be­fore reach­ing re­turn on in­vest­ment, and that more than 40 per cent of agen­tic-AI pro­jects will be can­celled by the end of 2027 [Gart­ner, 26 Janu­ary 2026; 7 April 2026; 25 June 2025]. McKinsey, ex­amin­ing the Gulf, has found cap­ital ex­pendit­ure ra­cing ahead of data and gov­ernance ma­tur­ity [McKinsey, 13 Novem­ber 2025]. None of these find­ings im­plic­ates the mod­els; all im­plic­ate the found­a­tion.

The more the Gulf invests in artificial intelligence, the more exposed it becomes to the weakness of the foundation beneath it.

The en­gin­eer­ing re­quire­ment that "build­ing for AI" im­poses is ex­act­ing. An AI-ready found­a­tion de­mands high data qual­ity, com­plete and cur­rent metadata, en­forced clas­si­fic­a­tion, trace­able lin­eage, and the re­trieval in­fra­struc­ture to serve groun­ded con­text to a model at query time. These are the dis­cip­lines of data en­gin­eer­ing, and they are the dis­cip­lines the re­gion has under-in­ves­ted in re­l­at­ive to com­pute. The ad­di­tion of an ex­pli­cit AI-read­i­ness scope to the Dubai Data Manual on 1 July 2026 is a sign that the more ad­vanced en­tit­ies now re­cog­nise this [DDSE, 1 July 2026].

The im­plic­a­tion for 2027 is dir­ect and ap­pears in the Calls. A ma­jor­ity of the en­ter­prise gen­er­at­ive-AI ini­ti­at­ives launched across the Gulf in 2025 and 2026 are ex­pec­ted to stall or be re-scoped pending found­a­tion work, and at least one na­tional en­tity is ex­pec­ted to re­port pub­licly an AI fail­ure or re­scope ex­pli­citly at­trib­uted to data qual­ity or en­gin­eer­ing. Both are the pre­dict­able con­sequence of load­ing in­tel­li­gence onto a found­a­tion that was built last.

Cost discipline

As cloud-and-data spend outpaces value, financial discipline arrives for the data estate as a named function.

The era of un­metered data spend­ing in the Gulf is end­ing. With cap­ital abund­ant and com­pute cheap, data and cloud budgets have grown with little scru­tiny of the value re­turned; in 2027 that scru­tiny

Data EngineeringVolge Quarterly08 •

ar­rives. The trig­ger is the vis­ible gap between spend and value (pro­jec­ted global cloud waste of 44.5 bil­lion US dol­lars in 2025, with Flex­era put­ting waste at 29 per cent of global in­fra­struc­ture and plat­form spend in 2026, the highest in five years, and an es­tim­ated 12.9 mil­lion US dol­lars a year in poor-data-qual­ity cost per or­gan­isa­tion), which is now large enough that fin­ance func­tions can no longer treat data as an un­ques­tioned over­head [Har­ness, 2025; Gart­ner via IBM, 23 Janu­ary 2026].

The dis­cip­line has a name: FinOps, ap­plied to data. It treats the data es­tate as a cost centre to be meas­ured, at­trib­uted and op­tim­ised: stor­age tiered to ac­cess, pipelines ra­tion­al­ised, du­plic­ate tool­ing re­tired, and spend at­trib­uted to the busi­ness unit that in­curs it. The mech­an­ism mat­ters be­cause the Gulf's cost ad­vant­age in raw in­puts, with in­dus­trial power at 0.05 to 0.07 US dol­lars per kilo­watt-hour, has masked in­ef­fi­ciency in how those in­puts are con­sumed [Ox­ford Eco­nom­ics, 22 Janu­ary 2026]. Cheap power makes waste af­ford­able, which is pre­cisely why the waste has ac­cu­mu­lated.

Tool sprawl is the clearest tar­get. The pro­lif­er­a­tion of over­lap­ping in­ges­tion, cata­logue, qual­ity and ob­serv­ab­il­ity tools (each pro­cured for a pro­ject, none re­tired when the pro­ject ended) is both a cost and an en­gin­eer­ing li­ab­il­ity, be­cause every ad­di­tional tool is an­other sur­face to gov­ern across an es­tate already car­ry­ing 109 ma­chine iden­tit­ies per human [Palo Alto Net­works, 2026]. Con­sol­id­a­tion onto a smal­ler, gov­erned plat­form stack re­duces spend and im­proves the found­a­tion sim­ul­tan­eously, which aligns the fin­an­cial and en­gin­eer­ing cases.

Cheap power made waste affordable, which is precisely why the waste accumulated. In 2027 someone starts counting.

The Call that fol­lows ex­pects FinOps-for-data to be­come a named func­tion or ex­pli­cit line item at major Gulf en­tit­ies dur­ing 2027, as the gap between data spend and data value be­comes too large to leave un­man­aged. It is the mech­an­ism by which cost dis­cip­line, ar­riv­ing late, be­gins to force the found­a­tion work that abund­ance had al­lowed the re­gion to defer.

Eight rows for six states: the United Arab Emir­ates gov­erns data at three levels, and the fed­eral in­stru­ment, the Abu Dhabi policy and the Dubai policies doc­u­ment im­pose dif­fer­ent du­ties on dif­fer­ent en­tit­ies.

The buyer's ground

What the rules require of data itself, where the region actually stands, what the platforms cost in the Gulf, and which maturity model a buyer can use without a licence.

What the rules actually require of your data

Three of the six GCC states mandate a data catalogue in an instrument a buyer can read. None of the six mandates data lineage, and no state has published a penalty under a data-management instrument as distinct from a privacy law.

The first ques­tion to settle in a data pro­gramme is not what good prac­tice re­com­mends but what the jur­is­dic­tion com­pels. On that test the Gulf po­s­i­tion is con­sid­er­ably nar­rower than the mar­ket as­sumes. Three of the six GCC states man­date a data cata­logue in an in­stru­ment that can be read in full: Abu Dhabi, Oman and Kuwait. Not one of the six man­dates data lin­eage. And no GCC state has pub­lished a pen­alty at­tached spe­cific­ally to a data-man­age­ment or data-clas­si­fic­a­tion in­stru­ment, as op­posed to a pri­vacy law. The table below sets out the po­s­i­tion state by state, against the in­stru­ment, what is man­dated, who is bound and how com­pli­ance is as­sessed. It is con­fined to what was read­able. Where an ob­lig­a­tion rests on vendor re­state­ment rather than on an op­er­at­ive clause, that is said in terms.

StateInstrumentWhat is mandatedWho is boundAssessment regime
Saudi ArabiaNational Data Governance Interim Regulations, SDAIA/NDMO; Data Management and Personal Data Protection Standards, NDMO, v1.0 June 2020A Data Management Office, a Data Management Committee and a Chief Data Officer at each entity. Catalogue, retention and master data appear in vendor restatements, not verified against the instrumentAll Kingdom public entities, extending to business partners handling government data (vendor source)Annual compliance audit, report submitted to NDMO in Q3 each year; ad-hoc audits possible; NDMO states it reports at entity, sector and whole-of-government level
UAE, federalFederal Decree-Law 44 of 2021, in force 2 January 2022Creation of the Data Office and the empowering of a regulator. No catalogue, quality or lineage duty of its ownFederal levelNone stated in the instrument
Data EngineeringVolge Quarterly09 •
StateInstrumentWhat is mandatedWho is boundAssessment regime
Abu DhabiGovernment Data Management Policy v1.0, ADSIC-DM-PD03; issue date not publicThirteen domains. A bilingual Data Catalogue, mandatory metadata keys, data quality metrics, lifecycle and destruction controls, reference and master data toolsets, a single enterprise data modelAbu Dhabi government entities, which must bind their suppliersProgramme status reported against Data Management Programme Plan milestones, frequency stated only as regular; central assessment of programmes and technical controls
DubaiLaw 26 of 2015 and Resolution 2 of 2017 Policies DocumentClassification before dissemination or exchange (Article 7). Open Data and Shared Data, the latter split Confidential, Sensitive and SecretDubai government entitiesNo assessment regime found; no chief data officer mandate found; no penalty found
QatarNational Data Policy, October 2024; National Data Management Guidelines, July 2025; National Data Classification Policy v3.0, launched 7 May 2023Reported as five layers and twelve domains binding entities to national standards. Guidelines published per domain, including Data Catalog and Metadata Management; mandatory or advisory not established. Five classification tiers, C0 Public to C4 Top SecretAll government entities; the classification policy also binds vital-sector entities, with a minimum of four levels for non-governmentInternal review against the National Data Standards and Qatar Data Trust Index scoring; cycle not dated
BahrainNone issued. The iGA AI Readiness Assessment Methodology report of 12 November 2025 describes a National Data Governance Framework as still in developmentNo catalogue, quality, chief data officer or master data mandate foundNot applicableNone
StateInstrumentWhat is mandatedWho is boundAssessment regime
OmanRegulatory Framework for National Data Governance and Management, Ministerial Circular 113/2025, MTCIT; policies document dated 1 January 2025Thirteen policies, including Data Catalogue and metadata, Data Classification, Data Quality, Reference and Master Data, Open Data and Data Monetisation. A Data Governance and Management Office with named rolesAll government units; the private sector reached indirectly through sector regulatorsAnnual periodic assessment by the ministry's compliance team; provisions split across three years, Priority 1 in year one, Priority 2 in year two, Priority 3 in year three
KuwaitData Classification Policy v2.3, CITRA; no issue date and no commencement date on the documentThe data owner must create and maintain a data catalogue including metadata in a unified format, updated periodically (clause 3.1.3). Encryption of Tiers 3 and 4 in transit, a classification team headed by senior management, a classification road mapPublic and private sector entities. The only GCC data-classification instrument binding private business directlyQuarterly implementation reports to CAIT; CITRA periodic compliance reports must contain the full catalogue with metadata, approved tiers with criteria, storage locations by tier and progress against the road map (clause 3.4.2.3)

What the table shows

What the table shows is a re­gion le­gis­lat­ing at three quite dif­fer­ent levels of spe­cificity. At one end sit Abu Dhabi, Oman and Kuwait, whose in­stru­ments can be read clause by clause and whose ob­lig­a­tions can there­fore be cos­ted. At the other sits Bahrain, which has is­sued noth­ing, and the UAE fed­eral layer, where De­cree-Law 44 of 2021 is in­sti­tu­tional rather than op­er­a­tional: it cre­ates a reg­u­lator and em­powers it, and im­poses no duty on an en­tity's data es­tate at all. Between the two sit Saudi Ar­a­bia and Qatar, both of which have built sub­stan­tial ap­par­atus whose op­er­at­ive text is not reach­able. The po­s­i­tion is that the weight of a state's data re­gime and the legib­il­ity of it are not cor­rel­ated, and the two states with the largest pro­grammes are the two about which least can be stated with con­fid­ence.

Data EngineeringVolge Quarterly10 •

On cata­loguing, the three man­dates dif­fer in char­ac­ter. Kuwait's is the bluntest: clause 3.1.3 of the CITRA policy puts the duty on the data owner to cre­ate and main­tain a cata­logue in­clud­ing metadata in a uni­fied format, and clause 3.4.2.3 re­quires the cata­logue it­self to be sur­rendered in peri­odic com­pli­ance re­port­ing. Abu Dhabi's is the most de­mand­ing: a bi­lin­gual cata­logue cov­er­ing own­er­ship, proven­ance, com­pos­i­tion and qual­ity re­quire­ments, which must be checked be­fore any new data defin­i­tion is cre­ated. Oman's sits within a numbered policy on Data Cata­logue and metadata in­side a thir­teen-policy frame­work. Saudi Ar­a­bia is widely de­scribed as man­dat­ing a cata­logue, but the ob­lig­a­tion ap­pears in vendor re­state­ments rather than in any NDMO text that could be read, and that dis­tinc­tion mat­ters when a buyer is asked to jus­tify a budget.

The lineage gap

Data lin­eage, the re­cor­ded path by which a fig­ure in a re­port traces back to the sys­tem that pro­duced it, is not an ex­press ob­lig­a­tion any­where in the GCC. It was not found in any read­able in­stru­ment in any of the six states. Abu Dhabi comes closest: its cata­logue re­quire­ment ex­tends to proven­ance, which is ad­ja­cent to lin­eage without being it. The con­sequence is worth stat­ing plainly, be­cause it cuts against the dir­ec­tion of the mar­ket. Lin­eage is the con­trol that an­swers the ques­tion an aud­itor, a reg­u­lator or a board will even­tu­ally ask of an AI out­put, which is where the num­ber came from. No Gulf state cur­rently com­pels it. A buyer who de­fers it until it is com­pelled will be build­ing it under time pres­sure, ret­ro­spect­ively, across sys­tems already in pro­duc­tion.

The region's data-management instruments are mandatory and unenforced, while its privacy laws are enforced and largely unpublished.

An enforcement record that is empty

No GCC state pub­lishes a pen­alty at­tached spe­cific­ally to a data-man­age­ment or data-clas­si­fic­a­tion in­stru­ment. Oman pub­lishes no pen­alty under Cir­cu­lar 113/2025. Dubai pub­lishes none under Res­ol­u­tion 2 of 2017. The Abu Dhabi policy is man­dat­ory and car­ries no pub­lished sanc­tion. The con­trast with pri­vacy law is sharp. Saudi Ar­a­bia's Per­sonal Data Pro­tec­tion Law, Royal De­cree M/19 of 2021 as amended by M/148 of 2023, car­ries fines to SAR 5m which may be doubled, and SDAIA an­nounced 48 en­force­ment de­cisions in

the year to early 2026 (Latham and Watkins, 15 May 2026), al­though the de­cision texts and the re­spond­ents are not pub­lic. Bahrain's Law 30 of 2018, in force 1 Au­gust 2019, car­ries up to one year's im­pris­on­ment and fines to BHD 20,000. The asym­metry is in­struct­ive: the Gulf has chosen to put its sanc­tions be­hind per­sonal data and its am­bi­tions be­hind data man­age­ment.

Who else is bound

It is to be noted that these in­stru­ments do not stop at the peri­meter of gov­ern­ment. Abu Dhabi re­quires en­tit­ies to bind their sup­pli­ers, which places a con­tractor in­side the policy whether or not the con­tract re­peats its terms. Kuwait's clas­si­fic­a­tion policy binds private sec­tor en­tit­ies dir­ectly and is, on the evid­ence avail­able, the only GCC data-clas­si­fic­a­tion in­stru­ment that does so. Oman reaches the private sec­tor in­dir­ectly, through sec­tor reg­u­lat­ors. Saudi Ar­a­bia is re­por­ted to ex­tend to busi­ness part­ners hand­ling gov­ern­ment data, though that reach too rests on a vendor source rather than on a clause. For a com­mer­cial buyer with pub­lic sec­tor rev­enue in the Gulf, the prac­tical read­ing is that the ob­lig­a­tion ar­rives by con­tract be­fore it ar­rives by stat­ute.

The struc­ture quoted al­most uni­ver­sally for the Saudi NDMO Stand­ards, fif­teen do­mains, 77 con­trols and 191 spe­cific­a­tions, ap­pears only in vendor com­pli­ance ma­ter­ial and in no reach­able NDMO doc­u­ment. SDAIA's own Na­tional Data Index gives four­teen do­mains. The fig­ures are treated here as vendor-ori­gin­ated. Sim­il­arly, the Saudi clas­si­fic­a­tion tiers con­flict between two of­fi­cial SDAIA doc­u­ments, one giv­ing Top Secret, Secret, Con­fid­en­tial and Pub­lic, the other Top Secret, Secret, Re­stric­ted and Pub­lic. The two are un­re­con­ciled.

The takeaway for a buyer is that the com­pli­ance floor in the Gulf is a cata­logue, a clas­si­fic­a­tion scheme and a gov­ernance body with a named ac­count­able of­ficer, and that noth­ing bey­ond that floor is cur­rently com­pelled any­where in the re­gion. Everything above it, lin­eage in­cluded, is built for the buyer's own de­fens­ib­il­ity rather than for a reg­u­lator's. That is a bet­ter reason to build it, and a harder one to fund.

Abu Dhabi, read clause by clause

The Abu Dhabi Government Data Management Policy is the most demanding readable data instrument in the GCC. Its obligations pass down to suppliers, which means a vendor to an Abu Dhabi entity is inside the policy whether or not the contract repeats it.

If a buyer in the Gulf is to read one data in­stru­ment in full, it should be the Abu Dhabi Gov­ern­ment Data Man­age­ment Policy v1.0, doc­u­ment ref­er­ence ADSIC-DM-PD03. It is the richest read­able data in­stru­ment in the re­gion: thir­teen do­mains, man­dat­ory in ap­plic­a­tion, and spe­cific enough that its re­quire­ments can be con­ver­ted into a work plan without in­ter­pret­a­tion. Two fea­tures set it apart from everything else in the GCC.

Data EngineeringVolge Quarterly11 •

The first is that its cata­logue re­quire­ment op­er­ates as a design con­trol rather than a doc­u­ment­a­tion ex­er­cise. The second is that en­tit­ies must bind their sup­pli­ers to it, which ex­tends the policy well bey­ond the gov­ern­ment payroll. One oddity should be re­cor­ded at the out­set: the re­lease date field in the ver­sion table of the doc­u­ment is blank, so the issue date of the most de­mand­ing data in­stru­ment in the Gulf is not pub­lic.

The governance clauses

The policy be­gins where a buyer would ex­pect an in­stru­ment of this kind to begin, with ac­count­ab­il­ity, and then goes fur­ther than most. An en­tity must ap­point a Data Man­ager who is ac­count­able to the en­tity board. The seni­or­ity of that re­port­ing line is the point: the policy does not per­mit data man­age­ment to be housed as a tech­nical func­tion re­port­ing into an in­form­a­tion tech­no­logy dir­ect­or­ate. An en­tity must also con­sti­tute a Data Gov­ernance Board. And it must set key per­form­ance in­dic­at­ors for every one of the thir­teen do­mains, which con­verts each do­main from a state­ment of prin­ciple into some­thing meas­ur­able. The prac­tical ef­fect is that an en­tity can­not claim par­tial ad­op­tion. Thir­teen do­mains with in­dic­at­ors at­tached to each pro­duce thir­teen sep­ar­ate re­port­ing ob­lig­a­tions, and cent­ral as­sess­ment cov­ers both pro­grammes and tech­nical con­trols.

The catalogue clause, and why it is the hardest

The cata­logue re­quire­ment is the clause most likely to be un­der­es­tim­ated. An en­tity must main­tain a bi­lin­gual Data Cata­logue cov­er­ing own­er­ship, proven­ance, com­pos­i­tion and qual­ity re­quire­ments, and it must pop­u­late man­dat­ory metadata keys. Each of those four cov­er­age re­quire­ments car­ries its own cost. Own­er­ship re­quires a named ac­count­able per­son for every data asset, which in most es­tates does not exist and can­not be in­ferred. Proven­ance re­quires a re­cord of where the data came from. Com­pos­i­tion re­quires a de­scrip­tion of what is in it. Qual­ity re­quire­ments re­quire a state­ment of the stand­ard the asset is meant to meet, which pre­sup­poses that someone has de­cided what that stand­ard is. The bi­lin­gual re­quire­ment doubles the draft­ing work.

A catalogue that must be consulted before any new data definition is created is not a documentation exercise: it is a control on the design of systems.

The clause that changes be­ha­viour, how­ever, is the re­quire­ment that the cata­logue be checked be­fore any new data defin­i­tion is cre­ated. That is not a re­port­ing duty. It is a gate in the de­vel­op­ment li­fe­cycle. An en­tity com­ply­ing with it can­not stand up a new table, a new field or a new ex­tract without first es­tab­lish­ing that the concept does not already exist in the cata­logue under an­other name. It is the only pro­vi­sion in any read­able GCC in­stru­ment that at­tacks the ac­tual mech­an­ism by which data es­tates be­come un­man­age­able, which is un­con­trolled pro­lif­er­a­tion of near-du­plic­ate defin­i­tions.

Quality, lifecycle and the National Archives

On qual­ity the policy re­quires data qual­ity met­rics sup­por­ted by a pri­or­it­ised pro­fil­ing and cleans­ing plan. The word pri­or­it­ised is doing work there: the policy does not re­quire an en­tity to clean everything, which would be un­achiev­able, but it does re­quire the en­tity to have de­cided and re­cor­ded what it will clean first. On li­fe­cycle, an en­tity must doc­u­ment li­fe­cycle at­trib­utes in­clud­ing archive and de­struc­tion, and the de­struc­tion pro­vi­sion in­tro­duces a third party into what is oth­er­wise an in­ternal pro­cess: writ­ten ap­proval from the Na­tional Archives is re­quired be­fore de­struc­tion. A re­ten­tion sched­ule built on the as­sump­tion that an en­tity may de­lete its own data at the end of a defined period will not com­ply.

Master data and the single model

Two fur­ther re­quire­ments carry sig­ni­fic­ant im­ple­ment­a­tion weight. An en­tity must stand up ref­er­ence and mas­ter data tool­sets with match-and-merge rules, which is a tool­ing ob­lig­a­tion and not merely a policy one: match-and-merge im­plies a res­ol­u­tion en­gine and a doc­u­mented rule set, with all the gov­ernance that at­taches to de­cid­ing when two re­cords de­scribe the same cus­tomer or the same asset. And an en­tity must hold its en­ter­prise data model in a single re­pos­it­ory. The word single is the op­er­at­ive one. Most gov­ern­ment es­tates of any age hold sev­eral par­tial mod­els in sev­eral tools, main­tained by sev­eral teams, and the con­sol­id­a­tion of those into one re­pos­it­ory is typ­ic­ally a longer ex­er­cise than the cata­loguing it­self.

Why your suppliers are inside the policy

The pro­vi­sion with the widest com­mer­cial reach is the shortest. The policy is man­dat­ory and en­tit­ies must bind their sup­pli­ers. The con­sequence is that a sys­tems in­teg­rator, an ana­lyt­ics vendor, a man­aged ser­vice pro­vider or a cloud host­ing part­ner work­ing for an Abu Dhabi gov­ern­ment en­tity is op­er­at­ing in­side an

Data EngineeringVolge Quarterly12 •

in­stru­ment it may never have read. Re­port­ing re­in­forces this: the cent­ral as­sess­ment cov­ers pro­grammes and tech­nical con­trols, and the con­tract­ors sit in­side both. A buyer on the sup­ply side should there­fore treat the Abu Dhabi policy as a tender re­quire­ment rather than a cli­ent-side com­pli­ance mat­ter, be­cause the cata­logue, the metadata keys and the li­fe­cycle doc­u­ment­a­tion will in prac­tice be pro­duced by the sup­plier even where the ob­lig­a­tion rests on the en­tity.

What is not in the policy

Three ab­sences de­serve re­cord­ing, be­cause they are as use­ful to a buyer as the ob­lig­a­tions. First, the re­port­ing fre­quency is stated only as reg­u­lar: en­tit­ies re­port pro­gramme status against Data Man­age­ment Pro­gramme Plan mile­stones, and no in­ter­val is pub­lished. An en­tity can­not there­fore plan its re­port­ing cal­en­dar from the in­stru­ment. Second, no pen­alty is pub­lished. The policy is man­dat­ory and car­ries no stated sanc­tion, which is the re­gional pat­tern de­scribed in the pre­ced­ing chapter. Third, data lin­eage is not re­quired. Proven­ance in the cata­logue is the nearest thing to it, and proven­ance re­cords an ori­gin rather than a path. The most de­mand­ing data in­stru­ment in the Gulf does not com­pel lin­eage, and nor does any other.

The takeaway is that the Abu Dhabi policy is the most use­ful doc­u­ment in the re­gion for a buyer any­where in the Gulf, in­clud­ing a buyer with no Abu Dhabi ex­pos­ure at all. It is spe­cific, it is read­able, and it de­scribes in op­er­at­ive lan­guage what the other five states de­scribe in as­pir­a­tion. Used as a tem­plate rather than as a com­pli­ance bur­den, it sup­plies the work plan that the Saudi and Qatari in­stru­ments do not make legible.

Saudi Arabia: the apparatus, and the number nobody publishes

Saudi Arabia has built the heaviest data governance apparatus in the Gulf and publishes the least about its results. The National Data Index has existed since November 2023 and no result from it has been published for any year.

The Saudi po­s­i­tion is a para­dox that a buyer must plan around rather than re­solve. The King­dom has the most elab­or­ate data gov­ernance ma­chinery in the re­gion: a ded­ic­ated au­thor­ity, a na­tional data man­age­ment of­fice, a bundled set of na­tional policies, man­dated gov­ernance bod­ies in­side every pub­lic en­tity, an an­nual audit with a fixed quarter and a pur­pose-built na­tional index to score the res­ult. What it does not have is a pub­lished out­come. No Na­tional Data Index res­ult has been made pub­lic for any year,

des­pite NDMO's stated com­mit­ment to pub­lish, and the op­er­at­ive text of the stand­ards against which en­tit­ies are audited was not reach­able. The con­sequence for a buyer is spe­cific: com­pli­ance in Saudi Ar­a­bia must be planned against an ap­par­atus whose re­quire­ments are known largely by re­state­ment and whose scor­ing is in­vis­ible.

The instruments, and their uncertain dates

Two in­stru­ments carry the re­gime. The Na­tional Data Gov­ernance In­terim Reg­u­la­tions, is­sued by SDAIA and NDMO, ap­pear as v1 dated 1 June 2020 on the of­fi­cial doc­u­ment, while sec­ond­ary ana­lysis gives ap­proval on 28 Septem­ber 2020 and pub­lic­a­tion on 20 Oc­to­ber 2020. Those dat­ings are un­re­con­ciled, and sep­ar­ate com­mence­ment dates for the com­pon­ent policies were not found. The Reg­u­la­tions bundle five policies: Data Clas­si­fic­a­tion, Data Shar­ing, Free­dom of In­form­a­tion, Open Data and the Per­sonal Data Pro­tec­tion In­terim Reg­u­la­tions. Along­side them sit the Data Man­age­ment and Per­sonal Data Pro­tec­tion Stand­ards, NDMO, v1.0 of June 2020, is­sued under Cab­inet Res­ol­u­tion 292 of 27/04/1441H. Whether a second ver­sion of those Stand­ards ex­ists could not be es­tab­lished. For an in­stru­ment bind­ing every pub­lic en­tity in the largest eco­nomy in the Gulf, that is a not­able gap in the pub­lic re­cord.

What is actually mandated

Three or­gan­isa­tional re­quire­ments are clear. Each pub­lic en­tity is to es­tab­lish a Data Man­age­ment Of­fice, con­sti­tute a Data Man­age­ment Com­mit­tee and ap­point a Chief Data Of­ficer. These are the struc­tural ob­lig­a­tions and they are not in doubt. Bey­ond them the pic­ture be­comes de­pend­ent on sec­ond­ary ma­ter­ial. Cata­logue, re­ten­tion and mas­ter data du­ties are lis­ted in vendor re­state­ments of the Stand­ards and were not veri­fied against the in­stru­ment it­self. The fre­quently quoted struc­ture of fif­teen do­mains, 77 con­trols and 191 spe­cific­a­tions ap­pears only in vendor com­pli­ance writ­ing and in no NDMO doc­u­ment that could be reached, and SDAIA's own Na­tional Data Index is built on four­teen do­mains rather than fif­teen. The scope is re­por­ted to ex­tend bey­ond gov­ern­ment to busi­ness part­ners hand­ling gov­ern­ment data, which, if cor­rect, is com­mer­cially sig­ni­fic­ant for every sup­plier to the Saudi pub­lic sec­tor. That too rests on a vendor source.

Be­cause the ver­batim op­er­at­ive clauses of the NDMO Stand­ards were not ob­tain­able, the ques­tion of which of cata­logue, lin­eage, chief data of­ficer and re­ten­tion are strictly man­dat­ory rests, for all but the chief data of­ficer, on vendor re­state­ment. A buyer re­ly­ing on those du­ties in a board paper should say so.

Data EngineeringVolge Quarterly13 •

The audit that does happen

The as­sess­ment re­gime is the most con­crete part of the Saudi ap­par­atus and the part a buyer can plan against. Every pub­lic en­tity con­ducts an an­nual com­pli­ance audit and sub­mits an En­tity Com­pli­ance Re­port to NDMO in the third quarter of each year. No fixed day within the quarter is pub­lished. NDMO may ad­di­tion­ally run ad-hoc audits, and it states that it re­ports at en­tity, sec­tor and whole-of-gov­ern­ment level. That is a re­cur­ring, bind­ing and dated ob­lig­a­tion in a re­gion where al­most noth­ing else is dated, and it is the single most use­ful fact in Saudi data com­pli­ance. An en­tity that treats Q3 as the plan­ning an­chor will be work­ing to the only pub­lished rhythm the re­gime has.

The index, and the silence

The Na­tional Data Index was un­veiled by SDAIA in Novem­ber 2023 and is un­usu­ally well spe­cified. It has three com­pon­ents: a ma­tur­ity ques­tion­naire of 42 ques­tions across four­teen do­mains scored on a six-level scale run­ning from Level 0, Ab­sence of Cap­ab­il­it­ies, to Level 5, Pi­on­eer; a com­pli­ance as­sess­ment scored per spe­cific­a­tion, at 100 per cent for full im­ple­ment­a­tion and zero for par­tial; and a set of op­er­a­tional in­dic­at­ors. The bin­ary treat­ment of par­tial com­pli­ance is worth not­ing, since it means an en­tity nine-tenths of the way through a con­trol scores the same as one that has not star­ted. Res­ults have not been pub­lished for any year. The only en­tity-level out­come that could be found any­where is the re­cog­ni­tion of the Min­istry of Health as an ad­vanced en­tity on 26 Decem­ber 2024, which car­ries no score.

An index that scores every public entity and publishes nothing creates assurance for the centre and none at all for the market.

The con­sequence is that no Saudi en­tity can bench­mark it­self, and no sup­plier can cal­ib­rate a pro­posal against the ac­tual state of the cli­ent's es­tate. It also leaves a va­cuum that other in­dices are drawn into filling. The Di­gital Gov­ern­ment Au­thor­ity's Di­gital Trans­form­a­tion Index re­por­ted an av­er­age of 87.14 per cent for 2024, up from 69.39 per cent in 2021, across an en­tity pop­u­la­tion re­por­ted as both 233 and 235. That fig­ure is fre­quently cited as evid­ence of Saudi data ma­tur­ity. It is not: it meas­ures di­gital gov­ern­ment, not data man­age­ment, and it should not be offered as a proxy.

The unsettled tiers

A fur­ther un­re­solved point af­fects any­one im­ple­ment­ing clas­si­fic­a­tion in the King­dom. Two of­fi­cial SDAIA doc­u­ments give dif­fer­ent clas­si­fic­a­tion schemes: one gives Top Secret, Secret, Con­fid­en­tial and Pub­lic, the other Top Secret, Secret, Re­stric­ted and Pub­lic. The third tier is there­fore un­settled on of­fi­cial sources. The prac­tical ad­vice is to im­ple­ment the struc­ture and label the third tier in ac­cord­ance with whichever doc­u­ment the con­tract­ing en­tity cites, and to re­cord the dis­crep­ancy in the design doc­u­ment­a­tion rather than re­solve it si­lently.

Privacy, where enforcement is real

The Per­sonal Data Pro­tec­tion Law, Royal De­cree M/19 of 2021 as amended by M/148 of 2023, came into force on 14 Septem­ber 2023 with a grace period that ended on 14 Septem­ber 2024. Fines run to SAR 5m and may be doubled. SDAIA an­nounced 48 en­force­ment de­cisions in the year to early 2026 (Latham and Watkins, 15 May 2026), though neither the de­cision texts nor the re­spond­ents are pub­lic. The con­trast with the data-man­age­ment side of the same au­thor­ity is the find­ing: the pri­vacy in­stru­ment is en­forced and the en­force­ment is un­pub­lished, while the data-man­age­ment in­stru­ment is audited an­nu­ally and the audit is un­pub­lished. On cap­ab­il­ity, the one Gulf fig­ure on spe­cial­ist sup­ply is also Saudi: SDAIA has trained more than 779,000 cit­izens in­clud­ing 9,775 spe­cial­ists and 260 sci­ent­ists and ex­perts, per the 2024 Vis­ion 2030 re­port (SPA N2311819, 5 May 2025), against an an­nounced NSDAI tar­get of ap­prox­im­ately 20,000 data and ar­ti­fi­cial in­tel­li­gence spe­cial­ists by 2030.

The takeaway is that a buyer in Saudi Ar­a­bia should build to the Q3 audit, doc­u­ment every ob­lig­a­tion that rests on re­state­ment as rest­ing on re­state­ment, and res­ist the tempta­tion to im­port the Di­gital Trans­form­a­tion Index av­er­age as a meas­ure of data read­i­ness. The ap­par­atus is real. The score is not avail­able, to any­one.

Oman and Kuwait: the two regimes a buyer can read

Oman and Kuwait have produced the Gulf's two most usable data instruments. Oman tiers its thirteen policies across three years by priority, and Kuwait's classification policy is the only one in the region that binds private business directly.

The two smal­lest data re­gimes in the Gulf are the two a buyer can ac­tu­ally work from. Oman's

Data EngineeringVolge Quarterly14 •

Reg­u­lat­ory Frame­work for Na­tional Data Gov­ernance and Man­age­ment is the best-doc­u­mented GCC re­gime on read­able primary sources: three pub­lished doc­u­ments, thir­teen numbered policies, named roles and an ex­pli­cit three-year se­quen­cing of ob­lig­a­tions by pri­or­ity. Kuwait's CITRA Data Clas­si­fic­a­tion Policy v2.3 is shorter but more far-reach­ing in one re­spect that no other Gulf in­stru­ment matches: it binds private sec­tor en­tit­ies dir­ectly, and it re­quires them to build and sur­render a data cata­logue. Both re­gimes, how­ever, share the re­gional de­fect. Oman has not pub­lished the cal­en­dar-year map­ping of its three pri­or­ity tiers, and Kuwait has not pub­lished the ex­e­cu­tion date from which its two dead­lines run. The ob­lig­a­tions are legible; the dates are not.

Oman: thirteen policies, three years

The Omani frame­work was is­sued by MTCIT under Min­is­terial Cir­cu­lar 113/2025, with the policies doc­u­ment dated 1 Janu­ary 2025. It com­prises three doc­u­ments: the Policies them­selves, a Guideline for Es­tab­lish­ing a Data Gov­ernance and Man­age­ment Of­fice, and a Guide for As­sess­ing Com­pli­ance. That triad is un­usual in the Gulf, be­cause it sup­plies not only the du­ties but the or­gan­isa­tional design to dis­charge them and the method by which com­pli­ance will be judged. Thir­teen policies are covered, in­clud­ing Data Cata­logue and metadata, Data Clas­si­fic­a­tion, Data Qual­ity, Ref­er­ence and Mas­ter Data, and Open Data. One of the thir­teen has no coun­ter­part else­where in the re­gion: Data Mon­et­isa­tion. Its pres­ence sig­nals that Oman is treat­ing gov­ern­ment data as an asset with po­ten­tial rev­enue at­tached rather than solely as a li­ab­il­ity to be con­trolled.

Or­gan­isa­tion­ally, a gov­ern­ment unit must es­tab­lish a Data Gov­ernance and Man­age­ment Of­fice in­side its di­gital trans­form­a­tion di­vi­sion, and must fill four named roles: busi­ness data spe­cial­ist, data tech­ni­cian, data ar­chi­tect and per­sonal data pro­tec­tion of­ficer. The frame­work per­mits these to be staffed from ex­ist­ing posts, which is a real­istic con­ces­sion and should be read as such: it lowers the entry cost con­sid­er­ably, and it also means that a unit can re­port the of­fice as es­tab­lished without hav­ing added a single head. Scope cov­ers all gov­ern­ment units, with the private sec­tor reached in­dir­ectly through sec­tor reg­u­lat­ors. A private op­er­ator in a reg­u­lated Omani sec­tor should there­fore ex­pect the frame­work to ar­rive through its reg­u­lator rather than through the min­istry.

As­sess­ment is an an­nual peri­odic ex­er­cise con­duc­ted by the min­istry's com­pli­ance team under the pub­lished Com­pli­ance As­sess­ment Guide. The se­quen­cing is the fea­ture most worth copy­ing. The pro­vi­sions of the thir­teen policies are split across three years by pri­or­ity: Pri­or­ity 1 pro­vi­sions in year one, Pri­or­ity 2 in year two, Pri­or­ity 3 in year three. That is a genu­inely pro­por­tion­ate design and the only staged im­ple­ment­a­tion path pub­lished in the Gulf. The de­fect is that the cal­en­dar-year map­ping has not been pub­lished, so a buyer can­not es­tab­lish from the in­stru­ment which cal­en­dar year con­sti­tutes year one. Clas­si­fic­a­tion runs on four levels bor­rowed from the Doc­u­ment and Archives Law, Top Secret, Secret, Re­stric­ted and Con­fid­en­tial, with Un­clas­si­fied along­side. The data owner clas­si­fies, and the re­cip­i­ent may not alter the clas­si­fic­a­tion. No pen­alty is pub­lished.

Kuwait: the only instrument that reaches a private company

The CITRA Data Clas­si­fic­a­tion Policy v2.3 car­ries neither an issue date nor a com­mence­ment date on its face, and that omis­sion is the cent­ral prob­lem with an oth­er­wise clear doc­u­ment. Its scope is what dis­tin­guishes it: it binds pub­lic and private sec­tor en­tit­ies, mak­ing it the only GCC data-clas­si­fic­a­tion in­stru­ment that reaches private busi­ness dir­ectly. Four tiers apply: Pub­lic, Private In­sens­it­ive, Private Sens­it­ive and Highly Sens­it­ive. Tiers 3 and 4 must be en­cryp­ted in transit. An en­tity must form a data clas­si­fic­a­tion team headed by senior man­age­ment and in­clud­ing the Dir­ect­ors of In­form­a­tion Se­cur­ity and In­form­a­tion Tech­no­logy, build a clas­si­fic­a­tion road map and con­duct peri­odic in­ternal re­view.

The ob­lig­a­tion with the longest tail is clause 3.1.3, under which the data owner must cre­ate and main­tain a data cata­logue in­clud­ing metadata in a uni­fied format, up­dated peri­od­ic­ally. Re­port­ing then makes the cata­logue an ex­ternal de­liv­er­able rather than an in­ternal re­cord. The en­tity focal point sub­mits an im­ple­ment­a­tion re­port to CAIT every quarter, and under clause 3.4.2.3 CITRA draws peri­odic com­pli­ance re­ports which must con­tain the full data cata­logue with metadata, the ap­proved clas­si­fic­a­tion tiers with their cri­teria, stor­age loc­a­tions by tier and pro­gress against the road map. A buyer should read that as a re­quire­ment to keep the cata­logue in a state fit to be handed over four times a year, which is a sig­ni­fic­antly dif­fer­ent en­gin­eer­ing prob­lem from keep­ing one for in­ternal use.

Two dead­lines exist in the policy and neither can be cal­cu­lated. Clause 3.4.2.4 re­quires an im­ple­ment­a­tion

Data EngineeringVolge Quarterly15 •
Kuwait has written the Gulf's only data-cataloguing duty that reaches a private company, and has not published the date from which it runs.

plan within three months of the policy's ex­e­cu­tion date, and re­quires en­tit­ies to be re­con­ciled within two years. The ex­e­cu­tion date is not pub­lished. Both dead­lines are there­fore un­place­able on a cal­en­dar, while the quarterly re­port­ing duty runs re­gard­less. The prudent read­ing for a private en­tity in scope is that the two-year re­con­cili­ation win­dow has in all like­li­hood already begun, and that there is no pub­lished basis on which to argue oth­er­wise.

One sec­ond­ary source states that De­cision 26 of 2024 re­pealed the Data Clas­si­fic­a­tion Policy. CITRA con­tin­ues to pub­lish it. The point is un­re­solved, and the policy is treated here as live be­cause the reg­u­lator still pub­lishes it.

What the two together demonstrate

Read side by side, Oman and Kuwait make the case that legib­il­ity and reg­u­lat­ory weight are in­de­pend­ent vari­ables. Neither state has the ap­par­atus of Saudi Ar­a­bia or the spe­cificity of Abu Dhabi, and both have pro­duced in­stru­ments from which a com­pet­ent team could write a work plan in an af­ter­noon. They also demon­strate the lim­its of that achieve­ment. Oman has pub­lished a three-year pri­or­ity se­quence without say­ing when year one began. Kuwait has pub­lished two dead­lines without pub­lish­ing the date they run from. Neither pub­lishes a pen­alty.

The takeaway is two­fold. For a buyer in either state, the un­dated char­ac­ter of the ob­lig­a­tions ar­gues for treat­ing them as live now rather than wait­ing for a date to be an­nounced, be­cause an un­dated man­dat­ory duty car­ries no grace period. And for a buyer else­where in the Gulf, the Omani Com­pli­ance As­sess­ment Guide is the most use­ful free arte­fact in the re­gion for struc­tur­ing a self-as­sess­ment, whatever the jur­is­dic­tion of the en­tity using it.

The ninety-five per cent that is not what you think

The MIT study does not find that 95 per cent of artificial intelligence projects fail, and the Gartner figure does not say that 60 per cent of them will be abandoned. Both carry qualifiers that are routinely dropped, and a buyer who repeats either without them will lose the argument.

Two num­bers now carry most of the weight in Gulf board­room ar­gu­ments about data found­a­tions, and both are being quoted in­cor­rectly. The first is the

MIT fig­ure, usu­ally rendered as the find­ing that 95 per cent of ar­ti­fi­cial in­tel­li­gence pro­jects fail. The study did not meas­ure pro­ject fail­ure and did not find that. The second is the Gart­ner fig­ure, usu­ally rendered as a pre­dic­tion that 60 per cent of ar­ti­fi­cial in­tel­li­gence pro­jects will be aban­doned. The pre­dic­tion car­ries a qual­i­fier, un­sup­por­ted by AI-ready data, and without it the fig­ure is simply wrong. The value of this chapter is de­fens­ive. A buyer who brings either num­ber to a board in its pop­u­lar form will be cor­rec­ted by the first per­son in the room who has read the source, and the cor­rec­tion will cost more cred­ib­il­ity than the num­ber was ever worth.

Figure as usually quotedWhat the study actually measuredSampleThe correctly stated finding
95 per cent of AI projects fail (MIT)Whether deployed generative artificial intelligence produced sustained productivity or profit-and-loss impact noticed by users or executives at around six months153 survey responses, 52 executive interviews and more than 300 publicly disclosed implementationsOnly about 5 per cent of custom generative artificial intelligence tools reach production with measurable impact
60 per cent of AI projects will be abandoned (Gartner)Analyst prediction, with a stated condition on the projects in scope1,203 data management leaders, surveyed July 2024Through 2026, organisations will abandon 60 per cent of AI projects unsupported by AI-ready data
63 per cent lack the data practices for AI (Gartner)Whether organisations have the right data management practices for artificial intelligence1,203 data management leaders, surveyed July 202463 per cent either do not have, or are unsure whether they have, the right data management practices for artificial intelligence
More than 80 per cent of AI projects fail (RAND)Root causes of failure, drawn from practitioner interviews65 data scientists and engineers, each with five or more years of experienceAn outside estimate that RAND cites, not a rate RAND measured
Two-thirds say data is the obstacle (McKinsey)Identification of the primary obstacle to enabling artificial intelligenceSample reported variously in secondary coverage; the primary document was not readMore than two-thirds of high-performing companies identify data as the primary obstacle to enabling artificial intelligence

The MIT study, properly described

The primary source is The GenAI Di­vide: State of AI in Busi­ness 2025, pub­lished by MIT Pro­ject NANDA in July 2025, au­thored by Chal­lap­ally, Pease, Ras­kar and Chari, run­ning to 26 pages. The doc­u­ment de­scribes it­self as pre­lim­in­ary find­ings and it is not peer re­viewed. Its evid­ence base com­bines 153 sur­vey

Data EngineeringVolge Quarterly16 •

re­sponses, 52 ex­ec­ut­ive in­ter­views and more than 300 pub­licly dis­closed im­ple­ment­a­tions. What it set out to es­tab­lish was whether gen­er­at­ive ar­ti­fi­cial in­tel­li­gence, once de­ployed, pro­duced sus­tained pro­ductiv­ity or profit-and-loss im­pact that users or ex­ec­ut­ives ac­tu­ally no­ticed at around the six-month mark. That is a ques­tion about real­ised be­ne­fit, not about de­liv­ery fail­ure. A pro­ject that was de­livered on time and on budget, and that nobody could show had moved a num­ber six months later, falls on the wrong side of the MIT line while being a suc­cess­ful pro­ject by any con­ven­tional meas­ure.

Two fur­ther lim­its mat­ter for a Gulf reader spe­cific­ally. The study con­tains no geo­graphic break­down and no pub­lic-sec­tor ana­lysis, so it says noth­ing whatever about the Gulf, and noth­ing about gov­ern­ment, which is the largest buyer of data and ana­lyt­ics cap­ab­il­ity in every one of the six states. Crit­ics have also noted that the 95 per cent de­rives from a sub­set of the evid­ence base rather than from the whole of it. The de­fens­ible ver­sion of the find­ing is the one the study it­self frames: only about 5 per cent of cus­tom gen­er­at­ive ar­ti­fi­cial in­tel­li­gence tools reach pro­duc­tion with meas­ur­able im­pact. That is a ser­i­ous claim and it sup­ports the data found­a­tions ar­gu­ment per­fectly well. It simply is not the claim being re­peated.

The honest version of the number is less dramatic and considerably harder to answer.

The Gartner figure, and the clause that is always dropped

The Gart­ner source is a press re­lease of 26 Feb­ru­ary 2025 titled Lack of AI-Ready Data Puts AI Pro­jects at Risk, quot­ing Rox­ane Ed­jlali. The pre­dic­tion is that through 2026, or­gan­isa­tions will aban­don 60 per cent of ar­ti­fi­cial in­tel­li­gence pro­jects un­sup­por­ted by AI-ready data. The final five words are the whole of it. The pre­dic­tion is con­di­tional on a pop­u­la­tion of pro­jects that lack AI-ready data, and it says noth­ing about the aban­don­ment rate of pro­jects that have it. Drop­ping the qual­i­fier con­verts a con­di­tional state­ment about a sub­set into a uni­ver­sal claim about the whole, which is both wrong and, awk­wardly for any­one mak­ing the found­a­tions ar­gu­ment, self-de­feat­ing: the con­di­tional ver­sion is the one that ac­tu­ally at­trib­utes fail­ure to data.

The ori­ginal re­lease car­ry­ing the 30 per cent pre­dic­tion was not re-veri­fied. It should be cited as a Gart­ner pre­dic­tion without a date at­tached un­less the buyer loc­ates the re­lease.

The com­pan­ion stat­istic from the same sur­vey of 1,203 data man­age­ment lead­ers, con­duc­ted in July 2024, is cleaner and more use­ful. Sixty-three per cent of or­gan­isa­tions either do not have, or are un­sure whether they have, the right data man­age­ment prac­tices for ar­ti­fi­cial in­tel­li­gence. The two clauses should be kept to­gether. The un­cer­tainty is it­self the find­ing, be­cause an or­gan­isa­tion that can­not say whether its data prac­tices are ad­equate has already answered the ques­tion. Gart­ner sep­ar­ately pre­dicts that through 2026, 30 per cent of gen­er­at­ive ar­ti­fi­cial in­tel­li­gence pro­jects will be aban­doned after proof of concept, owing to poor data qual­ity, in­ad­equate risk con­trols, es­cal­at­ing costs or un­clear busi­ness value. That pre­dic­tion has four stated causes and data qual­ity is one of four.

RAND, McKinsey, and the discipline of attribution

RAND's The Root Causes of Fail­ure for Ar­ti­fi­cial In­tel­li­gence Pro­jects and How They Can Suc­ceed, pub­lished 13 Au­gust 2024 by Ryseff, De Bruhl and New­berry, rests on in­ter­views with 65 data sci­ent­ists and en­gin­eers each car­ry­ing five or more years of ex­per­i­ence. It iden­ti­fies five root causes, of which the second is that the or­gan­isa­tion lacks the data needed to train an ef­fect­ive model. That is a well-sourced qual­it­at­ive find­ing and a buyer can use it. What a buyer can­not use is the fig­ure most often at­tached to the paper, that more than 80 per cent of ar­ti­fi­cial in­tel­li­gence pro­jects fail. RAND cites that as an out­side es­tim­ate. It did not meas­ure it, and it should not be at­trib­uted to RAND. McKinsey's po­s­i­tion that more than two-thirds of high-per­form­ing com­pan­ies identify data as the primary obstacle to en­abling ar­ti­fi­cial in­tel­li­gence is quot­able as stated. The ad­op­tion fig­ures cir­cu­lat­ing from its State of AI work, 88 per cent ad­op­tion against 7 per cent fully scaled, come from sec­ond­ary cov­er­age with the sample re­por­ted vari­ously as 1,933, 1,993 and about 1,500. The primary doc­u­ment was not read, so no sample fig­ure should be cited with them.

The takeaway is a rule rather than a num­ber. Any fig­ure brought into a Gulf data ar­gu­ment should be stated with its qual­i­fier, its sample and its date, and should be dropped en­tirely if any of the three can­not be pro­duced. The evid­ence that data found­a­tions de­term­ine ar­ti­fi­cial in­tel­li­gence out­comes is strong enough to sur­vive that dis­cip­line. The head­line ver­sions are not, and they are the ones a scep­tical fin­ance dir­ector will check first.

Data EngineeringVolge Quarterly17 •

Where the Gulf actually stands

One Gulf state publishes an aggregate data maturity score and none publishes entity-level scores. On the one substantial regional survey, artificial intelligence spending is rising at more than a hundred per cent a year while fewer than one organisation in seven has replaced the systems its data sits in.

The hon­est an­swer to where the Gulf stands on data found­a­tions is that al­most nobody will say. No Gulf state pub­lishes en­tity-level data ma­tur­ity scores. The United Arab Emir­ates is the only one of the six to pub­lish even an ag­greg­ate, and Saudi Ar­a­bia has op­er­ated a pur­pose-built na­tional index since Novem­ber 2023 without re­leas­ing a res­ult from it. What can be as­sembled in­stead is a com­pos­ite: one pub­lished ag­greg­ate, one set of open data­set counts, one vendor-com­mis­sioned re­gional sur­vey and a tal­ent pic­ture with a single source­able fig­ure in it. Taken to­gether they de­scribe a re­gion spend­ing very fast on ar­ti­fi­cial in­tel­li­gence and mov­ing slowly on the found­a­tions be­neath it, which is the gap this edi­tion is named for.

The one published aggregate

The UAE Data Ma­tur­ity Index is op­er­ated by the Fed­eral Com­pet­it­ive­ness and Stat­ist­ics Centre and was presen­ted to the United Na­tions Stat­ist­ical Com­mis­sion Net­work on Eco­nomic Stat­ist­ics on 4 June 2024. It re­por­ted an av­er­age data ma­tur­ity of fed­eral en­tit­ies of 80 per cent. The meth­od­o­logy is pub­lished in full, which is rarer and more valu­able than the score. Four bands apply: Basic from 0 to 49, De­vel­op­ing from 50 to 69, Ma­ture from 70 to 95 and Pi­on­eer from 96 to 100. Three weighted pil­lars pro­duce the score: Data Reg­u­lat­ory and Op­er­a­tional Frame­work at 10 per cent, Data Qual­ity Stand­ards at 40 per cent and Data Avail­ab­il­ity and Ex­change at 50 per cent. The index draws on 233 mil­lion data re­cords re­ceived monthly from fed­eral en­tit­ies. En­tity ranks from 1 to 15 are shown, but the names and in­di­vidual scores are not pub­lic.

The weight­ing re­pays at­ten­tion, be­cause it tells a buyer what the fed­eral gov­ern­ment con­siders data ma­tur­ity to con­sist of. Half the score sits in avail­ab­il­ity and ex­change, two-fifths in qual­ity stand­ards and only a tenth in the reg­u­lat­ory and op­er­a­tional frame­work. An en­tity that has writ­ten policies and ap­poin­ted of­ficers, and has not made its data avail­able or demon­strably ac­cur­ate, scores badly on this index by design. That is a use­ful cor­rect­ive to a mar­ket that often meas­ures gov­ernance doc­u­ments in­stead.

What is not published

Against that single dis­clos­ure sits an ex­tens­ive si­lence. No Saudi Na­tional Data Index res­ult has been found for any year; the only en­tity-level out­come loc­ated any­where was the re­cog­ni­tion of the Min­istry of Health as an ad­vanced en­tity on 26 Decem­ber 2024, without a score. The Di­gital Dubai Data Ma­tur­ity Frame­work was launched at GITEX 2022 with four pil­lars, Data Gov­ernance, Data Shar­ing, Data Ana­lysis and De­cision-Mak­ing, and Data Qual­ity, and its scores are not pub­lic; a data qual­ity ini­ti­at­ive began with el­even gov­ern­ment en­tit­ies in July 2024. Qatar Data Trust Index scores are not pub­lic. No count of chief data of­ficers ap­poin­ted ex­ists for any GCC state. Dubai ap­proved the ap­point­ment of 22 Chief Ar­ti­fi­cial In­tel­li­gence Of­ficers on 9 June 2024, and those are ar­ti­fi­cial in­tel­li­gence of­ficers, not chief data of­ficers: the two should not be con­flated, and the ap­point­ment of the former is fre­quently offered as evid­ence of the lat­ter.

Open data, as a visible proxy

StateOpen datasets, as at 6 May 2025Later figure where published
Saudi Arabia10,916Not published
United Arab Emirates (Bayanat)3,146Not published
Bahrain448515 datasets from 42 publishers, currently shown on data.gov.bh
Qatar159More than 1,100, on relaunch, 27 June 2025
Oman60Not published
KuwaitNo centralised portalNot published
Gulf total14,729Superseded for two states

Counts taken from the na­tional portals on a single date, 6 May 2025, and re­por­ted by ORF Middle East on 20 May 2025. The com­par­ison is already su­per­seded for Qatar and Bahrain, as shown. Data­set counts meas­ure pub­lic­a­tion, not qual­ity or ma­tur­ity.

The spread is the point. Open data is an im­per­fect proxy for ma­tur­ity, since a count says noth­ing about whether the data­sets are cur­rent, doc­u­mented or used, but it is one of the very few meas­ures in the re­gion that is both pub­lic and com­par­able, and the dis­tri­bu­tion broadly tracks the other in­dic­at­ors. On the United Na­tions E-Gov­ern­ment Sur­vey 2024, pub­lished in the UN DESA Tech­nical Ap­pendix of 30 Oc­to­ber 2024, Saudi Ar­a­bia ranked 6th with an index score of 0.9602, the UAE 11th at 0.9533, Bahrain 18th at 0.9196, Oman 41st at 0.8576, Qatar 53rd at 0.8244 and Kuwait 66th at 0.7812. On the Ox­ford In­sights Gov­ern­ment AI Read­i­ness Index 2025, Saudi Ar­a­bia placed 15th glob­ally at 69.79, the UAE 23rd at 66.87 and Qatar 54th; the

Data EngineeringVolge Quarterly18 •

data-di­men­sion sub-scores are not pub­lic, so neither index can be de­com­posed to isol­ate data cap­ab­il­ity.

Spend against foundations

The most dir­ectly rel­ev­ant Gulf evid­ence comes from the ServiceNow En­ter­prise AI Ma­tur­ity Index 2026, built on ThoughtLab field­work cov­er­ing 4,500 ex­ec­ut­ives across 19 coun­tries, in­clud­ing 100 in Saudi Ar­a­bia and 100 in the United Arab Emir­ates, and re­por­ted by Middle East AI News on 18 Au­gust 2026. Sev­enty-seven per cent of UAE ex­ec­ut­ives and 67 per cent of Saudi ex­ec­ut­ives named in­ad­equate data ac­cur­acy, ac­cess and man­age­ment as a major bar­rier to scal­ing ar­ti­fi­cial in­tel­li­gence. Ma­tur­ity scores were 48 out of 100 for the UAE and 50 out of 100 for Saudi Ar­a­bia. Ar­ti­fi­cial in­tel­li­gence spend­ing rose 105 per cent year on year in the UAE and 124 per cent in Saudi Ar­a­bia. And only 14 per cent of UAE or­gan­isa­tions and 13 per cent of Saudi or­gan­isa­tions had re­placed leg­acy sys­tems with plat­forms cap­able of in­teg­rat­ing ar­ti­fi­cial in­tel­li­gence.

Spending rose by more than a hundred per cent in both markets while fewer than one organisation in seven had replaced the systems the data sits in.

The Gulf samples are 100 ex­ec­ut­ives per coun­try and the study is vendor-com­mis­sioned. The dir­ec­tion is cred­ible and the pre­ci­sion of the per­cent­ages should not be re­lied upon. A Cloudera sur­vey of Saudi data read­i­ness, with its sample size not pub­lished, found 62 per cent of Saudi in­form­a­tion tech­no­logy lead­ers cit­ing data ac­cess re­stric­tions as a major bar­rier and 32 per cent re­port­ing com­plete vis­ib­il­ity into their data.

The talent position

On people the pub­lic re­cord is thin­ner still. The only Gulf fig­ure on spe­cial­ist cap­ab­il­ity is Saudi: SDAIA has trained more than 779,000 cit­izens, in­clud­ing 9,775 spe­cial­ists and 260 sci­ent­ists and ex­perts, per the 2024 Vis­ion 2030 re­port (SPA N2311819, 5 May 2025), against an an­nounced NSDAI tar­get of ap­prox­im­ately 20,000 data and ar­ti­fi­cial in­tel­li­gence spe­cial­ists by 2030. Later counts of 11,000, 14,000 and 14,495 cir­cu­late widely and trace to no dated primary re­lease, and are not used here. No na­tional la­bour of­fice in Saudi Ar­a­bia or the UAE pub­lishes va­cancy counts at oc­cu­pa­tion level, so the num­ber of open data en­gin­eer­ing roles in the Gulf is not pub­lic and can­not be es­tim­ated from of­fi­cial sources. GulfTalent re­ports ar­ti­fi­cial in­tel­li­gence skills ap­pear­ing in about 3.4 per cent of pro­fes­sional va­can­cies across the UAE, Saudi Ar­a­bia and Qatar in the first half of 2026, up from 1.2 per cent in 2022, drawn from 118,000 dir­ect-em­ployer va­can­cies

on its own job board; that meas­ures ar­ti­fi­cial in­tel­li­gence skills, not data en­gin­eer­ing. Korn Ferry's Q1 2026 sur­vey of 105 chief human re­sources of­ficers across the six states found 93 per cent ex­plor­ing, pi­lot­ing or de­ploy­ing ar­ti­fi­cial in­tel­li­gence and 1 per cent con­sid­er­ing them­selves fully ready to scale, a find­ing that is human-re­sources-scoped rather than data en­gin­eer­ing.

The takeaway is that a Gulf buyer can­not bench­mark against peers, be­cause peer scores are not pub­lished any­where in the re­gion. The prac­tical sub­sti­tute is to self-score against the UAE index, whose bands and weights are pub­lic, and to treat the 80 per cent fed­eral av­er­age as the only pub­lished ref­er­ence point in the Gulf. It is a UAE fed­eral fig­ure and should be used as such, not as a re­gional norm.

What it costs, and the Gulf premium nobody quotes

A Snowflake Standard credit costs 35 per cent more in Dammam than in Northern Virginia and 30 per cent more in Azure UAE North. On the published evidence the premium is a Snowflake characteristic rather than a regional one, because most platform vendors do not publish Gulf rates at all.

There is a Gulf cloud data premium, it is doc­u­mented in one vendor's pub­lished rate card, and it is lar­ger than most budgets as­sume. On the Snow­flake Ser­vice Con­sump­tion Table ef­fect­ive 9 Novem­ber 2026, a Stand­ard on-de­mand credit costs 35 per cent more in GCP Middle East Cent­ral 2 at Dammam than in AWS US East at North­ern Vir­ginia, and 30 per cent more in Azure UAE North. Whether that premium gen­er­al­ises can­not be es­tab­lished, and the reason is the second find­ing of this chapter: four of the five vendors ex­amined render their price tables in cli­ent-side JavaScript, so sev­eral Middle East rates are not pub­lished in re­triev­able form at all. Mi­crosoft Fab­ric pub­lishes no ca­pa­city price in any re­gion, BigQuery pub­lishes no Gulf rate, and Amazon Web Ser­vices pub­lishes no Middle East rate for Red­shift, Glue, EMR or Lake Form­a­tion.

Two fea­tures of the table de­serve more at­ten­tion than the head­line premium. The premium com­pounds up the edi­tion lad­der: a buyer on Busi­ness Crit­ical in Dammam pays USD 6.50 against USD 4.00 in North­ern Vir­ginia, so the host­ing de­cision and the edi­tion de­cision mul­tiply against each other. And Dammam stor­age is flat at USD 30.00 per tera­byte per month across all seven ca­pa­city tiers, where the usual ex­pect­a­tion is that volume at­tracts a lower unit rate. A busi­ness case that as­sumed stor­age eco­nom­ies of

Data EngineeringVolge Quarterly19 •

scale at the point of com­mit­ment will not find them in that re­gion.

The one published premium

RegionStandardEnterpriseBusiness CriticalVPSStorage per TB per month
AWS US East, Northern VirginiaUSD 2.00USD 3.00USD 4.00USD 6.00USD 23.00
Azure UAE NorthUSD 2.70USD 4.00USD 5.40USD 8.10USD 25.40
GCP Middle East Central 2, DammamUSD 3.25USD 4.90USD 6.50USD 9.75USD 30.00

Snow­flake on-de­mand credit prices and stor­age rates from the Ser­vice Con­sump­tion Table ef­fect­ive 9 Novem­ber 2026, re­trieved 10 Novem­ber 2026. No rate shown was de­rived from any other rate. Dammam stor­age is flat at USD 30.00 across all seven ca­pa­city tiers with no tier dis­count. AWS me-cent­ral-1 in the UAE and me-south-1 in Bahrain are not lis­ted at all. Ca­pa­city credit prices are not pub­lic as rates.

Three of the five platform vendors publish no retrievable Gulf rate at all, which leaves a quotation as the only honest basis for a Gulf budget.

Databricks, where there is no uplift

Azure Databricks Premium, per DBU-hourUAE NorthQatar Central
Jobs ComputeUSD 0.30USD 0.30
All-purposeUSD 0.55USD 0.55
SQL Compute ProUSD 0.74USD 0.74
Serverless SQLUSD 0.95Not published
Delta Live Tables, CoreUSD 0.30USD 0.30
Delta Live Tables, ProUSD 0.38USD 0.38
Delta Live Tables, AdvancedUSD 0.54USD 0.54

Dat­ab­ricks presents the op­pos­ite pic­ture, and it is the evid­ence against treat­ing the Snow­flake premium as a re­gional fact. The rates are identical in both Gulf re­gions and carry no re­gional up­lift at all. Qatar Cent­ral pub­lishes no server­less rate, which is a cap­ab­il­ity dif­fer­ence rather than a pri­cing one. The con­clu­sion a buyer should draw is that the Gulf premium is a prop­erty of a par­tic­u­lar vendor's re­gional rate card rather than of Gulf in­fra­struc­ture, and that it must there­fore be tested vendor by vendor rather than as­sumed.

Rates taken from the Mi­crosoft re­tail prices ap­plic­a­tion pro­gram­ming in­ter­face that bills the ser­vice, re­trieved 10 Novem­ber 2026. Dat­ab­ricks is not avail­able in any AWS Middle East re­gion, and GCP me-central2 at Dammam is sup­por­ted, per doc­u­ment­a­tion up­dated 21 Oc­to­ber 2026. Dat­ab­ricks own-brand DBU list prices for AWS and GCP, Dammam in­cluded, are not pub­lic by static fetch.

Google BigQuery pub­lishes a com­plete rate card for us-central1: on-de­mand at USD 6.25 per te­bi­b­yte scanned with the first te­bi­b­yte each month free,

Stand­ard edi­tion at USD 0.04 per slot-hour, En­ter­prise at USD 0.06 and En­ter­prise Plus at USD 0.10, with one-year and three-year com­mit­ment dis­counts, and act­ive lo­gical stor­age at USD 0.000031507 per gib­i­b­yte-hour. The Gulf re­gions me-central1 at Doha, me-central2 at Dammam and me-west1 all ap­pear in the re­gion se­lect­ors, so rates exist, but the served page re­turns us-central1 only. Gulf BigQuery rates are there­fore not pub­lic by static fetch. Two fig­ures are pub­lished and re­gion-blind, and they are the ones a Gulf ar­chi­tect should price care­fully: cross-re­gion trans­fer into the Middle East at USD 0.11 per gib­i­b­yte from North Amer­ica, Europe or Asia, and USD 0.08 per gib­i­b­yte for Middle East to Middle East.

Fabric and BigQuery, where the price is absent

Mi­crosoft Fab­ric pub­lishes its full SKU lad­der from F2 to F8192 with ca­pa­city units at­tached, and every price cell renders as a dash. Pay-as-you-go and re­served F-SKU prices are not pub­lic by static fetch in any re­gion, not merely in the Gulf, while the re­gion se­lector lists UAE North, UAE Cent­ral and Qatar Cent­ral: ca­pa­city is avail­able in three Gulf re­gions at prices that can­not be read. One re­ser­va­tion line is re­triev­able for UAE North, where the meter Fab­ric Ca­pa­city CU shows a one-year re­tail price of 1146.0 and a three-year price of 3438.0, with the unit of meas­ure la­belled by the in­ter­face as 1 Hour. OneLake stor­age in UAE North is pub­lished per giga­byte per month at USD 0.0253 for Hot, USD 0.015 for Cool, USD 0.005 for Cold and USD 0.264 for Cache.

Amazon Web Services, priced in Virginia

AWS service, US East Northern VirginiaPublished rate
Glue ETL SparkUSD 0.44 per DPU-hour
Glue FlexUSD 0.29 per DPU-hour
Glue crawlersUSD 0.44 per DPU-hour
Glue Data Quality, recommendation and evaluationUSD 0.44 per DPU-hour, two-DPU minimum
Glue Data Catalog, metadata objectsFirst 1,000,000 free, then USD 1.00 per 100,000 per month
Glue Data Catalog, requestsFirst 1,000,000 per month free, then USD 1.00 per million
Lake Formation permissions, tag-based access control, cross-account sharingNo charge
Redshift ServerlessUSD 0.375 per RPU-hour
Redshift ra3.4xlargeUSD 3.26 per node-hour, from AWS's own worked example
Redshift managed storageUSD 0.024 per GB-month
Redshift SpectrumUSD 5.00 per TB scanned
Data EngineeringVolge Quarterly20 •

AWS pri­cing pages up­dated 5 Novem­ber 2026. The RA3 node tables render empty. Middle East rates in me-cent­ral-1 and me-south-1 for Red­shift, Glue, EMR and Lake Form­a­tion are not pub­lic by static fetch, and the only source­able AWS state­ment on the point is that pri­cing var­ies by Re­gion. The rates above are there­fore a Vir­ginia baseline and not a Gulf quo­ta­tion.

Contracts, announcements and the difference between them

Below the rate cards the po­s­i­tion wor­sens. No pub­lished re­gister of GCC data con­tract awards ex­ists, and not one award across the six states could be sourced with buyer, scope and value pub­lished to­gether by the buyer. Etimad is blocked to auto­mated ac­cess by CAPTCHA and Qatar's Mon­aqasat re­turns an empty body. The one Etimad ag­greg­ate avail­able is that con­tracts awar­ded through the plat­form ex­ceeded SAR 100bn in 2023, with no cat­egory break­down, and the plat­form ex­cludes Ara­mco, SABIC, STC and PIF port­fo­lio com­pan­ies. The only Gulf con­tracts car­ry­ing a named gov­ern­ment buyer and a pub­lished value are Tadawul dis­clos­ures by Al Moam­mar In­form­a­tion Sys­tems with SDAIA as buyer, set out below, and every one of them is in­form­a­tion tech­no­logy in­fra­struc­ture sup­ply rather than data gov­ernance or ana­lyt­ics work.

Disclosed valueScope, SDAIA as buyerTerm or date
SAR 58.7m including VATSDAIA artificial intelligence network enhancement36 months
SAR 64.63mNational Information Center infrastructure supportNot stated
SAR 83.95mScope not stated in the sourced disclosureSigned 20 November
SAR 98.04mScope not stated in the sourced disclosureSigned 19 November
SAR 134.84mFramework agreement36 months, awarded 17 March 2026
SAR 228mExpansion of the Naqaa data centre, RiyadhNot stated

Sign­ing dates do not re­con­cile across re­ports, which is why two rows above carry a month without a year. One fur­ther con­tract is pub­lic without a value: Elm's sub­si­di­ary Thiqah holds an agree­ment with the Min­istry of Com­merce to di­git­ise min­istry ser­vices on a rev­enue-shar­ing basis, with no fixed total and a five-year term from 3 Septem­ber 2026.

Three cat­egor­ies should be kept sep­ar­ate and routinely are not. Spent: noth­ing, be­cause no Gulf gov­ern­ment pub­lishes real­ised spend on data or ana­lyt­ics pro­grammes. Com­mit­ted with a dis­closed value: only the Al Moam­mar con­tracts above. An­nounced only: Snow­flake's USD 300m in Saudi Ar­a­bia through 2030, stated on 3 Novem­ber 2026, Dat­ab­ricks at USD 300m over five years, Hua­wei Cloud at USD 400m, Alibaba Cloud at USD 500m and Abu Dhabi's AED 13bn di­gital strategy. The Abu Dhabi De­part­ment of Gov­ern­ment En­able­ment agree­ment with Mi­crosoft and Core42 on sov­er­eign cloud has no dis­closed value, and the AED 13bn fig­ure at­tached to it in

sec­ond­ary cov­er­age is the whole-of-gov­ern­ment Di­gital Strategy 2025 to 2027 plan rather than the con­tract. That is a com­mon and con­sequen­tial error. On ana­lyst fram­ing, Gart­ner put MENA in­form­a­tion tech­no­logy spend­ing at USD 169bn in 2026, up 8.9 per cent, with data centre sys­tems the fast­est-grow­ing seg­ment at 37.3 per cent to about USD 13bn and soft­ware up 13.9 per cent to USD 20.4bn (press re­lease, 4 Au­gust 2025); MENA there in­cludes Egypt, Mo­rocco and Al­geria. IDC put MEA in­form­a­tion tech­no­logy spend­ing at USD 155bn in 2025 with 5 per cent fore­cast growth in 2026. No Gart­ner or IDC data-and-ana­lyt­ics-spe­cific Middle East fig­ure is pub­lic.

Role, United Arab EmiratesAED per month, basic plus fixed allowances
Data Governance Director74,000 to 85,000
Data Architect48,000 to 59,000
Data Engineer, industry27,000 to 44,000
Data Engineer, start-ups25,000 to 35,000
Data Scientist44,000 to 55,000

Cooper Fitch UAE Salary Guide 2025, pub­lished Decem­ber 2024. These are re­cruiter es­tim­ates, not a stat­ist­ical sur­vey. The Robert Wal­ters Middle East Salary Sur­vey 2026 con­tains no Data En­gin­eer and no Data Ar­chi­tect line, and its Saudi tech­no­logy table is identical line for line to its UAE table under an AED head­ing, so no 2026 Saudi fig­ure should be taken from it. Hays, Mi­chael Page, Cooper Fitch KSA and Mer­cer are gated, and role-level fig­ures for Qatar, Oman, Bahrain and Kuwait are not pub­lic from any named guide.

Four widely quoted mar­ket fig­ures should not ap­pear in a buyer's busi­ness case at all. PwC's USD 320bn is a GDP-con­tri­bu­tion model from a press re­lease of 11 Feb­ru­ary 2018 cov­er­ing the Middle East rather than the GCC. Statista's MEA big data item is a re­pub­lished third-party fore­cast re­leased in March 2017 with pay­walled val­ues. Six com­mer­cial houses size the same 2025 Middle East data ana­lyt­ics mar­ket between USD 4bn and USD 17.75bn, a spread of 4.4 times. And Presight's AED 2.48bn second-quarter 2026 order in­take and AED 4.9bn back­log are neither con­tract val­ues nor awards.

The takeaway is that Gulf data plat­form budget­ing can­not be done from pub­lic list prices, be­cause for three of the five vendors there are none to read in the re­gion. What can be done is the arith­metic that the Snow­flake table per­mits: price the same work­load in a Gulf re­gion and in North­ern Vir­ginia, carry the dif­fer­ence as an ex­pli­cit sov­er­eignty cost, and put it in front of who­ever de­cided that the data must stay in the coun­try. That de­cision is usu­ally cor­rect. It is rarely cos­ted.

A maturity model a buyer can use without a licence

The strongest freely available data maturity model for a Gulf buyer is the UAE Federal Data Maturity Index, which publishes its bands and its pillar weights in full. The better-known alternatives are licensed, unreachable or retired.

A buyer want­ing to score its own data ma­tur­ity has six can­did­ate mod­els and should choose the UAE Fed­eral Data Ma­tur­ity Index. It is free, it is Gulf-nat­ive,

Data EngineeringVolge Quarterly21 •

and alone among the op­tions it pub­lishes both its scor­ing bands and the weights of its pil­lars, which means a self-as­sess­ment pro­duced against it can be re­pro­duced and chal­lenged. The al­tern­at­ives fail on avail­ab­il­ity rather than on qual­ity: the best-known in­ter­na­tional model, DCAM, is avail­able only to mem­ber firms of the body that owns it; the most widely cited name, CMMI's data man­age­ment model, ap­pears to have been re­tired; and Gart­ner's own model is not sourced in the open at all. The re­com­mend­a­tion that fol­lows is there­fore not that the UAE index is the most soph­ist­ic­ated in­stru­ment avail­able. It is that it is the most soph­ist­ic­ated in­stru­ment a Gulf buyer can ac­tu­ally ob­tain and apply.

ModelStructureAvailabilityRecommended use
UAE Federal Data Maturity IndexFour bands, Basic 0 to 49, Developing 50 to 69, Mature 70 to 95, Pioneer 96 to 100; three weighted pillars, Regulatory and Operational Framework 10 per cent, Data Quality Standards 40 per cent, Data Availability and Exchange 50 per centFree; bands and weights fully publishedThe scoring spine. Self-score against it at no cost
Saudi National Data Index maturity scaleSix levels, Level 0 Absence of Capabilities to Level 5 Pioneer; 42 questions across 14 domainsFree; methodology published, but the compliance component depends on NDMO specifications whose operative text is not reachableUseful as a question set for a Saudi entity. Not usable as a complete assessment
Oman Guide for Assessing ComplianceThirteen policies with provisions tiered Priority 1 to Priority 3 across three yearsFree; publishedThe control checklist and the staging plan, usable in any Gulf jurisdiction
DAMA DMBOK, 2nd edition 2017A body of knowledge. Its Data Management Maturity Assessment chapter sets out a six-level scale from Level 0 absence through Initial or Ad Hoc, Repeatable, Defined and Managed to Optimised, assessed on activity, tools, standards, and people and resourcesThe book is purchasable; the framework is not licensed for assessment useThe vocabulary and the domain taxonomy. Not a scoring model
ModelStructureAvailabilityRecommended use
EDM Council DCAMVersion 3 has 8 components, 34 capabilities and 101 sub-capabilities, each rated on a six-level maturity scale; v3.1 current as at April 2026Licensed. Available only to EDM Council member firms through the Knowledge Portal. Training and certification listed at USD 1,370 per seat for members, with a USD 950 upgrade course. Membership fees not publicAppropriate only where the buyer is already a member firm
CMMI Data Management MaturityReleased August 2014. Five levels and 25 process areas across six capability areas: data governance, data quality, data operations, data architecture, data development and data securityISACA is reported to have discontinued support in January 2022 and the model page removed. Treated as retired, not officially confirmedNot recommended
Qatar Data Trust IndexSeven dimensions and 50 sub-indicators, producing a Data Quality Score and a Data Compliance Score, with a stated action threshold at 85Methodology document not reachableBinding on Qatari government entities. Not usable as a self-assessment by anyone else

Why the UAE index wins

The case for the UAE index rests on three prop­er­ties. It is trans­par­ent: the four bands and the three pil­lar weights are pub­lished, so a score can be re­com­puted by any­one who dis­putes it. It is opin­ion­ated in a way that is use­ful: half the weight sits in Data Avail­ab­il­ity and Ex­change and two-fifths in Data Qual­ity Stand­ards, leav­ing only a tenth for the reg­u­lat­ory and op­er­a­tional frame­work, which pre­vents an or­gan­isa­tion scor­ing well on the strength of policy doc­u­ments alone. And it is Gulf-nat­ive, built by a Gulf stat­ist­ical au­thor­ity for Gulf gov­ern­ment en­tit­ies, which makes it easier to de­fend in­tern­ally than an Amer­ican or European frame­work. Its lim­it­a­tion is one it shares with every other in­stru­ment in the re­gion: en­tity names and in­di­vidual scores are not pub­lished, so the index yields a score and no peer group. A buyer can es­tab­lish that it sits in the De­vel­op­ing band. It can­not es­tab­lish whether that is bet­ter or worse than its neigh­bours, and the pub­lished fed­eral av­er­age of 80 per cent is the only ex­ternal ref­er­ence point avail­able any­where in the Gulf against which to read its own res­ult.

Data EngineeringVolge Quarterly22 •
A model whose scoring bands are published is worth more to a buyer than a better model behind a membership.

The licensed option, and its price

DCAM, the Data Man­age­ment Cap­ab­il­ity As­sess­ment Model of the EDM Coun­cil, is the most gran­u­lar in­stru­ment on the list. Ver­sion 3 sets out 8 com­pon­ents, 34 cap­ab­il­it­ies and 101 sub-cap­ab­il­it­ies, each rated on a six-level ma­tur­ity scale, and ver­sion 3.1 is cur­rent as at April 2026. That gran­u­lar­ity is genu­ine and it comes with a gate: DCAM is avail­able only to mem­ber firms through the Coun­cil's Know­ledge Portal. Train­ing and cer­ti­fic­a­tion are lis­ted at USD 1,370 per seat for mem­bers, with a USD 950 up­grade course, and mem­ber­ship fees are not pub­lished. For an in­sti­tu­tion that is already a mem­ber, typ­ic­ally a bank, DCAM is the right choice and this chapter's re­com­mend­a­tion does not apply. For every­body else the po­s­i­tion is that the as­sess­ment can­not be star­ted without a pro­cure­ment de­cision, and a Gulf buyer whose im­me­di­ate prob­lem is to es­tab­lish a baseline will reach the baseline faster with the UAE index.

DAMA, correctly positioned

DAMA DMBOK, second edi­tion 2017, is the most widely held ref­er­ence in the field and the most fre­quently mis­used. It is a body of know­ledge, not a scor­ing model. Its Data Man­age­ment Ma­tur­ity As­sess­ment chapter does set out a six-level scale, Level 0 for ab­sence of cap­ab­il­ity through Ini­tial or Ad Hoc, Re­peat­able, Defined and Man­aged to Op­tim­ised, as­sessed across activ­ity, tools, stand­ards, and people and re­sources. Sources dis­agree over whether the level names in­clude Meas­ured, which is it­self a reason not to present the scale as au­thor­it­at­ive. The book is pur­chas­able and the frame­work is not li­censed for as­sess­ment use. The right role for DMBOK in a Gulf pro­gramme is as the vocab­u­lary and the do­main tax­onomy: it sup­plies the lan­guage in which the Abu Dhabi, Omani and Saudi do­main lists can be re­con­ciled with one an­other, which is a genu­ine ser­vice and not a ma­tur­ity score.

The model to stop citing

The CMMI Data Man­age­ment Ma­tur­ity model, re­leased in Au­gust 2014 with five levels and 25 pro­cess areas across six cap­ab­il­ity areas, re­mains the most quoted name in Gulf tender doc­u­ments. ISACA is

re­por­ted to have dis­con­tin­ued sup­port for it in Janu­ary 2022, and the model page has been re­moved. That re­port is not of­fi­cially con­firmed, which is the proper caveat, but the prac­tical con­clu­sion does not change: a model whose owner ap­pears to have with­drawn it and whose doc­u­ment­a­tion is no longer pub­lished should not be writ­ten into a spe­cific­a­tion. A buyer who finds it named in a draft re­quire­ment should sub­sti­tute the UAE index and re­cord why.

The takeaway is a three-part con­struc­tion that costs noth­ing. Use the UAE Fed­eral Data Ma­tur­ity Index as the scor­ing spine, be­cause its bands and weights are pub­lic. Use Oman's Guide for As­sess­ing Com­pli­ance as the con­trol check­list and the sta­ging plan, be­cause it is the only pub­lished in­stru­ment in the Gulf that se­quences ob­lig­a­tions by pri­or­ity across three years. Use DAMA DMBOK for the vocab­u­lary. Then over­lay the bind­ing in­stru­ment of the buyer's own jur­is­dic­tion, which in Abu Dhabi, Oman and Kuwait is read­able and in Saudi Ar­a­bia and Qatar is not. Noth­ing in that stack re­quires a li­cence, and the whole of it can be as­sembled from doc­u­ments a buyer can down­load today.

The dated calendar, and what is not on it

Almost no GCC data instrument carries a published deadline. Across the six states the only dated obligations in the 2026 to 2028 window are recurring reporting cycles with no fixed day, and the hardest dated data obligation facing a Gulf group with Indian operations is Indian.

The most strik­ing find­ing in this edi­tion is an ab­sence. Across the six GCC states, al­most no data in­stru­ment car­ries a pub­lished dead­line. The ob­lig­a­tions that bind now are re­cur­ring re­port­ing cycles, and of those only one, the Saudi third-quarter audit, even names a quarter. Oman tiers its pro­vi­sions across three years without pub­lish­ing which cal­en­dar year is year one. Kuwait sets two dead­lines that run from an ex­e­cu­tion date it has not pub­lished. Abu Dhabi re­quires reg­u­lar re­port­ing and defines reg­u­lar nowhere. Qatar's re­view cycle is un­dated. There is no dated Saudi NDMO dead­line, no dated Qatar Na­tional Data Policy dead­line and no dated Bahraini ob­lig­a­tion of any kind. The hard­est dated data ob­lig­a­tion in the win­dow be­longs to a jur­is­dic­tion out­side the Gulf, and falls in May 2027 in India.

Data EngineeringVolge Quarterly23 •
WhenObligationWho is boundStatus
Q3 of 2026, 2027 and 2028NDMO compliance audit and Entity Compliance ReportAll Saudi public entitiesRecurring and binding. No fixed day published
AnnuallyMTCIT periodic compliance assessment under the Compliance Assessment GuideAll Omani government unitsRecurring and binding. Priority 1 provisions in year one, Priority 2 in year two, Priority 3 in year three. Calendar-year mapping not published
Every quarterImplementation report to CAIT, including the full data catalogue with metadataKuwaiti public and private sector entities in scopeRecurring and binding. No fixed day published
Frequency stated only as regularProgramme status report against Data Management Programme Plan milestonesAbu Dhabi government entities and their contractorsRecurring and binding. No interval published
Cycle not datedInternal review against the National Data Standards and Qatar Data Trust Index scoringQatari government entitiesRecurring. Cycle not published
During 2026 or 2027, no published datePDPL Executive Regulations expected from the Federal Authority for Artificial Intelligence and DataUAE controllers and processorsA forecast, not an obligation
Two years from the policy's execution dateEntities reconciled with the CITRA Data Classification PolicyKuwaiti public and private sector entities in scopeNot placeable, because the execution date is not published
About 13 November 2026DPDP Rules 2025, Rule 4, consent manager registrationConsent managers in IndiaDated and notified
About 13 May 2027DPDP Rules 2025, Rules 3, 5 to 16, 22 and 23, and the substantive sectionsIndian data fiduciaries, including Gulf groups with Indian operationsDated and notified. The hardest dated data obligation across the jurisdictions examined

What an undated obligation actually means

It would be com­fort­able to read the ab­sence of dead­lines as a sign that Gulf data ob­lig­a­tions are soft. The op­pos­ite is the bet­ter read­ing. A dead­line is a con­ces­sion. It tells the reg­u­lated party when the duty be­gins and, by im­plic­a­tion, that it does not bind be­fore. An in­stru­ment that is man­dat­ory on issue and sets no dead­line has been in force since the day it was is­sued, which means there is no grace period to rely on and no trans­itional de­fence to run. Oman's Cir­cu­lar 113/2025 has bound all gov­ern­ment units since Janu­ary 2025. Qatar's clas­si­fic­a­tion policy has bound gov­ern­ment and vital-sec­tor en­tit­ies since 7 May 2023 and its Na­tional Data Policy since Oc­to­ber 2024. Dubai's Res­ol­u­tion 2 has bound since 2017. Abu Dhabi's policy

is man­dat­ory on its own terms and its issue date is not even pub­lished. In each case the en­tity is already late if it is not com­pli­ant, and noth­ing in the in­stru­ment says how late.

An obligation with no deadline is not a soft obligation: it is one that has been in force since the day it was issued.

The regimes already standing

  • Qatar, 7 May 2023: Na­tional Data Clas­si­fic­a­tion Policy v3.0 launched, bind­ing gov­ern­ment and vital-sec­tor en­tit­ies.
  • Saudi Ar­a­bia, 14 Septem­ber 2024: the PDPL grace period ended, fol­low­ing entry into force on 14 Septem­ber 2023.
  • Qatar, Oc­to­ber 2024: Na­tional Data Policy dated, bind­ing all gov­ern­ment en­tit­ies as re­por­ted.
  • Oman, Janu­ary 2025: Min­is­terial Cir­cu­lar 113/2025 bind­ing, with the policies doc­u­ment dated 1 Janu­ary 2025.
  • Bahrain, 20 May 2025: Open Gov­ern­ment Data Policy v2.0, su­per­sed­ing v1.3 of 10 Janu­ary 2023. One copy of v2.0 is dated 30 July 2025.
  • Dubai, 2017: Res­ol­u­tion 2 Policies Doc­u­ment, fol­low­ing Law 26 of 2015.
  • DIFC, 1 Oc­to­ber 2020: Data Pro­tec­tion Law 5 of 2020 en­force­able, amended with ef­fect from 15 July 2025.
  • ADGM, 14 Feb­ru­ary 2022: Data Pro­tec­tion Reg­u­la­tions 2021 trans­ition com­pleted.

Two items in that list are pri­vacy in­stru­ments rather than data man­age­ment, and the dis­tinc­tion holds through­out this edi­tion: the DIFC and ADGM re­gimes gov­ern per­sonal data and im­pose no cata­logue, qual­ity or lin­eage duty. The UAE fed­eral pic­ture is the one genu­inely await­ing an event. Fed­eral De­cree-Law 45 of 2021 has been in force since 2 Janu­ary 2022 and its Ex­ec­ut­ive Reg­u­la­tions had still not been is­sued as at early Oc­to­ber 2026. A claim of a 1 Janu­ary 2027 com­pli­ance dead­line cir­cu­lates on vendor sites, cites no in­stru­ment and is con­tra­dicted by the ma­jor­ity of sources: it should not be used. The Fed­eral Au­thor­ity for Ar­ti­fi­cial In­tel­li­gence and Data, cre­ated by Cab­inet de­cision on 14 June 2026 and ab­sorb­ing the Data Of­fice, is the body ex­pec­ted to issue them, and the ex­pect­a­tion is a fore­cast rather than an ob­lig­a­tion.

India's data-man­age­ment in­stru­ment, by con­trast, be­haves ex­actly like the Gulf's. The Na­tional Frame­work

Data EngineeringVolge Quarterly24 •

for Data Shar­ing was ap­proved by the Com­mit­tee of Sec­ret­ar­ies on 15 May 2026, hav­ing been pre­pared by MeitY, and rests on four pil­lars. It re­quires every Min­istry and De­part­ment to con­sti­tute a Data Gov­ernance Com­mit­tee, with State com­mit­tees chaired by Chief Sec­ret­ar­ies. It car­ries no com­mence­ment date, no dead­line and no pen­alty: it is ad­min­is­trat­ive rather than stat­utory. The live tracker shows State com­mit­tees con­sti­tuted in about 19 of 36 States and Union Ter­rit­or­ies, and State data shar­ing policies pub­lished in nine. An earlier draft Na­tional Data Gov­ernance Frame­work Policy, from 2022, was never no­ti­fied at all. The pat­tern is there­fore not a Gulf pe­cu­li­ar­ity. Across eight jur­is­dic­tions, per­sonal data at­tracts dates and pen­al­ties, and data man­age­ment at­tracts com­mit­tees and audits.

The Indian contrast, which matters commercially

For a Gulf group with In­dian op­er­a­tions, and many have them, the In­dian cal­en­dar is firmer than any­thing in the Gulf. The Di­gital Per­sonal Data Pro­tec­tion Act 2023 was en­acted on 11 Au­gust 2023 with com­mence­ment de­ferred, and the DPDP Rules 2025 were no­ti­fied on 13 Novem­ber 2025 with a phased sched­ule. Rule 4, cov­er­ing con­sent man­ager re­gis­tra­tion, takes ef­fect about 13 Novem­ber 2026. Rules 3, 5 to 16, 22 and 23, to­gether with the sub­stant­ive sec­tions, take ef­fect about 13 May 2027. That May 2027 date is the hard­est dated data ob­lig­a­tion across all the jur­is­dic­tions ex­amined here, Gulf and In­dian to­gether, and it is the one around which a group pro­gramme can ac­tu­ally be se­quenced.

The takeaway is to stop plan­ning against dead­lines in the Gulf, be­cause there are al­most none, and to plan in­stead against the audit cal­en­dar. For a Saudi pub­lic en­tity that means the third quarter of every year. For a Kuwaiti en­tity, pub­lic or private, it means a quarterly sub­mis­sion that must in­clude the cata­logue it­self. For an Omani gov­ern­ment unit it means an an­nual min­istry as­sess­ment against a pri­or­ity tier whose year is not pub­lished, which ar­gues for treat­ing all three tiers as live. For an Abu Dhabi en­tity and its sup­pli­ers it means pro­gramme re­port­ing at an in­ter­val the en­tity will have to fix for it­self. In every case the com­pli­ance date is now, and the only ques­tion the in­stru­ment leaves open is when some­body will ask. For a group with In­dian op­er­a­tions the se­quen­cing is sim­pler, be­cause May 2027 is fixed and can be worked back­wards from. The Gulf of­fers no such an­chor, and an or­gan­isa­tion wait­ing for one will be wait­ing dur­ing the period in which it is already in de­fault.

The Calls

Ten dated, falsifiable judgements on Gulf data engineering in 2027: each written to be graded, hit, miss or partial, in the 2028 Scorecard.

The Calls at a glance

Ten judgements on Gulf data engineering in 2027, each dated, falsifiable and confidence-rated. The reasoning for each follows.

A forward-look that refuses to mark its own homework is a brochure. Each Call below is dated and falsifiable, and the 2028 edition will grade every one.

These are reasoned judge­ments, not guar­an­tees. Each is writ­ten to be graded, hit, miss or par­tial, in the 2028 Score­card, with a run­ning hit-rate re­por­ted by con­fid­ence band. High-con­fid­ence Calls carry struc­tural or already-vis­ible sup­port; lower-con­fid­ence Calls turn on ex­e­cu­tion or dis­clos­ure that can­not yet be ob­served.

  • Call 1: Cross-en­tity data shar­ing reaches pro­duc­tion, Me­dium, Re­solves 2027
  • Call 2: The data-en­gin­eer­ing tal­ent gap stays bind­ing, High, Re­solves 2027
  • Call 3: Sov­er­eignty-by-design be­comes a de­fault RFP re­quire­ment, High, Re­solves 2027
  • Call 4: The Mi­crosoft Saudi re­gion an­chors a mi­gra­tion wave, Me­dium–High, Re­solves 2027
  • Call 5: A ma­jor­ity of GenAI ini­ti­at­ives stall or are re-scoped, Me­dium, Re­solves 2027
  • Call 6: A na­tional en­tity re­ports a data-at­trib­uted AI fail­ure, Me­dium, Re­solves 2027
  • Call 7: SI-led builds begin shift­ing to in-house teams, Low–Me­dium, Re­solves 2027
  • Call 8: FinOps-for-data be­comes a named func­tion, Me­dium, Re­solves 2027
  • Call 9: A data-mesh op­er­at­ing model is ad­op­ted at an an­chor, Me­dium, Re­solves 2027
  • Call 10: Ca­pa­city passes 1.7 GW while ma­tur­ity lags, High, Re­solves 2027

The gov­ernance scaf­fold­ing ex­ists (SDAIA in­teg­rates 360+ sys­tems and Qatar's pro­gramme stood up in Janu­ary 2025), but pro­duc­tion-grade cross-en­tity shar­ing is an en­gin­eer­ing mile­stone bey­ond gov­ernance, and the tal­ent con­straint makes the pace un­cer­tain [SDAIA, 2025; NPC, 2025].

Data EngineeringVolge Quarterly25 •

Calls one to three

1. Cross-entity data sharing reaches production

By end-2027, at least three GCC na­tional data plat­forms (the SDAIA Na­tional Data Bank tier, Qatar's Na­tional Data Pro­gram, and a UAE fed­eral or emir­ate plat­form) reach pro­duc­tion-grade cross-en­tity data shar­ing.

Con­fid­ence: Me­dium · Re­solves end-2027

2. The data-engineering talent gap stays binding

The GCC data-en­gin­eer­ing tal­ent gap re­mains bind­ing through 2027, with de­mand out­strip­ping loc­ally avail­able data en­gin­eers des­pite loc­al­isa­tion pro­grammes.

Con­fid­ence: High · Re­solves through 2027

This is a struc­tural, long-ap­pren­tice­ship short­age that cap­ital can­not clear at speed; even Saudi Ar­a­bia's own strategy con­cedes the scale, tar­get­ing 20,000+ spe­cial­ists by 2030. The con­straint is human and slow-mov­ing [NSDAI, 2020].

3. Sovereignty-by-design becomes a default RFP requirement

Sov­er­eignty-by-design, clas­si­fic­a­tion-driven res­id­ency, be­comes a de­fault re­quire­ment in re­quests for pro­posal for GCC gov­ern­ment data plat­forms by end-2027.

Con­fid­ence: High · Re­solves end-2027

The dir­ec­tion is already set by Abu Dhabi's 100% sov­er­eign-cloud tar­get, vendor of­fer­ings (Mi­crosoft Cloud for Sov­er­eignty, Or­acle Sov­er­eign Cloud) and the Dubai Data Manual's man­dat­ory stand­ards of July 2026 [DGE, 2025; DDSE, 2026].

Calls four to six

4. The Microsoft Saudi region anchors a migration wave

The Mi­crosoft Saudi Ar­a­bia East re­gion an­chors a wave of KSA data-plat­form mi­gra­tions through 2027, graded on mi­gra­tion volume rather than on the avail­ab­il­ity date, which is tar­geted for Novem­ber 2026 and un­con­firmed.

Con­fid­ence: Me­dium–High · Re­solves through 2027

Con­struc­tion on three sites was com­plete by Decem­ber 2024 and cus­tomer avail­ab­il­ity is tar­geted for Novem­ber 2026, un­con­firmed as at 9 Novem­ber 2026; a new in-re­gion hy­per­scaler with sov­er­eign con­trols is the nat­ural an­chor for com­pli­ant mi­gra­tions, though tim­ing risk re­mains [Mi­crosoft, 2024–26].

5. A majority of GenAI initiatives stall or are re-scoped

A ma­jor­ity of GCC en­ter­prise gen­er­at­ive-AI ini­ti­at­ives launched in 2025–26 stall or are re-scoped in 2027 pending data-found­a­tion work.

Con­fid­ence: Low–Me­dium · Re­solves through 2027

The base rate is con­tested: the MIT Pro­ject NANDA fig­ure meas­ures or­gan­isa­tions re­cord­ing no profit-and-loss im­pact rather than pi­lots fail­ing, and 2026 sur­veys spread widely on re­turns. Gart­ner puts fail­ure near 50%, and the Gulf's found­a­tion lags its com­pute; the "ma­jor­ity" threshold and the con­tested base rate are what tem­per con­fid­ence [MIT NANDA, 2025; Gart­ner, 2026].

6. A national entity reports a data-attributed AI failure

At least one GCC na­tional en­tity pub­licly re­ports an AI pro­ject fail­ure or re­scope ex­pli­citly at­trib­uted to data qual­ity or en­gin­eer­ing by end-2027.

Con­fid­ence: Me­dium · Re­solves end-2027

Given the fail­ure base rates and the re­gion's AI am­bi­tion, such a fail­ure is prob­able; the un­cer­tainty is whether it is dis­closed pub­licly and at­trib­uted ex­pli­citly to data rather than to other causes [Gart­ner; McKinsey, 2025–26].

Calls seven and eight

7. SI-led builds begin shifting to in-house teams

Sys­tem-in­teg­rator-led builds begin shift­ing to­wards in-house plat­form teams at at least two flag­ship GCC en­tit­ies by 2027.

Con­fid­ence: Low–Me­dium · Re­solves through 2027

The plat­form logic fa­vours in-house own­er­ship and con­sol­id­a­tion is vis­ible in in­fra­struc­ture, but the tal­ent bot­tle­neck dir­ectly con­strains the abil­ity to staff in-house teams, which caps con­fid­ence [DGE, 2025; Palo Alto Net­works, 2026].

8. FinOps-for-data becomes a named function

FinOps-for-data be­comes a named func­tion or ex­pli­cit line item at major GCC en­tit­ies as cloud-data spend out­paces value in 2027.

Con­fid­ence: Me­dium · Re­solves through 2027

The spend-value gap is now large enough to force scru­tiny ($44.5bn pro­jec­ted cloud waste, $12.9m per-org data-qual­ity cost), and cheap power has masked in­ef­fi­ciency that fin­ance func­tions will begin to meter [Har­ness, 2025; Gart­ner via IBM, 2026].

Data EngineeringVolge Quarterly26 •

Calls nine and ten

9. A data-mesh operating model is adopted at an anchor

A data-mesh or data-product op­er­at­ing model is ad­op­ted at at least one GCC an­chor (Ara­mco, ADNOC or a major bank) by end-2027.

Con­fid­ence: Me­dium · Re­solves end-2027

The largest Gulf en­ter­prises have the scale, the do­main com­plex­ity and the in-house en­gin­eer­ing depth to adopt a data-product model; the con­straint is or­gan­isa­tional change rather than cap­ab­il­ity at these spe­cific an­chors.

10. Capacity passes 1.7 GW while maturity lags

GCC data-centre live IT ca­pa­city passes 3 GW by end-2027, but data-en­gin­eer­ing ma­tur­ity lags the mega­watt buil­dout. The ori­ginal 1.7 GW bar was passed at pub­lic­a­tion: Knight Frank's 2026 out­look already places the Middle East at 2.2 GW in 2026.

Con­fid­ence: High · Re­solves through 2027

Live ca­pa­city has run about 1.2 GW at end-2025 to 2.2 GW in 2026 and 3.2 GW fore­cast for 2027 [Knight Frank, 2026] and is com­pound­ing, while the found­a­tion-gap evid­ence shows ma­tur­ity trail­ing; the di­ver­gence of the two curves is the edi­tion's thesis in one line [Knight Frank / DC Byte; MIT NANDA, 2025].

How the Calls are made

Each Call is constructed to be falsifiable, dated to a resolution year, confidence-rated, and graded in the following edition.

Volge Quarterly ex­ists to make a for­ward-look that can be held to ac­count, and the method is de­signed to that end. A can­did­ate Call must sat­isfy four tests be­fore it is ad­mit­ted: it must be falsifi­able, so that a reader can state in ad­vance what evid­ence would prove it wrong; it must be dated, resolv­ing within a stated year; it must be groun­ded, tra­cing to sourced and year-stamped evid­ence in the pre­ced­ing parts; and it must be con­fid­ence-rated, so that the reader knows how much weight the ed­it­ors place on it. Calls that fail any test are set aside as watch­list items rather than dressed up as judge­ments.

The confidence bands

Four bands are used. High is re­served for Calls with struc­tural sup­port that is already vis­ible and slow to re­verse, such as the tal­ent con­straint and the di­ver­gence of ca­pa­city from ma­tur­ity. Me­dium–High marks Calls with strong sup­port but a re­sid­ual tim­ing or ex­e­cu­tion

risk. Me­dium marks Calls that are more likely than not but turn on a threshold (a "ma­jor­ity", a spe­cific count) that could fall either way. Low–Me­dium marks Calls where the dir­ec­tion is right but the pace is genu­inely un­cer­tain, typ­ic­ally be­cause the tal­ent bot­tle­neck con­strains it.

Grading

The 2028 edi­tion will open with the Score­card. Each of the ten Calls will be graded hit, miss or par­tial, and a run­ning hit-rate will be re­por­ted by con­fid­ence band, so that the cal­ib­ra­tion of the ed­it­ors, not merely the dir­ec­tion of their judge­ment, is ex­posed to scru­tiny. It is to be noted that a Call graded par­tial is a Call the method did not state sharply enough, and such out­comes are used to tighten the fol­low­ing year's draft­ing.

On sourcing

Every fig­ure in this edi­tion is drawn from a named third-party source and stamped with its year. Where sources dis­agree on scope, as they do on mar­ket size and ca­pa­city, a range is given rather than a false point es­tim­ate. Fig­ures were cur­rent to July 2026.

The year by domain

The same year, read through four lenses (technology, markets, regulation and delivery) and then stress-tested against three scenarios and a watchlist.

Technology and markets

The technology direction is consolidation and governance; the market direction is spend outpacing value until discipline arrives.

Technology

The tech­no­lo­gical through-line of 2027 is the move­ment from pro­cure­ment to op­er­a­tion. The re­gion has ac­quired the com­pon­ents of a mod­ern data stack (hy­per­scaler re­gions, sov­er­eign clouds, cata­logue and gov­ernance tool­ing), and the work of the year is to in­teg­rate and op­er­ate them as plat­forms rather than to buy more of them. The ar­rival of the Mi­crosoft Saudi Ar­a­bia East re­gion at cus­tomer avail­ab­il­ity tar­geted for Novem­ber 2026, on three avail­ab­il­ity zones with con­struc­tion com­plete since Decem­ber 2024, gives Saudi en­tit­ies an in-re­gion, sov­er­eign-cap­able an­chor for mi­gra­tion [Mi­crosoft, 2024–26]. Ex­pect con­sol­id­a­tion of tool­ing, en­forced clas­si­fic­a­tion car­ried through pipelines, and re­trieval in­fra­struc­ture built to feed mod­els groun­ded con­text.

Data EngineeringVolge Quarterly27 •

Markets

The mar­ket read is that spend con­tin­ues to com­pound faster than value is real­ised. The GCC cloud mar­ket above thirty bil­lion US dol­lars in 2025 and the data-centre mar­ket at 2.39 bil­lion, com­pound­ing at 26.78 per cent to 2034, de­scribe de­mand that is not in ques­tion [Mordor; IMARC, 2025]. What is in ques­tion is the re­turn: pro­jec­ted cloud waste of 44.5 bil­lion US dol­lars glob­ally in 2025, and 29 per cent of in­fra­struc­ture and plat­form spend wasted in 2026 [Flex­era, State of the Cloud 2026] and a per-or­gan­isa­tion data-qual­ity cost near 12.9 mil­lion sug­gest that a grow­ing share of the spend is not yet pro­duct­ive [Har­ness, 2025; Gart­ner via IBM, 2026]. The mar­ket cor­rec­tion of 2027 is not a fall in spend but the be­gin­ning of its meas­ure­ment.

The correction of 2027 is not a fall in spend but the beginning of its measurement.

Regulation and delivery

Regulation is tightening around classification and residency; delivery is shifting from the integrator to the in-house platform team.

Regulation

The reg­u­lat­ory dir­ec­tion is to­wards demon­strable, clas­si­fic­a­tion-driven gov­ernance. Saudi Ar­a­bia's NDMO frame­work of fif­teen do­mains, sev­enty-seven con­trols and 191 spe­cific­a­tions sets a con­crete com­pli­ance sur­face; Dubai's up­dated Data Manual of 1 July 2026 adds man­dat­ory stand­ards, self-as­sess­ment check­lists and an AI-read­i­ness scope; and Abu Dhabi's strategy com­mits to full sov­er­eign-cloud ad­op­tion for gov­ern­ment [NDMO, 2022; DDSE, 2026; DGE, 2025]. The not­able gap is the UAE's Per­sonal Data Pro­tec­tion Law, in force since 2 Janu­ary 2022 but with ex­ec­ut­ive reg­u­la­tions still un­pub­lished in mid-2026, a source of en­gin­eer­ing un­cer­tainty for any fed­eral-scope plat­form.

Delivery

The de­liv­ery model is the do­main in most vis­ible flux. The in­teg­rator-led pro­ject, which has sup­plied most Gulf data cap­ab­il­ity to date, is giv­ing way, slowly and at the lead­ing en­tit­ies first, to owned plat­forms op­er­ated by in-house teams. The Abu Dhabi Uni­fied Gov­ern­ment Data Centre for more than forty en­tit­ies, an­nounced with e& in Oc­to­ber 2025, is the pat­tern in

phys­ical form: build once, op­er­ate cent­rally, serve many [DGE, 2025]. The con­straint on the shift is the tal­ent bot­tle­neck, which is why the cor­res­pond­ing Call is held only at Low–Me­dium con­fid­ence. De­liv­ery, in short, knows where it must go and is lim­ited by whom it can hire.

The domain read in one line

Tech­no­logy con­sol­id­ates, mar­kets begin to meter, reg­u­la­tion de­mands demon­strable clas­si­fic­a­tion, and de­liv­ery moves in-house: all four point­ing at the same found­a­tion work the re­gion has de­ferred.

Three ways the year runs

The central case is that the foundation gap binds but begins to close at the leading entities; the tails turn on talent and on discipline.

  • Bull: the found­a­tion closes, The lead­ing tier con­verts its gov­ernance scaf­fold­ing into op­er­at­ing plat­forms. Cross-en­tity shar­ing reaches pro­duc­tion at three or more na­tional plat­forms; the Saudi re­gion an­chors a clean mi­gra­tion wave; FinOps-for-data and in-house teams take hold at flag­ship en­tit­ies; and the first groun­ded, pro­duc­tion AI use-cases re­turn meas­ur­able value. The 95-per-cent fail­ure rate be­gins to fall where the found­a­tion has been built. Tal­ent re­mains tight but is met by dur­able in-house teams rather than ren­ted cap­ab­il­ity.
  • Base: the gap binds, then nar­rows, The most likely path. Ca­pa­city passes 1.7 GW and keeps com­pound­ing while data-en­gin­eer­ing ma­tur­ity lags; a ma­jor­ity of 2025–26 GenAI ini­ti­at­ives stall or are re-scoped pending found­a­tion work; at least one na­tional en­tity re­ports a data-at­trib­uted fail­ure. But sov­er­eignty-by-design be­comes a de­fault RFP re­quire­ment, FinOps-for-data is named, and the lead­ing en­tit­ies begin the shift to plat­forms. The gap binds in 2027 and starts to nar­row at the top of the mar­ket.
  • Bear: abund­ance de­fers the reck­on­ing, Cheap power and deep cap­ital allow the re­gion to keep buy­ing com­pute and mod­els without con­front­ing the found­a­tion. Tool sprawl and in­teg­rator de­pend­ence per­sist; the tal­ent gap is met by more out­sourcing rather than in-house build; cloud waste com­pounds past its 44.5-bil­lion-dol­lar 2025 pro­jec­tion. The fail­ure rate holds near 95 per cent, the AI value PwC pro­jec­ted slips, and the reck­on­ing is de­ferred to 2028 at a higher cost.

Scen­arios are Volge con­struc­tions built on the sourced fig­ures in Parts I and III. They are il­lus­trat­ive reas­on­ing about the year, not fore­casts with as­signed prob­ab­il­it­ies. Cur­rent to July 2026.

Data EngineeringVolge Quarterly28 •

The watchlist

The signals to watch through the year: the seed of next year's Calls.

The Mi­crosoft Saudi Ar­a­bia East re­gion's gen­eral-avail­ab­il­ity date, and the mi­gra­tion volume that ac­tu­ally fol­lows it through 2027. Pub­lic­a­tion of the UAE Per­sonal Data Pro­tec­tion Law ex­ec­ut­ive reg­u­la­tions, un­re­solved since the law took force on 2 Janu­ary 2022. Pro­duc­tion cross-en­tity data shar­ing at the SDAIA Na­tional Data Bank tier, Qatar's Na­tional Data Pro­gram and a UAE fed­eral or emir­ate plat­form, the test of Call 1. Ad­op­tion of the up­dated Dubai Data Manual's AI-read­i­ness scope and self-as­sess­ment check­lists, launched 1 July 2026, across Dubai en­tit­ies. De­liv­ery of the Abu Dhabi Uni­fied Gov­ern­ment Data Centre for 40-plus en­tit­ies, and whether the "AI-nat­ive gov­ern­ment" 2027 tar­get is met. Any pub­lic dis­clos­ure by a na­tional en­tity of an AI fail­ure or re­scope at­trib­uted ex­pli­citly to data qual­ity or en­gin­eer­ing, the test of Call 6. The first named FinOps-for-data func­tion or line item at a major Gulf en­tity, and evid­ence of tool-stack con­sol­id­a­tion. Ad­op­tion of a data-mesh or data-product op­er­at­ing model at a Gulf an­chor such as Ara­mco, ADNOC or a major bank. Middle East live IT ca­pa­city against the 1,669-mega­watt 2026 fore­cast, and whether ma­tur­ity met­rics begin to close on the mega­watts.

Next year

The 2028 edi­tion opens with the Score­card: each of these ten Calls graded hit, miss or par­tial, with a run­ning hit-rate by con­fid­ence band. A for­ward-look that re­fuses to mark its own home­work is just a bro­chure; Volge Quarterly marks it.

Re­con­ciled from the sources lis­ted. Ranges re­flect genu­ine dis­agree­ment on scope and defin­i­tion. Cur­rent to July 2026.

What to watch, and the record

The operator's response to the year's foundation gap, and the reference matter (data appendix, sources and glossary) behind the narrative.

The data appendix

The numbers behind the narrative, in one place, each sourced and year-stamped.

MetricValueSource & year
GCC cloud market, 2025>$30B (KSA $15.39B; UAE $12.84B)Mordor Intelligence, 2025
KSA cloud, 2025→2030$15.39B → $27.93B (12.66% CAGR)Mordor Intelligence, 2025
UAE cloud, 2025→2031$12.84B → $56.26B (27.93% CAGR)Mordor Intelligence, 2025
GCC data-centre market$2.39B (2025) → $21.67B (2034), 26.78%IMARC, 2025
MEA colocation, 2025→2031$3.33B → $12.88B (25.32% CAGR)Arizton, 2025
ME live IT capacity761 → 1,115 → 1,669 MW (2024→26f)Knight Frank / DC Byte, 2024–26
UAE share of GCC capacity, 202535.2%IMARC / Business Wire, Jan 2026
GenAI pilots returning nothing~95%MIT NANDA / Forbes, Oct 2025
GenAI projects that fail~50%Gartner, Jan 2026
Agentic-AI projects cancelled by end-2027>40%Gartner, Jun 2025
Poor data quality cost, per org~$12.9M / yearGartner via IBM, Jan 2026
Projected cloud waste, 2025$44.5BHarness, 2025
Machine identities per human109 : 1Palo Alto Networks, 2026
Gulf industrial power$0.05–0.07 / kWhOxford Economics, Jan 2026
PIF assets~SAR 3.4T at end-2025 [PIF annual report]Sovereign disclosures, 2025–26
AI to the Middle East by 2030~$320B (KSA ~$135B; UAE ~$96B)PwC, 2023–24
SDAIA framework15 domains, 77 controls, 191 specificationsNDMO Standards, 2022
Saudi National Data Bank360+ government systems integratedSDAIA, 2025
Data EngineeringVolge Quarterly29 •

Sources

Every figure in this edition traces to a named source below, stamped with its year. Figures were current to July 2026.

  • MIT NANDA, "State of AI in Busi­ness 2025", re­por­ted via For­bes, 15 Oc­to­ber 2025 (~95% of en­ter­prise GenAI pi­lots de­liver no re­turn). https://www.for­bes.com/
  • Gart­ner, "Why Half of GenAI Pro­jects Fail", 26 Janu­ary 2026. https://www.gart­ner.com/
  • Gart­ner, AI in­fra­struc­ture-and-op­er­a­tions pro­jects stall pre-ROI, 7 April 2026. https://www.gart­ner.com/
  • Gart­ner, >40% of agen­tic-AI pro­jects can­celled by end-2027, 25 June 2025. https://www.gart­ner.com/
  • McKinsey & Com­pany, "The state of AI in GCC coun­tries: In pur­suit of scale and value", 13 Novem­ber 2025. https://www.mckin­sey.com/
  • Gart­ner, cost of poor data qual­ity ~$12.9M/yr per or­gan­isa­tion, via IBM, 23 Janu­ary 2026. https://www.ibm.com/
  • Har­ness, pro­jec­ted cloud waste $44.5bn in 2025. https://www.har­ness.io/
  • Mordor In­tel­li­gence, GCC / KSA / UAE cloud mar­ket siz­ing, 2025. https://www.mor­dor­in­tel­li­gence.com/
  • IMARC Group, GCC data-centre mar­ket $2.39bn (2025) → $21.67bn (2034), 2025. https://www.imar­cgroup.com/
  • Ar­izton, MEA coloca­tion mar­ket $3.33bn (2025) → $12.88bn (2031), 2025. https://www.ar­izton.com/
  • Knight Frank / DC Byte, Middle East live IT ca­pa­city 761→1,115→1,669 MW (2024–26). https://www.knight­frank.com/
  • IMARC / Busi­ness Wire, UAE 35.2% of GCC ca­pa­city, Janu­ary 2026. https://www.busi­ness­wire.com/
  • SDAIA, Royal Order A/471 (30 Au­gust 2019); Na­tional Data Bank; clas­si­fic­a­tion frame­work, 2019–2025. https://sdaia.gov.sa/
  • Na­tional Strategy for Data & AI (NSDAI), ap­proved 17 July 2020 (20,000+ spe­cial­ists; ~SAR 75bn). https://sdaia.gov.sa/
  • NDMO, Na­tional Data Man­age­ment & Per­sonal Data Pro­tec­tion Stand­ards: 15 do­mains, 77 con­trols, 191 spe­cific­a­tions, 2022. https://sdaia.gov.sa/ndmo/
  • Fed­eral Com­pet­it­ive­ness and Stat­ist­ics Centre (FCSC), UAE Smart Data Frame­work and bay­anat.ae. https://bay­anat.ae/
  • UAE Per­sonal Data Pro­tec­tion Law, Fed­eral De­cree-Law 45/2021, in force 2 Janu­ary 2022. https://u.ae/
  • Gov­ern­ment of Dubai, Dubai Data Law 26/2015; DDSE (Law 24/2023); Dubai Di­gital Au­thor­ity (Law 22/2023). https://www.di­git­al­dubai.ae/
  • Dubai Data and Stat­ist­ics Es­tab­lish­ment, up­dated Dubai Data Manual, launched 1 July 2026. https://www.di­git­al­dubai.ae/
  • De­part­ment of Gov­ern­ment En­able­ment (Abu Dhabi), Gov­ern­ment Di­gital Strategy 2025–2027; DGE & e& Uni­fied Gov­ern­ment Data Centre, 17 Oc­to­ber 2025. https://www.tamm.abud­h­abi/
  • Qatar Na­tional Plan­ning Coun­cil (Amiri De­cision 13/2024); Na­tional Data Pro­gram, Janu­ary 2025; Na­tional Data & Stat­ist­ics Strategy, 4 May 2025. https://www.npc.qa/
  • Mi­crosoft, Saudi Ar­a­bia East re­gion (cus­tomer avail­ab­il­ity tar­geted Novem­ber 2026); Cloud for Sov­er­eignty (GA Decem­ber 2023); Mi­crosoft–G42, April 2024. https://azure.mi­crosoft.com/
  • AWS, Bahrain (July 2019), UAE (Au­gust 2022), KSA me-cent­ral-2 (an­nounced March 2024, launch 2026). https://aws.amazon.com/
  • Google Cloud, Doha (2023) and Dammam (15 Novem­ber 2023) re­gions; Or­acle opened a second Saudi pub­lic cloud re­gion in Riy­adh, hos­ted by Center3, in Oc­to­ber 2024. https://cloud.google.com/
  • Or­acle, UAE and KSA re­gions; $14bn KSA com­mit­ment, May 2025. https://www.or­acle.com/
  • Ox­ford Eco­nom­ics, Gulf in­dus­trial power $0.05–0.07/kWh, 22 Janu­ary 2026. https://www.ox­for­de­co­nom­ics.com/
  • The Eco­nom­ist, "Saudi wants world's cheapest data centres", 17 Decem­ber 2025. https://www.eco­nom­ist.com/
  • PwC, AI to add ~$320bn to the Middle East by 2030 (KSA ~$135bn; UAE ~$96bn). https://www.pwc.com/
  • Palo Alto Net­works, 109 ma­chine iden­tit­ies per human, 2026 Iden­tity Se­cur­ity Land­scape. https://www.pa­loaltonet­works.com/

Defin­i­tions are Volge's, con­sist­ent with the na­tional-pro­gramme and ana­lyst sources cited through­out. Cur­rent to July 2026.

Data EngineeringVolge Quarterly30 •

Glossary

The terms this edition uses, defined for the operator and the board.

SDAIA: Saudi Data and Ar­ti­fi­cial In­tel­li­gence Au­thor­ity; the na­tional body hous­ing the NIC, NDMO and NCAI. NDMO: Na­tional Data Man­age­ment Of­fice (Saudi Ar­a­bia); au­thor of the 15-do­main, 77-con­trol, 191-spe­cific­a­tion data frame­work. Na­tional Data Bank: The Saudi na­tional data lake in­teg­rat­ing 360+ gov­ern­ment sys­tems. FCSC: Fed­eral Com­pet­it­ive­ness and Stat­ist­ics Centre (UAE); op­er­ator of the fed­eral open-data portal bay­anat.ae. PDPL: Per­sonal Data Pro­tec­tion Law; UAE Fed­eral De­cree-Law 45/2021, in force since 2 Janu­ary 2022, ex­ec­ut­ive reg­u­la­tions pending mid-2026. DDSE: Dubai Data and Stat­ist­ics Es­tab­lish­ment; the emir­ate data body under the Dubai Di­gital Au­thor­ity. Dubai Pulse: Dubai's cent­ral data plat­form, de­liv­er­ing plat­form, data and in­fra­struc­ture as a ser­vice. DGE: De­part­ment of Gov­ern­ment En­able­ment (Abu Dhabi); ab­sorbed the Abu Dhabi Di­gital Au­thor­ity in 2023. NPC: Na­tional Plan­ning Coun­cil (Qatar); es­tab­lished 2024, run­ning the Na­tional Data Pro­gram. Data en­gin­eer­ing: The dis­cip­line of in­gest­ing, clean­ing, cata­loguing, gov­ern­ing and in­teg­rat­ing data into a found­a­tion us­able by ana­lyt­ics and AI. Sov­er­eignty by design: Ar­chi­tect­ing a plat­form so that clas­si­fic­a­tion-driven res­id­ency and ac­cess are en­forced from the first line, not ap­plied af­ter­wards. FinOps: Fin­an­cial op­er­a­tions ap­plied to cloud and data spend; meas­ur­ing, at­trib­ut­ing and op­tim­ising the es­tate as a man­aged cost. Data mesh / data product: An op­er­at­ing model that treats data as an owned, served product per do­main rather than a cent­ral mono­lith. Data clas­si­fic­a­tion: The scheme (e.g. Saudi Ar­a­bia's four levels) by which data is la­belled to gov­ern res­id­ency, ac­cess and hand­ling. Agen­tic AI: AI sys­tems that plan and act across mul­tiple steps; noted by Gart­ner as fa­cing a >40% can­cel­la­tion rate by end-2027.

Data EngineeringVolge Quarterly31 •
Data Engineering

The Gulf can buy any model on the market. It cannot buy, at the speed it wishes, the engineered data foundation those models require to be worth anything.

Version · VSP-OTG-005 · 2026
volgePublished
every quarter
About Volge Quarterly

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.

Terms of use

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.