Quick Answer
Worldwide AI spending will reach $2.59 trillion in 2026, up 47% from 2025, according to Gartner. Yet only 37% of organizations report any earnings impact from AI, McKinsey finds. These AI statistics cover what companies spend, what AI costs to run, and the ROI they're actually getting.
That gap between the two headline numbers is the story of AI in 2026. Spending is compounding at venture speed while attributable returns sit flat, and we call the distance between them the AI accountability gap: the share of AI spend no one can trace to a product, feature, customer, or outcome.
The statistics that follow measure that gap from every angle, and each section ends with what the numbers mean for the finance leaders who now own them.
The ten numbers that frame 2026
| Category | Headline stat | Source, year |
|---|---|---|
| AI spend | $2.59 trillion worldwide, +47% | Gartner, 2026 |
| AI ROI | 37% report any EBIT impact, flat YoY | McKinsey, 2026 |
| Run costs | GPT-3.5-level queries fell from $20 to $0.07 per million tokens | Stanford AI Index, 2025 |
| Cost trends | Per-token prices for a fixed capability level fall 9x to 900x per year while total spend rises 47% | Epoch AI, 2025; Gartner, 2026 |
| Adoption | 19.8% of US businesses use AI; 37% of large firms | US Census Bureau, 2026 |
| Finance teams | Finance ranks last among functions in AI deployment | CFO Connect, 2026 |
| Infrastructure | Data center electricity doubles to ~950 TWh by 2030 | IEA, 2026 |
| Waste | At least 50% of GenAI projects will overrun budgets through 2028 | Gartner, 2026 |
| Pricing | Hybrid pricing is now the top model at 37%, up from 25% | Kyle Poyar survey, 2026 |
| Market | Anthropic leads enterprise LLM spend at ~40% | Menlo Ventures, 2025 |
How much are companies spending on AI in 2026?
Companies and technology vendors will spend $2.59 trillion on AI worldwide in 2026, a 47% jump year over year, per Gartner’s May 2026 forecast. Most of that money still comes from vendors and hyperscalers rather than enterprises, which makes 2026 the year finance teams inherit the bill.
- Worldwide AI spending will total $2.59 trillion in 2026, growing 47% year over year, according to Gartner’s May 2026 forecast. That’s an upward revision from the $2.52 trillion the firm projected in January 2026.
- AI infrastructure will account for over 45% of all AI spending through the next several years, making it the largest segment of the market, per Gartner (2026). That includes AI-optimized servers, network fabric, semiconductors, and infrastructure as a service.
- Gartner analyst John-David Lovelock calls the AI buildout “the largest infrastructure project ever attempted by humanity” in the firm’s July 2026 update. Worldwide IT spending overall will hit $6.37 trillion in 2026, up 14.2%, with AI the primary driver.
- Enterprises haven’t driven the spending wave yet. Gartner’s January 2026 analysis notes AI spending has primarily come from technology companies and hyperscalers, with 2026 positioned as the enterprise inflection year.
- Vendor buildout alone adds $401 billion in AI infrastructure spending in 2026 as technology providers construct AI foundations, per Gartner (January 2026).
- Global corporate AI investment reached $581.7 billion in 2025, a 130% increase year over year, according to the Stanford AI Index 2026. Generative AI investment alone hit $170.9 billion, up 404%.
- US private AI investment reached $285.9 billion in 2025, more than 23 times China’s $12.4 billion in tracked private investment, per Stanford (2026). The report cautions this understates China’s total, given state guidance funds.
- AI infrastructure spending will reach $497 billion in 2026, up roughly 56% year over year, according to IDC’s Worldwide AI Infrastructure Tracker (July 2026), an acceleration on the 53% IDC projected a quarter earlier. IDC expects the market to reach $1.08 trillion in 2029.
- AI infrastructure spending hit roughly $90 billion in Q4 2025 alone, the largest quarter ever recorded for the segment, per IDC (2026).
- Asia Pacific AI and generative AI spending is projected to reach $370 billion by 2029, growing fivefold as enterprises shift from experimentation to scaled deployment, per IDC (2026).
- The four largest cloud and AI providers, Amazon, Microsoft, Alphabet, and Meta, have guided to roughly $700 billion or more in combined 2026 capital expenditures, up from about $410 billion in 2025, per company guidance through Q2 2026 earnings. Amazon and Alphabet are neck and neck at roughly $200 billion each, with Alphabet’s guidance range topping out $5 billion higher.
- Alphabet raised its 2026 capex ceiling to $205 billion at Q2 2026 earnings, and Meta has raised its guidance twice, to a range of $130 billion to $145 billion, citing higher component prices among other drivers (company earnings calls, 2026).
- Goldman Sachs projects $5.3 trillion in combined capital spending from the four largest hyperscalers between fiscal 2025 and fiscal 2030, up from a prior $4.5 trillion estimate (reported June 2026).
- McKinsey estimates cumulative capital expenditure on data centers could reach $6.7 trillion through 2030, of which about $5.2 trillion is AI-specific (reported April 2025).
- Enterprise spending on generative AI reached $37 billion in 2025, a 3.2x increase from $11.5 billion in 2024 and up from just $1.7 billion in 2023, according to Menlo Ventures’ State of Generative AI in the Enterprise report (December 2025).
- AI applications captured more than half of enterprise AI investment in 2025, about $19 billion, and now represent 6% of the entire software market. Menlo Ventures calls it the fastest growth of any category in software history (2025).
- Global business leaders plan to invest a weighted average of $186 million each in AI over the next twelve months, per KPMG’s first Global AI Pulse survey of 2,110 senior leaders across 20 countries (March 2026). Three quarters of respondents run organizations with revenue above $1 billion.
- Planned AI investment is highest in Asia Pacific at $245 million per organization on average, versus $178 million in the Americas and $157 million in EMEA, per KPMG (2026).
- US organizations plan average AI spending of $207 million over the next twelve months, nearly double the figure from the same period a year earlier, per KPMG’s US Q1 AI Quarterly Pulse (2026).
- 74% of leaders say AI will remain a top investment priority even in the event of a recession, per KPMG (2026).
- End-user spending on AI models and platforms will total $64 billion in 2026, up 63.4% from $39 billion in 2025, with generative AI model spending growing 117%, according to Gartner’s July 2026 forecast.
- AI-optimized infrastructure as a service spending will grow 96% in 2026 to reach $42 billion, then $66 billion in 2027, per Gartner’s August 2026 forecast.
AI spending in 2026 at a glance
| Measure | Figure | Source, year |
|---|---|---|
| Worldwide AI spending, full stack | $2.59 trillion, +47% | Gartner, 2026 |
| Global corporate AI investment (2025) | $581.7 billion, +130% | Stanford AI Index, 2026 |
| AI infrastructure spending | $497 billion, +56% | IDC, 2026 |
| Big four hyperscaler capex, combined | ~$700 billion+ | Company guidance, Q2 2026 |
| Enterprise generative AI spend (2025) | $37 billion, 3.2x YoY | Menlo Ventures, 2025 |
| Planned per-company AI investment, next 12 months | $186 million average | KPMG, 2026 |
| AI models and platforms, end-user spend | $64 billion, +63.4% | Gartner, 2026 |
One caution when you use these numbers: they aren’t additive. Gartner’s $2.59 trillion counts the full stack including vendor buildout. IDC’s $497 billion counts infrastructure only. Stanford’s $581.7 billion counts corporate investment, and Menlo’s $37 billion counts enterprise generative AI purchases. Each answers a different question, and mixing them is the most common error in AI budget decks.
What this means for finance leaders
The money has moved faster than the accountability. Vendors and hyperscalers drove the first $2 trillion of this wave, but 2026 is when enterprise budgets take over, and Gartner says CIOs already struggle to prove value from what they’ve committed.
The accountability gap starts here: a nine-figure line item, KPMG’s $186 million average, is landing in budgets that were never built to trace AI spend to outcomes. The companies that close that gap first set the ROI bar everyone else gets measured against.
Report: Finance needs to prove AI’s return
260 senior finance leaders (more than half CFOs) told us why the speed of seeing AI spend, not the size of it, separates who pulls ahead on AI from who gets burned.
Report
Finance needs to prove AI’s return: CloudZero report
260 senior finance leaders (more than half CFOs) told us why the speed of seeing AI spend, not the size of it, separates who pulls ahead on AI from who gets burned.
What ROI are companies seeing from AI?
Most companies aren’t seeing measurable AI ROI yet. Only 37% of organizations attribute any earnings impact to AI, a figure unchanged from 2025, per McKinsey’s 2026 global survey.
Meanwhile 80% of individual users report productivity gains. That gap between personal productivity and enterprise returns defines AI economics in 2026.
- 37% of organizations attribute at least some EBIT impact to AI use, essentially unchanged from a year earlier, according to McKinsey’s State of AI 2026 survey of 1,719 leaders across 97 countries (August 2026).
- Only 6% of organizations qualify as AI high performers, meaning they attribute at least 5% of EBIT to AI and describe its impact as significant. That share is also flat year over year despite rising investment, per McKinsey (2026).
- 80% of respondents say AI has improved their individual productivity, and 50% say it helps them make better decisions, per McKinsey (2026). The enterprise P&L hasn’t followed: individual gains are running roughly two times ahead of reported earnings impact.
- 89% of organizations now use AI in at least one business function, and 44% have scaled it enterprise-wide, per McKinsey (2026). Adoption is no longer the bottleneck. Value capture is.
- About one in five organizations say AI-related operating costs are beginning to constrain their AI use, even as most plan to increase AI investment in the year ahead, per McKinsey (2026).
- McKinsey’s high performers look different on discipline, not just budget: nearly three quarters fundamentally redesigned workflows around AI versus one quarter of everyone else, and they’re twice as likely to have defined processes for measuring AI initiatives’ impact (2026).
- Where financial impact does show up, it’s function specific: respondents most often report AI cost reductions in supply chain management, service operations, and manufacturing, per McKinsey (2026).
- Revenue gains from AI concentrate in marketing and sales, followed by product and service development and software engineering, per McKinsey (2026). Cost impact and revenue impact live in different corners of the business.
- 95% of enterprise generative AI pilots delivered little or no measurable P&L return despite $30 billion to $40 billion in enterprise investment, according to MIT’s Project NANDA report, The GenAI Divide (2025). The research drew on 300 public AI initiatives, interviews with about 150 leaders, and a survey of 350 employees.
- The roughly 5% of pilots that succeed share a pattern, per the report’s lead author Aditya Challapally: they “pick one pain point, execute well, and partner smartly” with the companies that use their tools (Fortune, 2025).
- 42% of companies abandoned most of their AI initiatives in 2025, up sharply from 17% a year earlier, according to S&P Global Market Intelligence’s Voice of the Enterprise survey (2025).
- More than 80% of AI projects fail to deliver their intended business value, roughly twice the failure rate of comparable IT projects that don’t involve AI, according to RAND (2024).
- Only 11% of organizations have reached the stage of deploying and scaling AI agents in ways that produce enterprise-wide business outcomes, per KPMG’s Global AI Pulse (2026). Investment is broad, but realized value at scale remains rare.
- 64% of organizations say AI is already delivering meaningful business outcomes, per KPMG (2026). The distance between that self-report and McKinsey’s 37% EBIT figure is itself instructive: felt value is running well ahead of measured value.
- The demand side still argues boom over bubble: at least 10 AI products now generate over $1 billion in annual recurring revenue, and more than 50 have crossed $100 million, per Menlo Ventures (2025).
- AI-driven workforce reductions ran well below expectations. 14% of organizations reported AI contributed to a decline in workforce size over the past year, less than half the 32% that had anticipated one, per McKinsey (2026).
- 91% of organizations plan to increase AI investment in the next year, according to Deloitte’s AI ROI study (October 2025), and 94% plan to keep investing even if AI doesn’t drive immediate returns, per BCG’s AI Radar 2026 survey of 2,360 executives.
AI ROI in 2026 at a glance
| Measure | Figure | Source, year |
|---|---|---|
| Organizations reporting any EBIT impact from AI | 37%, flat YoY | McKinsey, 2026 |
| AI high performers (5%+ of EBIT from AI) | 6% | McKinsey, 2026 |
| Users reporting individual productivity gains | 80% | McKinsey, 2026 |
| GenAI pilots with little or no measurable P&L return | 95% | MIT NANDA, 2025 |
| Companies abandoning most AI initiatives | 42%, up from 17% | S&P Global, 2025 |
| Organizations scaling AI agents to enterprise outcomes | 11% | KPMG, 2026 |
| Companies investing regardless of immediate returns | 94% | BCG, 2026 |
What this means for finance leaders
Read these numbers together and the pattern is hard to miss. Adoption is nearly universal, individual productivity is real, and enterprise returns are flat. The difference between the 6% of high performers and everyone else isn’t model choice or budget size. It’s measurement discipline: the trait McKinsey found twice as often among high performers.
That’s an unusually fixable problem. You can’t redesign workflows overnight, but you can start attributing AI spend to the products, features, and teams that consume it, which is the precondition for every ROI conversation that follows. The 42% of companies abandoning AI initiatives aren’t necessarily abandoning bad ideas. Many are abandoning ideas they never instrumented well enough to defend.
One honest caveat belongs here, and it’ll build more trust than hiding it: the famous 95% figure measures early pilots from 2025, and its methodology drew criticism for how it defined failure. It’s a real signal about unmeasured deployments, not proof that AI can’t pay. McKinsey’s flat 37% is the more durable benchmark.
How much does it cost to run AI in 2026?
Running AI in 2026 costs far less per token and far more in total than most budgets assumed. GPT-3.5-level output that cost $20 per million tokens in late 2022 now costs about $0.07, per Stanford’s AI Index. What’s grown is volume: inference, not training, now dominates production AI bills.
- The price of querying an AI model at GPT-3.5-level quality fell from $20 to $0.07 per million tokens in roughly 18 months, a 280-fold drop, according to the Stanford AI Index (2025).
- The median on-demand price for an Nvidia H100 sits at about $3.27 per GPU hour across 53 providers, per pricing tracker getdeploying (September 2026).
- A single AWS p4d.24xlarge GPU instance costs about $21.96 per hour on demand in US East (N. Virginia), per AWS pricing (September 2026). An idle cluster left running over a long weekend can burn five figures before anyone notices.
- OpenAI’s inference bill reached roughly $8.4 billion in 2025, about four times the prior year, as reported by The Information (2026). Even the model makers feel inference economics at scale, which shows up in OpenAI’s own pricing structure.
- Nvidia’s GB300 NVL72 system reached $0.12 per million tokens in the company’s InferenceMAX benchmarks, which Nvidia describes as 35 times lower cost per token than its Hopper generation (2025).
- Inference accounts for an estimated 70% to 80% of total GPU cloud spend for teams running AI in production, per analyst estimates compiled in 2026. Training is a project. Inference is a run rate.
- 2026 is the year inference spending overtakes training in AI-optimized cloud infrastructure: $23.3 billion versus $19 billion, with inference reaching 55% of AI IaaS spending this year and 59% in 2027, per Gartner (2026).
- Google processed 3.2 quadrillion tokens per month by mid-2026, roughly seven times its year-earlier volume, per Google disclosures (2026). Falling unit prices have accelerated consumption rather than slowing it.
- Production-ready generative AI systems can cost orders of magnitude more than the pilots that preceded them, Gartner warns in its 2026 cost optimization research. Teams budgeting from pilot economics get, in Gartner’s words, a rude awakening.
- Consumer and team-tier AI subscriptions have become a visible budget line of their own: ChatGPT’s paid tiers and Claude’s plans and token rates now sit alongside per-token API costs in most enterprise AI bills.
- Coding and developer tools captured $7.3 billion of enterprise AI application spending in 2025, alongside $8.4 billion for general-purpose copilots and $3.5 billion for industry-specific AI, per Menlo Ventures (2025). Codex-class coding tools carry their own pricing structures.
- AI is starting to displace software purchases entirely: 32% of organizations skipped buying at least one software product because agentic coding tools let them build the capability in house, per McKinsey (2026). Build costs are becoming a run-cost line of their own.
- Structural costs hide beneath visible token prices: provisioned model endpoints bill around the clock whether used or not, cross-region data movement adds egress fees, and managed retrieval pipelines carry high minimum monthly storage fees (industry analyses, 2026).
What this means for finance leaders
Per-unit prices are collapsing while bills grow, and both facts are true because the unit of consumption changed: from a seat you could count once a year to tokens that scale with every feature, retry, and agent run.
When a single feature can silently multiply its token volume by ten, AI spend behaves like cloud spend did a decade ago, before anyone instrumented it. The fix is the same one cloud went through: measure consumption at the unit level, per feature and per customer, before negotiating a single discount.
How are AI costs changing?
AI costs are moving in two directions at once. The price of a fixed level of AI capability is collapsing, by 9x to 900x per year depending on the benchmark, per Epoch AI. Total AI spending is still rising 47% this year. Cheaper units are driving more consumption, not smaller bills.
- The price of reaching a fixed AI capability level falls between 9x and 900x per year depending on the benchmark, with GPT-4-level performance on PhD-level science questions getting 40x cheaper per year, according to Epoch AI’s LLM inference price analysis (2025).
- The price of GPT-3-quality output fell roughly 1,000-fold in three years, from $60 per million tokens in late 2021 to $0.06 by late 2024, a trend a16z calls LLMflation (2024).
- Gartner raised its worldwide IT spending forecast three times in 2026 alone, from $6.15 trillion in February to $6.31 trillion in April to $6.37 trillion in July, each revision driven by AI infrastructure demand (Gartner, 2026).
- Companies plan to double their AI spending in 2026, reaching about 1.7% of revenues, more than twice the increase they made in 2025, per BCG’s AI Radar 2026 survey.
- 90% of CEOs believe AI agents will enable their organizations to report measurable ROI in 2026, and CEOs have committed more than 30% of 2026 AI investment to agentic AI, per BCG (2026). Costs are shifting toward the most consumption-heavy workload class there is.
- Electricity use in accelerated servers, the hardware class that runs AI, is projected to grow 30% annually through 2030, more than three times the pace of conventional servers, per the IEA (2026).
- Generative AI model spending will grow 117% in 2026, the fastest-growing line inside the AI platforms market, per Gartner (2026).
- Gartner placed AI in the trough of disillusionment for 2026, predicting AI will most often be sold to enterprises by incumbent software providers rather than bought as moonshot projects, and that ROI predictability must improve before enterprises truly scale up (2026).
- Spending on AI-optimized servers will rise 49% in 2026, representing 17% of total AI spending, as technology providers build out AI foundations, per Gartner (January 2026).
- Software spending overall will grow 14.7% in 2026, per Gartner (February 2026). AI line items are growing three to eight times faster than the software budgets that contain them.
- Agentic AI amplifies the trend: Gartner attributes rising inference intensity to multistep autonomous execution, which makes each agent run consume far more tokens than a single prompt (2026).
What this means for finance leaders
This is a classic efficiency paradox: as the unit gets cheaper, total consumption grows faster than the price falls. Budgeting for AI by extrapolating unit prices will consistently underestimate spend, because volume, not price, is the variable that moves.
The forecasting discipline that matters in 2026 is consumption forecasting: how many tokens, agent runs, and GPU hours your roadmap implies, priced at falling but nonzero rates. Teams that built unit economics muscle on cloud spend already have the right mental model.
Who is adopting AI, and where?
About one in five US businesses uses AI in producing goods or services, per the Census Bureau’s biweekly survey, but adoption splits sharply by size: 37% of firms with 250 or more employees use AI versus under 20% of the smallest firms. Weighted by employment, coverage roughly doubles.
- 19.8% of US businesses reported using AI in a business function as of May 3, 2026, per the US Census Bureau’s Business Trends and Outlook Survey. Usage hovered between 17% and 20% from December 2025 through May 2026.
- 37% of firms with at least 250 employees use AI, versus 32% of firms with 100 to 249 employees and under 20% of firms with four or fewer employees, per Census BTOS (May 2026).
- Weighted by employment rather than firm count, AI use roughly doubles: 18% of firms used AI during the November 2025 to January 2026 reference period, rising to 32% on an employment-weighted basis, per the Census Bureau’s 2026 AI supplement.
- The Information sector leads adoption at 39.7%, double the 19.8% national rate, with Finance and Insurance second at 33.9%. Retail trails at about 14%, per Census BTOS (2026).
- Between 20% and 23% of US businesses expect to be using AI within six months, consistently a few points above current use, per Census BTOS (2026).
- At the start of 2025, only about 9% of firms planned to use AI in the following six months, per a Federal Reserve analysis (2026). Expected adoption has more than doubled in under two years.
- Before a late-2025 methodology change, the Census-measured AI adoption rate had grown 68% in a single year, per the same Federal Reserve analysis of adoption surveys (April 2026).
- 41% of working individuals report using generative AI for work as of November 2025, per the Real-Time Population Survey cited in the same Federal Reserve note (2026).
- 78% of the US labor force works at firms that have adopted AI, and about 54% works at firms using large language models, per the Atlanta Fed’s Survey of Business Uncertainty (November 2025).
- Definitions move the number dramatically: the Census measures AI used in producing goods or services and lands at 17% to 20%, while broad industry surveys counting any regular tool use put small business adoption at 68% to 77% (2026 survey comparisons). Always check what an adoption stat counts.
- Generative AI reached 53% global population adoption within three years of ChatGPT’s launch, faster than the personal computer or the internet, per the Stanford AI Index (2026). Adoption correlates strongly with GDP per capita, with Singapore at 61% and the UAE at 64%.
- 40% of companies with over $1 billion in revenue are scaling AI agents, up sharply from 27% a year earlier, per McKinsey (2026).
- The US led entrepreneurial activity with 1,953 newly funded AI companies in 2025, more than ten times the next closest country, per Stanford (2026).
- Among adopting businesses, 57% apply AI in three or fewer operational functions, per Census BTOS analyses (2026). Most adopters are still running AI in a narrow slice of operations.
AI adoption in 2026 at a glance
| Measure | Figure | Source, year |
|---|---|---|
| US businesses using AI | 19.8% | Census BTOS, May 2026 |
| Firms with 250+ employees using AI | 37% | Census BTOS, 2026 |
| Employment-weighted firm adoption | 32% | Census AI supplement, 2026 |
| Labor force at AI-adopting firms | 78% | Atlanta Fed SBU, 2025 |
| Global population using generative AI | 53% | Stanford AI Index, 2026 |
| $1B+ companies scaling AI agents | 40%, up from 27% | McKinsey, 2026 |
What this means for finance leaders
The adoption gap is a size gap, and it compounds. Large firms adopt at nearly twice the national rate, which means AI’s advantages are concentrating among companies that already had scale advantages.
For finance teams at those companies, the question has moved from whether to adopt to whether you can account for what adoption produces per unit of output. The 57% of adopters running AI in three or fewer functions are about to expand, and expansion without visibility is how the waste numbers three sections down happen.
How are finance teams adopting AI?
Finance is adopting AI fast and deploying it slowly. 56% of finance leaders now use AI tools in daily work, more than triple the 17% of 2023, yet finance still ranks last among all business functions in actual AI deployment, per CFO Connect’s State of AI in Finance 2026.
- 56% of finance leaders now use AI-powered tools in their daily work, up from 17% in 2023, per CFO Connect’s State of AI in Finance 2026 report.
- Finance and accounting ranks last among all major business functions in AI deployment, trailing engineering, marketing, and sales, per a General Atlantic survey cited in the same report (2025).
- 45% of finance teams remain in limited pilot mode, while only 17% actively use AI in their core workflows, per the General Atlantic poll (2025). The tinkerer-to-integrator gap defines finance AI in 2026.
- 59% of finance functions now use AI, barely up from 58% a year earlier, per Gartner’s survey of 183 CFOs and senior finance leaders (2025). By Gartner’s measure, finance adoption has plateaued.
- Gartner separately projects that 90% of finance teams will deploy at least one AI-enabled solution by 2026, up from 58% of finance professionals reporting AI use in 2024 and 37% in 2023.
- 87% of CFOs at large companies say AI will be extremely or very important to finance operations in 2026, per Deloitte’s Q4 2025 CFO Signals survey.
- 68% of CFOs are increasing IT and digital transformation spending, the highest level in the 21 quarters the survey has run, per Grant Thornton’s Q1 2026 CFO survey.
- Fewer than half of finance leaders, 47%, identify AI as the global trend expected to have the greatest impact on their organizations this year, per Wolters Kluwer’s 2026 Future Ready CFO report of over 1,600 senior finance executives across 20 markets.
- 43% of finance leaders say AI adoption now influences capital allocation and resource planning, directing funding toward analytics tools, automation platforms, and supporting data infrastructure, per Wolters Kluwer (2026).
- 79% of FP&A teams have adopted AI tools to some degree, though most deployments produce quick operational wins like spreadsheet automation and report polishing rather than transformed planning, per a Drivetrain survey of 258 FP&A professionals (2025).
- Upskilling lags adoption badly: 68% of FP&A professionals spent five or fewer hours on AI upskilling in the prior month, and 15% spent none, per Drivetrain (2025).
- 88% of senior finance leaders believe AI is the most transformative trend of the next 12 to 24 months, yet only 8% feel very well prepared for it, per AICPA and CIMA’s survey of 1,446 senior finance leaders (2025).
- 17% of finance teams already deploy generative AI agents, and 75% of finance leaders expect agentic AI to be a routine part of finance operations by 2028 (industry surveys, 2026).
- 98% of FinOps practitioners now manage AI spend as part of their practice, up from 63% in 2025 and just 31% in 2024, per the FinOps Foundation’s State of FinOps 2026 report. Managing AI cost has gone from a specialty to table stakes in two years.
- 68% of CFOs say they’ve been slow to adopt AI because they don’t know where to start, per CFO Connect (2026).
- ChatGPT is the most used AI tool in finance, adopted by 35% of finance teams, well ahead of specialized finance AI tools, per CFO Connect (2026).
Finance team AI adoption at a glance
| Measure | Figure | Source, year |
|---|---|---|
| Finance leaders using AI tools daily | 56%, up from 17% in 2023 | CFO Connect, 2026 |
| Finance’s rank among functions in AI deployment | Last | General Atlantic, 2025 |
| Finance teams using AI in core workflows | 17% | General Atlantic, 2025 |
| CFOs calling AI very important to 2026 operations | 87% | Deloitte, 2025 |
| Finance leaders feeling very well prepared for AI | 8% | AICPA and CIMA, 2025 |
| FP&A teams with some AI adoption | 79% | Drivetrain, 2025 |
What this means for finance leaders
There’s an irony in this data that should shape strategy: the function that will be asked to prove AI’s ROI is the function slowest to operationalize AI itself. Finance teams that stay in tinkerer mode inherit two problems at once, their own productivity gap and an unanswerable board question about everyone else’s AI value.
The way out of both is the same skill: treating AI as a measurable cost object with owners, budgets, and unit metrics, the discipline finance already applies to every other major cost line. The 8% who feel prepared aren’t better technologists. They started measuring earlier.
What do AI infrastructure and energy statistics show?
AI’s physical footprint is now a line item in national energy planning. Global data center electricity use will roughly double from 485 TWh in 2025 to about 950 TWh by 2030, slightly more than Japan’s total consumption today, per the IEA. AI is the primary driver.
- Global data center electricity consumption is projected to roughly double from 485 TWh in 2025 to about 950 TWh by 2030, reaching around 3% of global electricity demand, per the IEA’s 2026 Energy and AI update.
- At roughly 945 TWh, that 2030 total is slightly more than Japan’s entire electricity consumption today, per the IEA (2026).
- The IEA’s base case sees data center electricity demand continuing to about 1,200 TWh by 2035, with uncertainty widening after 2030 (2026).
- Electricity consumption from AI-focused data centers specifically is set to triple between 2025 and 2030, growing much faster than data centers overall, per the IEA’s 2026 update.
- In the United States, data centers account for nearly half of all electricity demand growth between now and 2030. By decade’s end, the US will use more electricity for data centers than for producing aluminum, steel, cement, chemicals, and all other energy-intensive goods combined, per the IEA (2026).
- US data center energy demand is projected to increase 130% by 2030, per IEA estimates synthesized by Brookings (2026).
- US data centers consumed about 4.4% of total US electricity in 2023 and are projected to reach between 6.7% and 12% by 2028, rising from 176 TWh to between 325 and 580 TWh, per Lawrence Berkeley National Laboratory for the US Department of Energy (2024).
- AI data center power capacity reached 29.6 gigawatts in 2025, roughly comparable to the peak electricity demand of New York State, per the Stanford AI Index (2026).
- Former Google CEO Eric Schmidt testified before Congress that data centers will need 29 gigawatts of additional power by 2027 and 67 more gigawatts by 2030 (2025, via Brookings).
- Renewables are projected to meet half of global data center demand growth, expanding by over 450 TWh through 2035, with natural gas adding 175 TWh, notably in the US, per the IEA (2026).
- Total worldwide data center spending will surpass $650 billion in 2026, up 31.7% from nearly $500 billion the prior year, per Gartner’s February 2026 forecast.
- Data center systems spending will grow 55.8% in 2026, the largest growth of any IT segment, per Gartner’s April 2026 forecast.
- Early estimates put a ChatGPT request at about 2.9 watt-hours versus roughly 0.3 watt-hours for a typical Google search, per IEA analysis (2024). Direct measurement has since collapsed that gap: Google reported a median Gemini text prompt at 0.24 watt-hours in 2025, down 33x in a year.
- Training a single frontier model now carries an industrial-scale footprint: Grok 4’s training emitted an estimated 72,816 tonnes of CO2 equivalent, per the Stanford AI Index (2026).
- The US accounted for roughly 45% of global data center electricity consumption in 2024, making it the world’s largest data center market, per Brookings’ synthesis of IEA and LBNL data (2026).
- Cloud remains the delivery vehicle: data center systems and infrastructure as a service are the two fastest-growing IT segments in 2026 per Gartner (July 2026), and hyperscaler AI platforms, including Azure’s OpenAI services, keep absorbing enterprise workloads while conventional server electricity use grows just 9% annually against 30% for accelerated servers (IEA, 2026).
What this means for finance leaders
Energy is becoming a cost input finance teams can no longer treat as someone else’s problem. Power constraints already shape where capacity gets built and what it costs, and those costs pass through to cloud and API prices. When your providers’ input costs are doubling on a five-year horizon, long-term AI budgeting needs a power-price assumption in it, the same way cloud cost planning learned to carry region and instance-mix assumptions.
How much AI spending is wasted?
A large share of AI spending buys nothing. Gartner projects at least 50% of generative AI projects will overrun their budgets through 2028, and roughly 29% of all cloud spend is wasted outright, per Flexera’s annual estimate. Idle GPUs are the fastest-growing waste category.
- At least 50% of generative AI projects will overrun their budgeted costs through 2028 due to poor architectural choices and lack of operational know-how, per Gartner’s cost optimization research (2026).
- 30% of generative AI projects were expected to be abandoned after proof of concept by the end of 2025, per Gartner (2024).
- Over 40% of agentic AI projects will be canceled by the end of 2027 due to escalating costs, unclear business value, and inadequate risk controls, Gartner projects (2025).
- An estimated 29% of cloud spend is wasted, up for the first time in five years and reversing a five-year downward trend, per Flexera’s 2026 State of the Cloud report of 753 cloud decision-makers. Flexera attributes the reversal to surging cloud-based AI workloads.
- Average GPU utilization sits at just 23%, meaning about three quarters of provisioned GPU capacity goes unused, with idle GPU time the fastest-growing cloud waste category, per Harness (2025).
- Fewer than 20% of organizations have automatic shutdown policies for GPU instances, per a CNCF survey (2024).
- Only 22% of companies have advanced beyond the proof-of-concept stage with AI, and only 4% are creating substantial value, per BCG’s research on the AI value gap (2025).
- None of the roughly 30 generative AI technologies Gartner tracks in its Hype Cycle for Generative AI has reached the plateau of productivity, per the firm’s 2026 analysis (reported by The Register).
- Organizations scrapped nearly half of their AI proofs of concept before they reached production, per S&P Global Market Intelligence (2025).
- The barriers behind the waste are worsening, not easing: 65% of organizations cite difficulty scaling use cases as a top obstacle to demonstrating AI ROI, up from 33% a quarter earlier, per KPMG’s US Q1 Pulse (2026).
- 62% point to skills gaps as a top ROI barrier, up from 25% the prior quarter, per KPMG (2026). The constraint has shifted from technology to operating capability.
- Most IT leaders exceeded their cloud budgets in 2023, with only about one third avoiding overruns through careful budgeting, monitoring, and resource optimization, per a Gartner Peer Community survey of 200 IT leaders. AI is inheriting cloud’s overrun habits at larger scale.
- Reported AI spending runs roughly 12x lower than actual AI-driven cloud consumption, per CloudZero’s ROI in the AI Era research. Most AI waste isn’t overspending on a known line item. It’s spending that never appears as AI at all.
- Organizations budget 30% to 36% of cloud spend for AI, yet AI-specific line items show up at just 2.5% of bills, per the same CloudZero research. The gap between those numbers is where unowned GPU hours and untagged inference live.
- 57% of organizations still track AI costs in spreadsheets, and 15% have no formal AI cost-tracking system at all, per CloudZero’s State of AI Costs report (2025).
- Only 51% of organizations strongly agree they can accurately track AI ROI, even though 91% feel confident in their ability to evaluate it, per CloudZero’s State of AI Costs (2025). Confidence is running 40 points ahead of capability.
- The mean Cloud Efficiency Rate dropped from 80% to 65% even as formal cloud cost programs nearly doubled, from 39% to 72% of organizations, per CloudZero’s ROI in the AI Era report of 475 senior leaders (2026). Maturity is rising while efficiency falls, and AI workloads are the wedge between them.
AI waste in 2026 at a glance
| Measure | Figure | Source, year |
|---|---|---|
| GenAI projects overrunning budgets through 2028 | 50%+ | Gartner, 2026 |
| Cloud spend wasted | 29% | Flexera, 2026 |
| Average GPU utilization | 23% | Harness, 2025 |
| GenAI projects abandoned after POC | 30% | Gartner, 2024 |
| Agentic AI projects canceled by 2027 | 40%+ | Gartner, 2025 |
| Companies beyond POC with AI | 22% | BCG, 2025 |
What this means for finance leaders
Waste at this scale isn’t a tooling gap. It’s a visibility gap, the accountability gap’s most expensive symptom. A GPU cluster idling at 23% utilization survives because nobody owns the number, and a project that’s 3x over budget survives because the budget never had unit-level checkpoints.
The pattern from cloud cost savings data repeats here: waste persists exactly as long as it stays unattributed. Assign every GPU hour and every token to a team, product, or customer, and the 29% number starts falling on its own.
How are AI pricing models changing?
AI is rewriting how software gets priced. Hybrid pricing, a base subscription plus usage-based charges, is now the most common model at 37% of B2B software companies, up from 25% a year ago, per Kyle Poyar’s 2026 monetization survey. Pure seat-based pricing is collapsing.
- Hybrid pricing is now the most common primary pricing structure among B2B software companies at 37%, up from 25% twelve months earlier, per Kyle Poyar’s 2026 State of B2B Monetization survey of over 230 software and AI companies.
- Only 5% of investors prefer seat-based pricing and 10% prefer flat-fee subscriptions. Investors instead favor hybrid (35%), outcome-based (26%), and usage-based (24%) models, per the same survey (2026).
- 97% of SaaS CEOs plan to retire seat-based pricing within two years, even though 94% say it currently aligns with their product’s value, per a Cruxy survey of 300 SaaS CEOs (April 2026).
- The top 500 B2B and AI companies made more than 1,800 pricing and packaging changes in 2025 alone, an average of 3.6 changes per company, per the PricingSaaS 500 Index (2026).
- Credit-based models became the defining pricing innovation of 2025: companies using them grew 126% year over year, from 35 to 79 among the top 500, per PricingSaaS (2026).
- Outcome-based pricing has crossed from theory to production: Fin charges $0.99 per resolved outcome, with lead qualification priced ten times higher at $9.99, and Salesforce’s Agentforce prices actions in Flex Credits at roughly $0.10 each, per company pricing (September 2026).
- Seat-based pricing fell from 21% to 15% of SaaS companies in 12 months, per Kyle Poyar’s 2025 State of B2B Monetization survey, the edition his 2026 survey has since superseded.
- Gartner estimates up to $234 billion in enterprise application software spend is exposed to agentic arbitrage between now and 2030, which would account for roughly 20% of enterprise application SaaS spending, per Gartner (July 2026).
- Consumption-priced vendors are outgrowing seat-based peers: Snowflake grew product revenue 34% and Datadog 32% year over year in early 2026, both more than double the roughly 14% median for public SaaS companies (company earnings, 2026).
- Usage-priced AI products can scale revenue at unprecedented speed: Cursor grew from roughly $2 billion in annualized revenue in February 2026 to about $4 billion by June, per industry reporting (2026).
- AI cost structures broke the old model’s economics: per-seat software has near-zero marginal cost per user, while every AI interaction carries real inference cost, which is why vendors are pushing pricing toward usage (industry analyses, 2026). The unit economics logic SaaS companies built now runs in reverse: the vendor’s cost of goods scales with your usage.
- Agentic AI pressures seat pricing from the demand side too: when agents do work people used to do, seat counts fall even as workload grows, breaking the link between headcount and software value (2026 pricing analyses).
What this means for finance leaders
Every one of these pricing shifts moves risk from vendor to buyer. A seat was a fixed cost you could budget in January. A hybrid or usage contract is a variable cost that scales with adoption you can’t fully predict.
Procurement teams that negotiated discounts now need to negotiate meters: what counts as usage, where the caps sit, and what visibility the vendor provides. Buying AI well in 2026 looks less like licensing and more like managing cloud commitments, because structurally, that’s what it is.
What do AI vendor, market, and leadership statistics show?
The AI vendor market re-ranked itself in under three years. Anthropic now leads enterprise LLM spend at roughly 40%, ahead of OpenAI at 27%, down from 50% in 2023, per Menlo Ventures. Meanwhile CEOs have taken personal ownership of AI decisions.
- Anthropic holds approximately 40% of enterprise LLM spend, the first time it has led OpenAI, which sits at 27%, down from 50% in 2023, per Menlo Ventures’ enterprise survey (December 2025).
- Google’s Gemini grew its enterprise LLM spend share from 7% to 21% over the same period, per Menlo Ventures (2025). Provider share is proving far less sticky than the underlying category growth.
- Healthcare captured $1.5 billion of vertical AI spending in 2025, 43% of the vertical market and more than the next four industries combined, per Menlo Ventures (2025).
- Industry-specific AI solutions attracted about $3.5 billion in enterprise investment in 2025, nearly tripling the prior year, per Menlo Ventures (2025).
- 72% of CEOs say they are now the main decision maker on AI in their organization, twice the level of the previous year, per BCG’s AI Radar 2026 survey of 2,360 executives including 640 CEOs.
- Half of CEOs believe their job stability depends on getting AI investments and strategy right, per BCG (2026). AI has become an executive survival question, not a technology preference.
- 82% of CEOs are more optimistic about AI’s ability to deliver ROI than they were twelve months earlier, per BCG (2026). Confidence is rising even as measured returns stay flat, a spread finance teams will be asked to explain.
- CEO conviction splits into three archetypes: roughly 15% Trailblazers driving transformation through decisive investment, 70% Pragmatists who invest only where value is evident and risk is low, and 15% Followers making cautious early bets, per BCG (2026).
- Trailblazing CEOs allocate 60% of their AI budgets to upskilling and retraining their current workforce, versus 27% for Pragmatists and 24% for Followers, per BCG (2026).
- Agentic AI already drives 17% of total AI value and is expected to nearly double its share by 2028, per BCG (2025).
- 70% of AI’s potential value concentrates in core business functions such as sales and marketing, manufacturing, supply chain, and pricing, per BCG (2025).
- AI leaders already generate 1.7 times more revenue growth and 1.6 times higher EBIT margins than other companies, and expect twice the revenue increase and 40% greater cost reductions by 2028, per BCG (2025).
- Leading companies allocate more than 80% of their AI investments to reshaping key functions and inventing new offerings rather than small productivity initiatives, per BCG (2025).
- AI agent deployment reached 80% among US technology companies, close to 30 percentage points higher than the 54% average across industries, per KPMG’s US Q1 Pulse (2026).
- AI agent scaling is highest in Asia Pacific at 49%, followed by the Americas at 46% and EMEA at 42%, per KPMG’s Global AI Pulse (2026).
- The AI talent map is shifting under the market: the number of AI researchers and developers moving to the US has dropped 89% since 2017, with an 80% decline in the last year alone, per the Stanford AI Index (2026).
- Employment of software developers aged 22 to 25 fell about 20% year over year, in fields where AI most boosts productivity, per Stanford (2026).
- US consumers capture an estimated $172 billion per year in surplus value from AI tools, per the Stanford AI Index (2026). Much of AI’s measured value is landing outside any company’s P&L, which is part of why enterprise ROI numbers look thin.
AI market and leadership at a glance
| Measure | Figure | Source, year |
|---|---|---|
| Anthropic share of enterprise LLM spend | ~40% | Menlo Ventures, 2025 |
| OpenAI share of enterprise LLM spend | 27%, down from 50% in 2023 | Menlo Ventures, 2025 |
| CEOs who are their org’s main AI decision maker | 72%, 2x YoY | BCG, 2026 |
| CEOs whose job security depends on AI | 50% | BCG, 2026 |
| AI value driven by agentic AI today | 17%, ~2x by 2028 | BCG, 2025 |
| AI leaders’ growth advantage | 1.7x revenue growth, 1.6x EBIT margin | BCG, 2025 |
What this means for finance leaders
Two market facts matter more than any vendor ranking. First, provider share is volatile: a market leader lost half its share in three years, which means multi-model architectures and portable workloads aren’t paranoia, they’re pricing power.
Second, the CEO now owns AI personally, half of them believing their job depends on it, while the returns data stays flat. That combination lands on finance: someone has to connect executive conviction to unit-level evidence, and the teams that can price what each AI feature and agent actually costs are the ones who’ll have the answer when the board asks whether the conviction was justified.
Close your accountability gap. You now have 156 statistics about everyone else’s AI spend. CloudZero shows you yours: every model, feature, and customer, tied to the ROI it produced. Schedule a free demo today.
Frequently asked questions
