Contents
The Voice of Volge is Volge’s thought-leadership series, unsigned and institutional, every figure sourced, quantified and year-stamped. Volge is a Flexera delivery partner; interest disclosed.
volge.com/voice
About Flexera
Flexera is a technology-spend management software company. What began as software-licence optimisation is now a single platform that discovers, values, and governs an organisation's entire technology landscape, the software it owns, the cloud it rents, the SaaS it subscribes to, and, since 2026, the AI it meters, built out over two decades and a sequence of pointed acquisitions.
+ Chaos Genius
The recent chapters matter most to a buyer today: ProsperOps and Chaos Genius folded into the FinOps portfolio, and, at FinOps X on 12 June 2026, an AI Cost Management platform, the FinOps Assist assistant, and outcome-based ProsperOps+ pricing.
| See | IT Visibility · Technopedia · SaaS Management discovery |
| Value & govern | Cloud Cost Optimization (FinOps) · SaaS Management · Software Asset Management |
| Optimise | ProsperOps (autonomous rate) · Chaos Genius (data-cloud) · AI Cost Management (2026) |
Source: Flexera product and customer materials, 2026. Customer outcomes as reported by Flexera and ProsperOps.
Flexera did not start in FinOps; it arrived through licence management and bought its way across cloud, SaaS, and now AI. That breadth is the pitch and the risk: one landscape view if the pieces integrate, another integration project if they do not.
The bill no one signed off.
Cloud, SaaS, and AI were each bought to save money. Metered, un-governed, and compounding, they have become the fastest-growing line item most boards cannot itemise.
The purchase became a meter,
and the meter never stops
A decade ago, technology was something a company bought. It signed a licence, took delivery, capitalised the cost, and owned the thing. Today it rents almost all of it. The servers are someone else's cloud, billed by the second. The software is a subscription, billed by the seat. And the newest line, artificial intelligence, is billed by the token, the call, and the query. The capital expense became an operating expense, and the operating expense became a meter that runs whether anyone is watching or not.
This is the technology tax: not a single charge but a compounding one, spread across three meters that were each installed for good reasons and are now, in most organisations, un-governed as a whole. Cloud was adopted to trade capital for elasticity. SaaS was adopted to trade maintenance for immediacy. AI is being adopted to trade labour for speed. Every one of those trades is defensible. The sum of them, unwatched, is not.
The figures are no longer marginal. Flexera's 2026 State of the Cloud Report, its fifteenth annual edition, finds 85 per cent of organisations citing the management of cloud spend as a top challenge, and, for the first time in five years, a 29 per cent rise in wasted cloud spend (Flexera, 18 March 2026). Zylo's 2026 SaaS Management Index, working from 75 billion US dollars of tracked spend, finds roughly 36 per cent of SaaS licences sitting unused against utilisation benchmarks (Zylo, 2026). And AI spend, the newest meter, is rising fastest of all.
What makes this a governance problem rather than a procurement one is the shape of the spending. A licence is a decision made once, by someone accountable. A meter is a decision made continuously, by everyone, the engineer who leaves a cluster running, the team that expenses a new tool, the application that calls a model in a loop. The cost accrues below the altitude at which anyone signs. By the time it reaches the board, it is a total, not a set of choices.
Flexera's answer is to put a single instrument across all three meters. Cloud Cost Optimization, with ProsperOps and Chaos Genius, governs the cloud meter. SaaS Management governs the subscription meter. And the AI Cost Management platform, announced in June 2026, extends the same discipline, visibility, benchmarking, optimisation, to the agents, models, and compute behind the AI meter. One landscape, one method, three meters.
Volge licenses, configures, and operates that instrument on our clients' behalf, and we wrote this book to make the case for buying that capability: with our commercial interest marked wherever it appears. The discipline it describes has a name the industry has settled on: FinOps. The argument of this book is that FinOps can no longer stop at cloud.
Read on for the three meters in turn, cloud, then SaaS, then AI, and then for the operator's read: how to govern all three as one landscape rather than three separate surprises.
What the money says
The FinOps discipline
Autonomous optimisation
2 · Cloud & AI, 2024–26
3 · Rate vs workload
The cloud bill runs away because no single person decides it
Cloud spending is not a purchase; it is the sum of a million small decisions, most of them made by people with no view of the bill and no reason to care about it. That is the elasticity trap. The very property that makes cloud valuable, that anyone can provision anything, instantly, without asking finance, is the property that makes it leak. The meter is democratic. The accountability is not.
Consider how the money actually moves. An engineer spins up a cluster to test something and forgets to turn it off. A team picks the instance size that is safe rather than the one that is right, because over-provisioning is invisible and under-provisioning gets you paged. A workload that ran fine at a tenth of the scale keeps its architecture as it grows, and the architecture, not the traffic, becomes the cost. None of these is a scandal. Each is a rational local choice. Summed across an enterprise, they are the waste line.
The pricing model compounds the problem. Cloud is priced to reward commitment, reserved instances, savings plans, committed-use discounts, and to punish the on-demand consumption that flexible teams default to. Capturing those discounts is a portfolio optimisation problem that changes every day as workloads shift. Done by hand, on a spreadsheet, once a quarter, it is always out of date. Most organisations leave a large fraction of the available discount on the table not through negligence but through arithmetic they cannot keep up with.
Then there is the visibility gap. A cloud invoice is a telephone directory: hundreds of thousands of line items, tagged inconsistently, spread across accounts and providers, arriving a month after the spending happened. Answering the simplest governance question, which team spent what, on what, and was it worth it, requires normalising that directory into something a business can read. That normalisation is the first thing a FinOps practice builds, and the first thing most organisations lack.
The result is predictable and, in 2026, measurable. Flexera's State of the Cloud Report finds 85 per cent of organisations naming cloud-spend management a top challenge, and 76 per cent of large enterprises now spending more than five million US dollars a month on cloud (Flexera, 18 March 2026). At that scale, a few points of un-captured discount or un-retired waste is not a rounding error. It is a hire, a project, a quarter.
| Meter | Billed by | Who spends it | Flexera instrument |
|---|---|---|---|
| Cloud | The second, compute, storage, egress | Engineers, continuously | Cloud Cost Optimization · ProsperOps |
| SaaS | The seat, subscriptions, renewals | Business units, on expense | SaaS Management |
| AI | The token, models, agents, inference | Applications and teams, in loops | AI Cost Management (2026) |
For the first time in five years, cloud waste is rising: while AI pours onto the same bill
The direction of cloud spending has been settled for a decade; what changed in 2026 is the direction of cloud waste. After years of slow improvement as FinOps practices matured, Flexera's fifteenth State of the Cloud Report records a 29 per cent rise in wasted cloud spend, the first increase in five years (Flexera, 18 March 2026). The proximate cause is not hard to find: artificial intelligence has arrived on the cloud bill, and it is being consumed faster than it is being governed.
GenAI adoption rose from 47% (2024) to 72% (2025) to 81% (2026); 45% now report extensive production use, up from 36% in 2025.
Read those two facts together and the mechanism is plain. Adoption of generative AI jumped from 47 per cent of organisations in 2024 to 81 per cent in 2026, and nearly half now run it extensively in production. That production traffic runs on cloud compute, the most expensive compute in the catalogue, and it runs in loops that a human never sees. The cloud meter did not slow down; it acquired a new, hungry consumer, and the waste line turned upward for the first time in five years.
The governance response has matured, but unevenly. Flexera finds 63 per cent of organisations with an established FinOps team and 71 per cent operating a Cloud Centre of Excellence (Flexera, 18 March 2026). That is real progress. But a discipline built to govern predictable virtual machines is now being asked to govern unpredictable inference, and the tooling, the benchmarks, and the habits are all a step behind the spending. The practice exists; the practice is behind.
FinOps is the discipline; Cloud Cost Optimization is where Flexera runs it
FinOps, the operating practice of bringing financial accountability to variable cloud spend, is not a Flexera invention; it is an industry discipline with its own foundation, its own certification, and its own open billing standard, FOCUS. What Flexera sells is the instrument that runs it: Cloud Cost Optimization, the platform descended from its 2018 acquisition of RightScale, now joined by ProsperOps and Chaos Genius. The distinction matters, because it is the honest way to describe the value. The method is the industry's. The tooling is Flexera's.
The FinOps loop is deliberately unglamorous: inform, optimise, operate. Inform means normalising the invoice into a shared, tagged, business-readable view of who spent what: the directory turned into an account. Optimise means acting on that view: right-sizing over-provisioned resources, retiring the forgotten ones, and capturing the commitment discounts the on-demand default leaves behind. Operate means making the loop a habit rather than a quarterly panic, with ownership assigned and budgets that mean something.
Cloud Cost Optimization supplies the inform and optimise layers across the major providers, reading the same FOCUS-formatted billing data the rest of the discipline has standardised on. It answers the governance questions, cost per team, per product, per unit of business output, that the raw invoice buries. And it turns the answers into a worklist: here is the waste, here is the discount you are missing, here is the resource no one has touched in ninety days.
The harder half of the loop is the discounts, because capturing them well is a continuous optimisation problem, not a one-off purchase. This is what ProsperOps automates. Rather than a human buying reserved instances once a quarter and hoping the workload does not shift, ProsperOps continuously manages the commitment portfolio, laddering, exchanging, and rebalancing to hold a high effective savings rate as consumption moves underneath it. Tealium, on ProsperOps, reports paying 37 cents per dollar of on-demand cloud; Duolingo reports a five-point gain in effective savings rate (Flexera, June 2026).
We will not oversell the automation. ProsperOps optimises the rate you pay for what you run; it does not decide whether you should be running it. A cluster that should not exist, optimally discounted, is still waste. That judgement, architectural waste, not rate waste, stays with engineering, and it is where the largest savings usually hide.
| Lever | What it changes | Done by |
|---|---|---|
| Rate optimisation | The price you pay for a given resource, commitments, discounts | ProsperOps, continuously and automatically |
| Workload optimisation | Right-sizing and scheduling of what runs | Cloud Cost Optimization + engineering |
| Architectural optimisation | Whether the workload should exist in that shape at all | Engineering judgement, not automatable |
The 2026 move: pay for the saving, not the software
The most interesting thing Flexera announced at FinOps X in June 2026 was not a feature but a pricing model. ProsperOps+ bundles autonomous optimisation with the wider FinOps portfolio on an outcome basis: the customer pays a share of quantified savings rather than an upfront licence (Flexera, 12 June 2026). It is a small change with a large signal: a vendor confident enough in the saving to be paid out of it.
Alongside it came Unified Autonomous Optimization, described as combining rate and workload optimisation under a single engine, generally available at announcement (Flexera, 12 June 2026). The ambition is to close the gap between the two levers in Exhibit 3: to stop treating "what price you pay" and "what you run" as separate projects owned by separate teams. Whether it delivers on that ambition at scale is a 2026 question we cannot yet answer from independent data, and we flag it as such.
The outcome-based model deserves a sceptical footnote as well as applause. Paying for quantified savings is attractive until you ask who quantifies them, against what baseline, and whether the baseline was optimised first. A saving measured against a deliberately un-optimised starting point is a smaller saving than it looks. This is precisely the kind of question a delivery partner exists to ask on the buyer's behalf, and we ask it.
None of this removes the need for the operating discipline. Automation captures the rate; it does not build the culture in which engineers see the cost of what they provision. The organisations that get cloud spend under control are the ones that make cost visible at the point of decision, in the pull request, in the dashboard the team actually reads, not the ones that buy the cleverest optimiser and leave the behaviour untouched.
Cloud spend runs away because consumption is decided continuously and accountably by no one. FinOps is the discipline that governs it; Flexera's Cloud Cost Optimization, ProsperOps, and Chaos Genius are where the discipline runs. Automation captures the rate; architecture and culture capture the rest. And in 2026 a new consumer: AI, turned the waste line upward. Part II turns to the second meter: the software you rent by the seat.
Who controls the spend
Discover, right-size, renew
5 · Where SaaS spend hides
SaaS sprawls because the person who buys it is not the person who pays for it
The defining fact about software-as-a-service is that IT no longer buys it. Marketing buys its own stack; sales buys its own; every function subscribes to the tools it likes, on a corporate card, without a purchase order. This is not a governance failure to be scolded away, it is the operating model SaaS was designed to enable, and it delivered speed the old procurement gate never could. But it means the spend is decided in one place and paid for in another, and the gap between them is where the waste lives.
The numbers describe a portfolio no one owns as a whole. Zylo's 2026 SaaS Management Index, built from 75 billion US dollars of tracked spend across 40 million licences, finds business units controlling 81 per cent of SaaS spend while IT directly manages just 15 per cent (Zylo, 2026). The median organisation spends 9,455 US dollars per employee per year on SaaS. And the applications keep arriving: large enterprises add an average of 21 new applications every month, even as the total count stays roughly flat: a churn that hides duplication behind a stable headline.
Duplication is the quiet cost. When each function buys its own tools, the enterprise ends up paying for four project trackers, three video platforms, and two of everything else, each under its own contract, each auto-renewing. No single tool is extravagant. The portfolio is. And because the contracts sit with the functions rather than with a central desk, no one sees the overlap until someone goes looking, which, in most organisations, no one is paid to do.
Then there is the expensed layer, growing fastest of all. Zylo finds expense-based SaaS spend up 267 per cent year on year, with ChatGPT now the single most expensed application (Zylo, 2026). This is shadow IT with a receipt: tools bought on personal expense, below the threshold anyone reviews, invisible to both IT and procurement until the aggregate appears in a finance report. The newest and fastest-growing of these tools are AI-native, which is where this part meets the next.
The consequence is not merely waste; it is exposure. Zylo reports 61 per cent of organisations forced to cut projects because of unplanned SaaS cost increases (Zylo, 2026). A portfolio that grows by expense and renews on autopilot is a portfolio whose cost is discovered, not decided, and a cost you discover is a cost that arrives at the worst possible time.
| Finding | Figure |
|---|---|
| Median SaaS spend per employee, per year | $9,455 |
| SaaS licences unused against utilisation benchmarks | ~36% |
| Share of SaaS spend controlled by business units | 81% |
| Growth in expense-based (shadow) SaaS spend, year on year | +267% |
| Organisations forced to cut projects on unplanned SaaS cost | 61% |
A third of what you buy, you do not use, and it renews anyway
If a single figure captures the SaaS opportunity, it is this: roughly 36 per cent of SaaS licences sit unused against utilisation benchmarks (Zylo, 2026). More than a third of the seats an organisation pays for are dormant, assigned to leavers, bought in bulk for a rollout that never happened, or provisioned by default to people who never signed in. Unlike cloud waste, which at least bought idle capacity, this is spend that bought nothing at all.
The spend is controlled where it cannot be seen centrally, and roughly a third of it buys a licence no one uses.
The waste persists because the renewal is where the bargaining power sits, and the renewal is exactly where attention is scarcest. A subscription auto-renews on a date the buying function has forgotten, at a quantity set a year ago, often with an uplift the contract permitted and no one contested. Reclaiming the dormant third requires knowing, before that date, who actually uses the tool, and that requires usage telemetry the finance system does not hold and the buying function never collected.
This is the case for treating SaaS as a managed landscape rather than a pile of contracts. The savings are not won by cancelling tools people love; they are won by right-sizing quantities to real usage, consolidating the duplicated four-into-one, and arriving at each renewal with data instead of a default. The discipline is the same FinOps loop as cloud, inform, optimise, operate, applied to the seat instead of the second.
Flexera SaaS Management turns a pile of contracts into a governed landscape
Flexera SaaS Management, the capability deepened by the 2024 Snow acquisition, exists to do for subscriptions what Cloud Cost Optimization does for compute: discover what exists, measure how it is used, and turn renewals from a default into a decision. The work divides into three jobs, and the order matters, you cannot right-size what you have not discovered, and you cannot renew well what you have not measured.
Discover. The first job is finding the landscape, including the parts bought on expense that no contract records. SaaS Management surfaces applications from expense feeds, single-sign-on logs, and financial data, assembling the shadow portfolio into a visible one. This is the step that turns "we think we have about eighty tools" into "we have two hundred and forty, and here are the eleven you are paying for twice."
Right-size. The second job is matching quantity to reality. By reading usage, who has logged in, how recently, how much, the platform separates the seats that earn their cost from the dormant third that does not. That evidence is what makes a reclaim defensible: not "we think finance over-bought Salesforce" but "these 340 licences have not been used in ninety days."
Renew. The third job is timing. A renewal calendar that fires before the date, with usage data attached, converts the renewal from an ambush into a negotiation. The organisation arrives knowing what it uses, what it duplicates, and what it can drop: which is the only position from which a subscription is ever renegotiated down.
We will state the limit plainly, because it is the same limit as every tool in this book. Discovery and telemetry find the waste and quantify it; they do not remove it. Cancelling a beloved-but-redundant tool, consolidating two functions onto one platform, and holding a hard line at renewal are organisational acts, not software outputs. The instrument makes the case unarguable. Someone still has to make it, and to absorb the internal friction of taking a tool away from the team that chose it.
SaaS sprawls because the business buys it and finance pays for it, and roughly a third of what is bought goes unused and renews regardless. Flexera SaaS Management discovers the landscape, right-sizes it to real usage, and turns renewals into decisions, but the decisions themselves stay human. Part III turns to the newest and fastest meter: the AI you consume by the token.
The meter you cannot predict
AI Cost Management & FinOps Assist
7 · Scope of AI cost governance
AI is the first cost centre that decides its own spend
Every meter so far has a human somewhere behind it, an engineer provisioning a cluster, a manager buying a seat. The AI meter is the first where the consumer is the software itself. An agent that reasons in a loop, calls a model, reads the answer, and calls again is spending money at machine speed, without a person in the transaction. That is what makes AI cost a genuinely new problem rather than a rebranded old one: the thing generating the bill is the thing you deployed to save money.
The pricing makes it worse. AI is billed by the token, fractions of a cent per unit of text in and out, which sounds trivial and compounds ferociously. A single agentic workflow can make dozens of model calls to answer one question; multiply by production traffic and the fractions become a material line. And because the cost scales with usage rather than with seats or instances, it is genuinely unpredictable: the same feature costs a different amount every day depending on how the world uses it.
The organisations feeling this first are the ones adopting fastest. Zylo finds AI-native application spend up 393 per cent year on year in organisations with more than 10,000 employees, and, the figure that should worry any CFO, 78 per cent of IT leaders reporting unexpected charges from consumption-based or AI pricing models (Zylo, 2026). "Unexpected" is the operative word. This is not overspending against a budget; it is spending with no budget to overspend against, because the meter is too new to have one.
And the AI meter does not respect the boundaries of the other two. A single AI feature might call a third-party model API (a SaaS-like charge), run inference on cloud GPUs (a cloud charge), and pull from a data platform billed separately (a data-cloud charge). The cost is smeared across all three meters at once, which means no single existing tool sees the whole of it. That is precisely the visibility gap Flexera's 2026 move is aimed at.
| Layer | Billed by | Governance difficulty |
|---|---|---|
| Models & APIs | Tokens in and out, per call | Non-deterministic; cost varies per query |
| Agents | Chained calls in a loop | Software, not humans, sets the volume |
| Compute | GPU-seconds of inference | The most expensive cloud capacity |
| Data platforms | Queries and storage for training/RAG | Billed separately, easily overlooked |
Flexera's 2026 move: put the FinOps meter across agents, models, and compute
At FinOps X on 12 June 2026, Flexera announced an AI Cost Management platform spanning, in its words, agents, models, data platforms, and compute, the four layers of Exhibit 6, with the same three-part promise it makes for cloud: see the spend, benchmark it, optimise it (Flexera, 12 June 2026). The claim is that AI cost is not a new discipline but the existing FinOps discipline extended to a new meter, and that the platform already governing cloud and SaaS is the natural place to govern it.
The visibility half is the part that matters first, because it is the part nobody has. Assembling AI spend into one view means pulling token bills from model providers, GPU costs from cloud, and query costs from data platforms into a single, normalised picture: the same normalisation Cloud Cost Optimization does for the cloud invoice, applied to a bill that currently arrives in three unrelated formats. Until that picture exists, "optimise" is premature; you cannot reduce what you cannot yet total.
Alongside the platform, Flexera introduced FinOps Assist, an AI-powered assistant inside Cloud Cost Optimization that answers cost questions and generates reports from natural-language prompts, with a stated path toward multi-agent orchestration of specialised tasks (Flexera, 12 June 2026). There is an irony worth naming: the tool for governing AI spend is itself an AI agent, and therefore itself a consumer of the meter it helps you read. Useful, and not free of the problem it addresses.
Here candour is required, and we would rather supply it than have a client discover it. This platform is weeks old at the time of writing. There is no independent, at-scale evidence yet that it governs AI spend as effectively as Cloud Cost Optimization governs cloud, because there has not been time for any to exist. We present it as the most complete answer currently on the market to a real and growing problem, not as a proven one. A buyer should treat 2026 AI cost governance as early, and price the risk accordingly.
| Capability | What it promises | Maturity, mid-2026 |
|---|---|---|
| Visibility | One view across models, agents, compute, data | New; the first and most needed step |
| Benchmarking | Compare AI spend against peers and baselines | Early; benchmarks still forming |
| Optimisation | Reduce consumption cost without cutting value | Unproven at scale; watch closely |
| FinOps Assist | Natural-language cost queries and reports | Rolling out in Cloud Cost Optimization |
The AI you rent is not the AI you own, and both need governing
This book governs the AI you rent: the consumption cost of models, agents, and inference, metered by the token and paid by the month. There is a second AI question it deliberately leaves aside: the AI you own and deploy as an asset: the models running in your landscape that must be inventoried, secured, and eventually retired. That is a matter of liability rather than of price, and it is the subject of the companion book. The distinction is worth drawing now, because the two are easily confused and governed differently.
The rented AI of this book is a spend problem: it appears on a bill, it scales with use, and the governance question is whether the value justifies the meter. The owned AI of the companion volume is a risk problem: a deployed model is an asset in your landscape with a vulnerability surface, a compliance status, and an end of life, whether or not it costs anything to run this month. One is measured in dollars per token; the other in exposure per asset.
Flexera sits on both sides, Cloud Cost Optimization and AI Cost Management on the spend side, IT Visibility and Software Vulnerability Manager on the asset side, which is the argument for reading the two books together. But keep the meters straight. Governing the price of the AI you rent will not tell you what the AI you own is exposing you to, and vice versa. This part has been about the price.
AI is the first meter where software, not people, sets the spend; it is priced by the token, consumed in loops, and smeared across cloud, SaaS, and data platforms at once. Flexera's June 2026 AI Cost Management platform is the most complete answer on the market and, candidly, the least proven. Part IV pulls the three meters together into a single governed landscape.
The independent case
Six buying questions
Objections · worked example · routes
Evidence · independent findings
9 · A bill, torn down
10 · Three routes
The three meters are one tax, and the point of a platform is to govern them as one
The argument of this book has treated cloud, SaaS, and AI in turn, but the reason to read them together is that a company does not experience three bills. It experiences one rising cost of running on rented technology, arriving from three directions, governed, if at all, by three teams who rarely compare notes. The case for a single platform is not that any one meter is unmanageable alone. It is that managing them separately guarantees the overlaps, the double-counting, and the AI charge that hides between the cloud and SaaS columns because it belongs to neither.
Consider where the meters bleed into each other. An AI feature runs on cloud GPUs, calls a SaaS-delivered model, and reads a data platform, one feature, three bills, three teams. A SaaS vendor re-prices to a consumption model and its cost migrates from the predictable seat column to the volatile usage column. A cloud team celebrates a saving that a SaaS team quietly reverses by buying the same capability as a subscription. Governed in silos, these moves are invisible; the enterprise optimises one meter and loses on another, and calls it progress.
A single landscape view is what makes the trade-offs visible. When cloud, SaaS, and AI spend sit in one normalised picture, the FOCUS-standardised, Technopedia-referenced view Flexera assembles, the questions a board actually wants to ask become answerable: what does this product line cost us across all three meters; is that AI feature cheaper as a subscription or as raw inference; where are we paying twice. That is the whole of the pitch, and it is a real one, provided the pieces integrate.
Which is where we keep our disclosed scepticism. Flexera assembled this breadth by acquisition, RightScale, Snow, ProsperOps, Chaos Genius, and a suite assembled by acquisition is a suite with integration seams. The single landscape view is the promise; the integration is the risk. A buyer should ask to see the three meters in one screen, with their own data, before believing the one-landscape story. We ask on our clients' behalf, and we have seen the seams as well as the joins.
What independent research finds about metered-technology waste
The argument of this book, that rented technology leaks money at every meter, should survive contact with sources that have no stake in a Flexera sale. The figures below come from Flexera's own research and from independent data houses; where a number is Flexera's, we say so, and the SaaS figures come from Zylo, which competes in the SaaS-management category rather than selling Flexera.
| Finding | Figure | Source |
|---|---|---|
| Organisations naming cloud-spend management a top challenge | 85% | Flexera, 18 Mar 2026 |
| Rise in wasted cloud spend: first in five years | +29% | Flexera, 18 Mar 2026 |
| SaaS licences unused against benchmarks | ~36% | Zylo, 2026 |
| SaaS spend controlled by business units, not IT | 81% | Zylo, 2026 |
| IT leaders hit by unexpected AI / consumption charges | 78% | Zylo, 2026 |
| Growth in AI-native app spend (10,000+ employees) | +393% | Zylo, 2026 |
Read across the three meters and one pattern holds: adoption has outrun governance on every one. Cloud waste is rising again after years of decline; a third of SaaS sits unused; AI spend is growing in triple digits and surprising the people who pay for it. None of these is a story about bad technology. Each is a story about a meter installed faster than the discipline to read it.
The corrective is cheap relative to the spend it governs. A FinOps and technology-spend practice is a small line against a cloud bill above five million dollars a month, and it is the line that decides whether the other line is aimed at value or at waste. Skipping the governance to save on the governance is, here as everywhere in this book, the expensive economy.
Six questions to ask before you spend on spend-management
The following six questions are the ones Volge puts to a client before recommending any platform, and they are as useful for interrogating us as for interrogating Flexera. A programme that can answer them with numbers and named owners is ready to buy. One that answers with intentions should start with a diagnostic, which costs little and prevents the expensive mistake of buying a platform to govern a bill no one has yet totalled.
Cloud, SaaS, and AI, in one number, this quarter. If the answer requires three teams and two weeks, that gap is the first thing to buy, and the reason the landscape view is the point.
Un-captured cloud discounts, the dormant SaaS third, or un-budgeted AI. Govern the biggest meter first; do not let a tool sell you effort on the smallest one.
If cloud sits with engineering, SaaS with procurement, and AI with no one, the AI meter is where the surprise will come from. Assign it before you tool it.
Outcome-based pricing is only honest if the baseline was optimised first. Ask what the saving is measured against, and who chose the baseline.
Ask to see cloud, SaaS, and AI in one screen, with your data, before believing the one-landscape story. The seams are where value leaks back out.
A spend-management platform un-operated becomes another subscription in Exhibit 4's dormant third. Decide who runs the loop, or do not buy the loop.
A programme that answers these six with numbers and names is ready to spend. One that answers with hopes should start with a diagnostic, which is small, fast, and the honest first step.
The four objections a sceptical board raises, and the honest reply to each
A well-run board does not buy on a waste statistic, and the four objections below are the ones we meet most often, each reasonable, none dismissed with a slogan. We set them out because an analysis that ducks the hard questions is not worth reading, and because the answers are, in most cases, stronger than the objection.
Very likely, and well. The question is whether that practice reaches SaaS and AI, which usually sit outside it. Cloud FinOps maturity is real and is also, in most organisations, an island. The technology tax is the whole archipelago.
It is if you buy it as a dashboard. The value is the operating loop, inform, optimise, operate, and the automation, like ProsperOps, that acts without waiting to be looked at. A dashboard no one reads is the failure mode we design engagements to avoid.
It is, and we have said so plainly. Announced in June 2026, it has no at-scale independent track record yet. We recommend it as the most complete current answer to a real problem, not as a proven one, and we would start with cloud and SaaS, where the evidence is settled, and treat AI as the fast-moving frontier it is.
You can, and the licence is the easy part. The value leaks in the work around it: connecting the billing feeds, sorting real waste from sensible spend, holding the renewal line, and turning a one-off clean-up into an operating habit. That is judgement and persistence, not a login, and it is what Volge delivers.
Each objection assumes the purchase is the software. It is not. The purchase is a governed, totalled, continuously-managed view of what your rented technology costs, and the discipline to act on it. The software is a means; the governance is the thing.
One bill, torn down: where a technology tax actually hides
A single teardown shows the method more clearly than any amount of description, so here is a composite mid-market technology bill, illustrative, drawn to match the shape governance commonly reveals rather than any one client. The point is not the totals but the pattern: how much of the tax is recoverable, and by which of the three levers.
The composite firm spends, in round numbers, 6 million US dollars a year on cloud, 4 million on SaaS, and a fast-growing 1 million on AI, eleven million on rented technology, governed by three teams who have never shared a spreadsheet.
| Meter | Annual spend | Typical recoverable | Lever |
|---|---|---|---|
| Cloud | ≈ $6.0M | Rate + waste | ProsperOps + right-sizing |
| SaaS | ≈ $4.0M | The dormant ~third | Discovery + renewal discipline |
| AI | ≈ $1.0M, rising | Un-budgeted, unseen | Visibility first, then optimise |
The cloud line yields to two levers at once: ProsperOps captures the rate, right-sizing captures the obvious waste, and only architecture reaches the rest. The SaaS line yields to discovery and the renewal calendar, the dormant third is the prize, and it is recovered by arriving at each renewal with usage data. The AI line yields to nothing yet, because it has not been totalled; the first move is visibility, and the saving comes later. We deliberately quote no single recovery percentage, because an honest one depends entirely on how un-governed the starting point was.
A vendor deck would put a number on the total saving. We will not, because the number depends on your baseline, and a large headline saving usually means the starting point was badly governed rather than the tool being brilliant. Your own diagnostic produces the only figure that matters.
Build, buy, or have it delivered: an honest comparison
There are three ways to acquire technology-spend governance, and each suits a different organisation, so we set them side by side rather than pretending only one is defensible. Build the practice yourself. Buy the platform and staff it. Or have it licensed and delivered through Volge. We sell the third, and we will still tell you when one of the others fits you better.
| Dimension | Build in-house | Buy software, self-run | Delivered through Volge |
|---|---|---|---|
| Time to first saving | Slow: months to a practice | Medium: licence then learn | Fast: diagnostic in weeks |
| Where value leaks | Reinventing FinOps | Feed connection, waste triage | Managed as part of delivery |
| Ongoing burden | You own it all | You staff and operate it | Single accountable partner |
| Best for | Firms with a mature FinOps team | Firms with spare capacity | Firms wanting the saving, not the project |
Building the practice yourself makes sense for organisations with a mature FinOps team who would rather own the whole discipline. For most, standing up cloud, SaaS, and AI governance from scratch is an expensive way to arrive where a platform already sits. We would talk you out of it unless you have a specific reason.
Buying the platform and running it yourself is the middle path, and it works when you have spare capacity and patience for the learning curve on billing feeds and waste triage. The familiar risk is that the licence gets bought, the loop never quite runs, and the platform joins the dormant third. Most of the stalled programmes we are asked to rescue started here.
Delivery through Volge suits the organisation that wants the saving rather than the project, a totalled landscape, a worked set of savings, and a running loop, with one partner accountable end to end. It is what we do, and we price the first step low enough that choosing it is a small decision, not a large one.
Glossary
Plain definitions of the terms used in this book, in the sense we have used them.
- The technology tax
- This book's term for the compounding, metered cost of rented technology, cloud, SaaS, and AI, governed, if at all, in silos.
- FinOps
- The operating practice of bringing financial accountability to variable, consumption-based technology spend. An industry discipline, not a product.
- FinOps Foundation
- The body that stewards the FinOps discipline, its certification, and the FOCUS standard. Flexera is a member.
- FOCUS
- FinOps Open Cost and Usage Specification: an open standard for normalising cloud billing data across providers.
- Flexera One
- Flexera's platform spanning IT visibility, SAM, SaaS management, and cloud/AI cost: the single landscape view.
- Cloud Cost Optimization
- Flexera's FinOps platform, descended from RightScale (2018). Informs and optimises cloud spend.
- ProsperOps
- Autonomous rate optimisation: continuously manages commitment discounts. Folded into Flexera's FinOps portfolio, 2026.
- Chaos Genius
- Data-cloud optimisation (Snowflake, Databricks). Folded into Flexera's FinOps portfolio, 2026.
- SaaS Management
- Flexera's capability for discovering, measuring, and right-sizing subscription software. Deepened by Snow (2024).
- AI Cost Management
- Flexera's platform for governing AI spend across agents, models, data platforms, and compute. Announced 12 June 2026.
- FinOps Assist
- An AI assistant inside Cloud Cost Optimization; answers cost questions from natural-language prompts. 2026.
- Effective savings rate (ESR)
- The blended discount actually achieved on cloud spend after commitments: the metric ProsperOps optimises.
- Rate optimisation
- Reducing the price paid for a given resource, via commitments and discounts. Automatable.
- Workload / architectural optimisation
- Reducing what runs, or whether it should run in that shape. Human judgement.
- Shadow SaaS
- Subscription software bought on expense, below procurement's threshold, invisible until aggregated.
- Technopedia
- Flexera's normalised catalogue of 5m+ hardware and software items: the reference behind the landscape view.
- The AI you rent
- Consumption cost of models and agents, metered by the token. The subject of this book.
- The AI you own
- Deployed AI as an asset to inventory, secure, and retire. The subject of the companion book.
Methodology & editorial standards
This book follows a small set of rules designed to keep a disclosed-interest analysis honest, and we state them so you can hold us to them. Every external figure is attributed and dated. Vendor positioning is labelled as positioning. Newly-shipped capability is labelled as unproven where it is unproven. Illustrative figures are labelled as illustrative. Our commercial interest is disclosed wherever it bears on the argument.
Where a number comes from Flexera, we say so and give the date: chiefly the 2026 State of the Cloud Report (18 March 2026) and the FinOps X announcements (12 June 2026). The SaaS and AI-adoption figures come from Zylo's 2026 SaaS Management Index, an independent SaaS-management data source that does not sell Flexera. Where we have interpolated or composited, we mark it inline.
Exhibit 9, the composite technology bill, is an illustrative composite, not a measurement of a named client. It shows the shape governance typically reveals, and we have labelled it so no reader mistakes a representative pattern for a benchmark. We deliberately quote no single headline recovery percentage, because an honest one depends entirely on how un-governed the starting point was.
We make no claim that Flexera's AI Cost Management platform is proven at scale; announced in June 2026, it has no independent at-scale track record yet, and we say so wherever it appears. We describe the one-landscape integration as a conditional benefit a buyer should verify with their own data, not as a settled fact. Corporate-history dates are best-effort from public sources.
Volge is a Flexera delivery partner and benefits if you buy. We have handled that conflict by sourcing the market figures independently, including the Zylo SaaS and AI data, which comes from a firm that competes with Flexera rather than serving it, labelling new capability as unproven, and keeping the first step small.
If any figure here has aged or any attribution is wrong, we would rather know. This is a first edition of a fast-moving subject, and the AI meter in particular will have moved by the time you read it.
Sources
Every external claim in this book, with source and date. Listed in order of first appearance.
- Flexera: 2026 State of the Cloud Report (18 March 2026): Fifteenth annual edition. 85% name cloud-spend management a top challenge; 29% rise in wasted cloud spend (first in five years); 76% of large enterprises spend >$5M/month; 63% have a FinOps team; 71% run a Cloud Centre of Excellence.
- Flexera: 2026 & prior State of the Cloud Reports: GenAI adoption 47% (2024) → 72% (2025) → 81% (2026); 45% extensive production use (up from 36% in 2025).
- Zylo: 2026 SaaS Management Index: From US$75bn tracked spend across 40m licences. Median $9,455 SaaS spend per employee; ~36% of licences unused; 81% of spend controlled by business units, 15% by IT; expense-based SaaS spend +267% YoY; ChatGPT the most expensed application; 61% forced to cut projects on unplanned SaaS cost.
- Zylo: 2026 SaaS Management Index (AI): AI-native app spend +393% YoY in organisations of 10,000+ employees; 78% of IT leaders reported unexpected charges from consumption-based or AI pricing.
- Flexera: FinOps X product announcements (12 June 2026): AI Cost Management platform spanning agents, models, data platforms, and compute; FinOps Assist assistant in Cloud Cost Optimization; ProsperOps+ outcome-based pricing; Unified Autonomous Optimization (GA).
- Flexera / ProsperOps customer figures (June 2026): Tealium paying $0.37 per $1.00 of on-demand cloud; Duolingo +5 points effective savings rate. Outcomes as reported by the vendors.
- Flexera corporate history: Spun out of Macrovision (2008); acquired RightScale (September 2018); completed acquisition of Snow Software (February 2024); folded in ProsperOps and Chaos Genius (6 January 2026). Majority-held by Thoma Bravo.
- Technopedia: Flexera's normalised catalogue of 5m+ hardware and software items.
- FinOps Foundation / FOCUS: Flexera is a member; FOCUS is the open cost-and-usage specification the discipline standardises on.
The cloud figures are Flexera's own research, labelled as such. The SaaS and AI-consumption figures are Zylo's, a firm competing in SaaS management, not a Flexera reseller, and are the closest thing to an independent check available on those meters.
The reader's checklist
Tear-out version. Six questions from the operator's read (page 19), on a single page you can take into a meeting.
| Ask | If the answer is vague |
|---|---|
| Can we total cloud, SaaS, and AI today? | Buy the landscape view first |
| Where is our largest single leak? | Govern the biggest meter first |
| Who owns each meter, by name? | The unowned meter is where surprise comes from |
| Is the saving measured against an optimised baseline? | A large headline saving means a bad baseline |
| Do the acquired pieces integrate? | Ask to see one screen, your data |
| Who operates it after signing? | Decide, or it joins the dormant third |
One accountable team, from strategy to shipped software
Volge delivers FinOps on the Flexera platform, cloud, SaaS, and AI, through one team and one arc. Four roles hold the work from the first workshop to operational handover; nothing is handed off between phases.
One FinOps approach, delivered tool-based or manual
The meter is the same whether it counts instances, seats, or tokens, and so is the discipline. The same five phases cover Cloud FinOps and AI FinOps alike; only the delivery changes.