State of AI Spending 2026

Home / Indexes / State of AI / Spending
Report AI · State of AI series · Q4 2026 edition · Quarterly

State of AI Spending 2026

What companies and governments spend on AI, 2024 to 2026: capex, forecasts, enterprise budgets, power and public money, each figure dated and rated

Thank you for reading this post, don't forget to subscribe!

By the Report AI editorial team · Published · Next edition: Q1 2027
$2.67T (forecast)
Worldwide AI spending (Gartner), 16 Sep 2026 MEDIUM
$521B (forecast)
AI infrastructure spending (IDC), 6 Nov 2026 MEDIUM
42% (forecast)
AI spending as a share of worldwide IT spending (Gartner on Gartner), Sep–Oct 2026 tables DERIVED
$176.8B
Enterprise AI deployment spending, annualised run-rate (Menlo Ventures), Sep 2026 run-rate (published 3 Nov 2026) MEDIUM
At a glance
  • Gartner forecast on 16 September 2026 that worldwide AI spending will reach about $2.67 trillion in 2026, up 49.5% on its estimate of $1.79 trillion for 2025, with AI infrastructure about 56% of the total.
  • Microsoft, Alphabet, Amazon and Meta spent about $508 billion on capital expenditure from January to September 2026, more than the roughly $413 billion they spent in all of 2025, according to Report AI’s sum of their reported results.
  • The same four companies’ full-year 2026 capex guidance, as given in late October 2026, adds up to roughly $720–740 billion, against about $245 billion of capex in calendar 2024.
  • IDC counted $318.7 billion of worldwide AI infrastructure spending in 2025, up 107% from $153.9 billion in 2024, and on 6 November 2026 raised its 2026 forecast to $521 billion.
  • Menlo Ventures estimated that US enterprises spent $37 billion on generative AI in 2025; its November 2026 report puts enterprise AI deployment spending at an annualised $176.8 billion in September 2026, on a new definition that is not comparable.
  • Anthropic’s share of US enterprise LLM API spending rose from 24% in 2024 to 40% in 2025 and 55% in 2026, according to Menlo Ventures, an Anthropic investor.
  • Government AI money remains small next to corporate capex: the largest verified public items in 2026 are one-off vehicles such as Abu Dhabi’s $49 billion MGX fund (July 2026) and the UK’s £500 million Sovereign AI Fund (April 2026), not annual budgets.
Cite this report
Report AI, State of AI Spending 2026, Q4 2026 edition, published 16 November 2026. https://report-ai.org/indexes/state-of-ai/state-of-ai-spending/
Download the data (CSV) · 26 metrics with source, date and confidence for every value. Free to quote with attribution and a link.
In this report
  1. 1. Worldwide spending — Forecasters raised their 2026 AI spending estimates all year, and most of the new money is going to hardware
  2. 2. Big-tech capex — The four largest spenders’ capex roughly doubled in two years and now runs ahead of the cash their businesses generate
  3. 3. Power — Electricity is the physical limit on AI spending, and data-centre demand is on course to double by 2030
  4. 4. Enterprise spending — Enterprise spending on AI models is growing faster than adoption, and one provider now takes most of it
  5. 5. Industries and budgets — AI is taking a growing share of technology spending, but industry rankings depend on what is measured
  6. 6. Governments — Government AI spending is small next to corporate capex, and its biggest headlines are not budgets
  7. Comparison
  8. The numbers, 2024–2026
  9. Outlook: what to watch in 2027
  10. FAQ · About the research
1. Worldwide spending

Forecasters raised their 2026 AI spending estimates all year, and most of the new money is going to hardware

Gartner forecast on 16 September 2026 that worldwide AI spending will reach about $2.67 trillion in 2026, up 49.5% on its estimate of about $1.79 trillion for 2025. It was the third time in eight months that Gartner raised its figure for the year: the January edition said $2.52 trillion and the May edition $2.59 trillion. Gartner expects a further rise to about $3.64 trillion in 2027. Its total is broad. It counts AI infrastructure, software, services, AI models, security and AI-enabled consumer devices, so it is best read as a measure of all spending that has some AI content rather than spending on AI alone. Gartner’s figures are built from vendor revenue and buyer surveys, and the company does not publish its full method. Treat the total as an order of magnitude and the direction of revision as the signal.

Infrastructure is the largest component and the main reason for the revisions. Gartner puts 2026 AI infrastructure spending at about $1.48 trillion, roughly 56% of its total. IDC, which counts only AI-optimised servers and storage, measured $153.9 billion of such spending in 2024 and $318.7 billion in 2025, a rise of 107%. On 6 November 2026 IDC raised its 2026 forecast to $521 billion, from $497 billion in July and $487 billion in April, citing higher component prices, demand for CPU-based systems and continued cloud-provider capex. IDC expects the series to pass $1 trillion in 2029. The infrastructure share matters because hardware is bought ahead of demand: servers and data centres are paid for before the services that run on them earn revenue, which is why infrastructure leads every other category in both firms’ figures.

Exhibit 1
Gartner raised its 2026 AI spending forecast three times in eight months, by about $150 billion in total
Gartner forecast of worldwide AI spending in 2026, by edition, US$ trillion
0123US$ trillionJan 2026 edition2026 forecast, by edition, Jan 2026 edition: 2.52 US$ trillion2.52May 2026 edition2026 forecast, by edition, May 2026 edition: 2.59 US$ trillion2.59Sep 2026 edition2026 forecast, by edition, Sep 2026 edition: 2.67 US$ trillion2.67
View the data
2026 forecast, by edition
Jan 2026 edition2.52
May 2026 edition2.59
Sep 2026 edition2.67
Source: Gartner press releases of 15 Jan, 19 May and 16 Sep 2026; Report AI analysis MEDIUM
Notes: All three values are forecasts of the same year, made at different times. The September edition restructured categories (for example, splitting out AI agents and assistants), so part of the last step is definitional. Gartner’s own release rounds the September figure to $2.7 trillion. Increase Jan to Sep = $2.67T − $2.52T ≈ $0.15T (DERIVED).REPORTAI report-ai.org

The quarterly data show why the forecasts moved. IDC counted about $90 billion of AI infrastructure spending in the fourth quarter of 2025 and $89.7 billion in the first quarter of 2026, a flat start to the year that IDC attributes to paused deployments. Spending then rose to $121.4 billion in the second quarter, up 53% year on year and about a third on the previous quarter. Servers made up 98% of the value. One strong quarter was enough to lift the full-year forecast by $24 billion. The quarterly line is the closest thing to a measured series in this report. It is still an estimate, built from vendor revenue rather than buyers’ accounts, and IDC revises past quarters when vendors restate, so the latest edition should always be used.

Two cautions apply to the headline totals. First, Gartner’s figures for completed years are estimates that it revises: its September 2025 forecast put 2025 at $1.48 trillion, its January 2026 table at $1.757 trillion and its September 2026 table at $1.79 trillion, partly because categories were added. A growth rate computed across editions is not a measured growth rate. Second, the IDC and Gartner totals measure different things and should never be added together or used to check one another. They agree on direction, not on size. For 2026 the safest statement is the range of forecasts and their direction: every published figure for the year was revised upward between January and November, and none was revised down. That pattern can change quickly if hardware deliveries slip.

Exhibit 2
Quarterly AI infrastructure spending jumped by a third in the second quarter of 2026 after a flat first quarter
IDC worldwide AI infrastructure spending per quarter, actuals, US$ billion, Q4 2025 to Q2 2026
0255075100125US$ billionQ2 2026Q1 2026Q4 2025AI infrastructure, actual, Q4 2025: 90 US$ billionAI infrastructure, actual, Q1 2026: 89.7 US$ billionAI infrastructure, actual, Q2 2026: 121 US$ billion12190
View the data
AI infrastructure, actual
Q4 202590
Q1 202689.7
Q2 2026121
Source: IDC Worldwide Quarterly AI Infrastructure Tracker releases of 16 Apr, 21 Jul and 6 Nov 2026 MEDIUM
Notes: Actuals only; IDC’s full-year 2026 forecast ($521B) is not plotted on this line. Q4 2025 is given by IDC as ‘about $90 billion’. Hardware only (servers were 98% of Q2 value). Q2 2026 was +53% year on year and about +34% on Q1. Seen via search extracts; primary pages could not be opened.REPORTAI report-ai.org
Why AI spending forecasts disagree

Gartner’s $2.67 trillion and IDC’s $521 billion for 2026 are both correct on their own terms. Gartner counts any spending with AI content, including AI-enabled smartphones and PCs, software with AI features and AI services. IDC’s tracker counts AI-optimised servers and storage only. IDC also publishes a separate, broader spending guide, whose 2029 figure of $1.3 trillion is often misquoted as ‘agentic AI spending’. Before comparing two figures, check four things: which categories are included, whether the year is complete (an estimate) or still ahead (a forecast), which edition of the forecast is being quoted, and whether the figure is an annual total or an annualised run-rate.

2. Big-tech capex

The four largest spenders’ capex roughly doubled in two years and now runs ahead of the cash their businesses generate

Capital expenditure is the most concrete measure of AI spending because it is reported quarterly in audited accounts. By Report AI’s sum of reported figures, Microsoft, Alphabet, Amazon and Meta spent about $245 billion in calendar 2024 and about $413 billion in calendar 2025, an increase of roughly 69%. From January to September 2026 they spent about $508 billion, already more than in all of 2025. The third quarter of 2026 alone accounted for about $207 billion: Amazon $75.9 billion, Alphabet $51.4 billion, Microsoft $45.2 billion and Meta $34.91 billion. Quarterly capex for the four has risen steeply since 2024: the third-quarter figure for 2026 alone was more than four-fifths of their combined spending for the whole of 2024.

Not all of this is AI. Capex also covers warehouses, offices, network equipment and conventional cloud capacity. But each company has said that the increase is driven by data centres, servers and accelerators for AI, and none breaks out an AI-only figure. The totals are also approximate because the companies use different bases: Amazon reports cash capex net of incentives, Microsoft and Meta include assets acquired under finance leases, and Alphabet reports purchases of property and equipment. Off-balance-sheet structures such as Meta’s roughly $27 billion Hyperion financing with Blue Owl are not in these numbers at all. Analysts often call these totals ‘AI capex’; it is more accurate to call them capex driven by AI, and to quote each company’s own basis alongside each number.

Exhibit 3
Each of the four largest spenders had passed or nearly matched its 2025 capex by the end of September 2026
Capital expenditure by company, US$ billion: calendar 2024 and 2025 (full year) and January–September 2026
2024 (full year)2025 (full year)2026 (Jan–Sep)050100150200US$ billionAmazon2024 (full year), Amazon: 77.7 US$ billion77.72025 (full year), Amazon: 132 US$ billion1322026 (Jan–Sep), Amazon: 172 US$ billion172Microsoft2024 (full year), Microsoft: 75.6 US$ billion75.62025 (full year), Microsoft: 118 US$ billion1182026 (Jan–Sep), Microsoft: 118 US$ billion118Alphabet2024 (full year), Alphabet: 52.5 US$ billion52.52025 (full year), Alphabet: 91.4 US$ billion91.42026 (Jan–Sep), Alphabet: 132 US$ billion132Meta2024 (full year), Meta: 39.2 US$ billion39.22025 (full year), Meta: 72.2 US$ billion72.22026 (Jan–Sep), Meta: 85.8 US$ billion85.8
View the data
2024 (full year)2025 (full year)2026 (Jan–Sep)
Amazon77.7132172
Microsoft75.6118118
Alphabet52.591.4132
Meta39.272.285.8
Source: company results as reported by Reuters, CNBC, Sherwood News and others; Report AI sums of quarterly figures DERIVED
Notes: Bases differ: Amazon is cash capex net of proceeds and incentives (gross 2024 ≈ $83B); Microsoft and Meta include finance leases; Alphabet is purchases of property and equipment. Microsoft values are DERIVED sums of fiscal quarters re-cut to calendar years; Microsoft’s 2026 figure equals its 2025 total to within $0.1B. Meta 2026 includes a single-source Q2 (~$31B). The 2026 bars cover nine months and are not a full year. Oracle reports fiscal years ending May and is excluded.REPORTAI report-ai.org

Guidance points higher still. In late October 2026 Amazon reaffirmed about $220 billion for 2026, Alphabet held $195–205 billion, Meta narrowed its range to $135–145 billion and Microsoft gave about $170 billion, which most outlets read as calendar 2026 and one as its fiscal year to June 2027. Together that is roughly $720–740 billion, which implies about $220 billion in the fourth quarter. At the start of 2026 the same four had guided to roughly $630–650 billion. Alphabet and Amazon have said spending will rise again in 2027, without giving numbers. Guidance has moved in one direction: Alphabet’s 2026 range rose from $175–185 billion in February to $180–190 billion in April and $195–205 billion in July, and Amazon’s from about $200 billion to about $220 billion.

Oracle, which reports fiscal years ending in May, shows the same pattern from a smaller base: gross capex of $21.2 billion in fiscal 2025, $55.7 billion in fiscal 2026 and guidance of $90–95 billion for fiscal 2027, with $28.5 billion spent in its first quarter to August 2026. For Oracle, Meta, Alphabet and Amazon, capex now exceeds operating cash flow. Meta’s free cash flow was negative $2.12 billion in the third quarter, Alphabet’s about negative $9.1 billion and Amazon’s negative $42.4 billion over twelve months. Between 26 October and 5 November 2026, Meta, Oracle and Alphabet raised more than $70 billion of new equity and debt. Oracle’s new notes were priced at coupons of up to 8.65%. The cost of financing is now part of the cost of the build-out.

Exhibit 4
Full-year 2026 guidance implies about $220 billion more capex in the fourth quarter
2026 capital expenditure guidance as of late October 2026, midpoint, US$ billion
US$ billionAmazonAmazon: 220 US$ billion220AlphabetAlphabet: 200 US$ billion200MicrosoftMicrosoft: 170 US$ billion170MetaMeta: 140 US$ billion140
View the data
2026 guidance (midpoint)
Amazon220
Alphabet200
Microsoft170
Meta140
Source: company guidance on Q3 2026 earnings calls (21–29 Oct 2026), as reported; Report AI analysis DERIVED
Notes: Guidance, not outcomes. Ranges: Alphabet $195–205B, Meta $135–145B; Amazon ~$220B; Microsoft ~$170B, which most outlets report as calendar 2026 and one as fiscal 2027 (CONTESTED). Sum ≈ $730B (range $720–740B). Implied Q4 = $730B − $508B Jan–Sep ≈ $222B (DERIVED, approximate because bases differ).REPORTAI report-ai.org
How we count capex

We use each company’s reported capital expenditure for calendar years. Microsoft’s fiscal year ends in June, so its calendar figures are sums of four fiscal quarters, marked DERIVED. Oracle’s fiscal year ends in May and is shown separately. We use the basis each company reports and do not adjust for it: Amazon net of proceeds and incentives, Microsoft and Meta including finance leases, Alphabet purchases of property and equipment. Combined totals are therefore an order of magnitude, not a like-for-like sum. Guidance is shown as guidance and never on the same line as outcomes. Off-balance-sheet leases and joint ventures are excluded.

3. Power

Electricity is the physical limit on AI spending, and data-centre demand is on course to double by 2030

The International Energy Agency estimates that data centres worldwide used about 415 TWh of electricity in 2024, roughly 1.5% of global demand, and about 485 TWh in 2025. Sources differ on the 2025 growth rate, 17% in most secondary coverage against ‘almost 20%’ in IEA presentation material, so we rate it CONTESTED. The IEA’s April 2026 update estimates that electricity used by AI-focused data centres rose about 50% in 2025, far faster than the total, and projects global data-centre consumption of about 950 TWh in 2030 in its base case, slightly above the 945 TWh in its 2025 edition. On those estimates, data centres would use a little under 3% of global electricity in 2030, against about 1.5% in 2024. The IEA says it will update its projections regularly.

There is no IEA estimate for 2026 yet. Gartner separately forecast in June 2026 that data-centre electricity demand will grow 26% in 2026. We do not apply Gartner’s growth rate to the IEA’s base because the two organisations measure differently. In the United States, where data centres are most concentrated, Lawrence Berkeley National Laboratory projects data-centre electricity rising from 176 TWh in 2023 to between 325 and 580 TWh by 2028, a range wide enough to show how uncertain the build-out remains. The LBNL range runs from below to well above a doubling in five years. For spending, the important point is that grid connections, transformers and turbines take years to deliver, so electricity limits how fast capex can be turned into working capacity, whatever the budget.

Exhibit 5
Data-centre electricity use grew by about 70 TWh in 2025 and the IEA expects it roughly to double by 2030
Global data-centre electricity consumption, TWh: IEA estimates for 2024–2025 and base-case projection for 2030
0100200300400500TWh2024IEA estimate, 2024: 415 TWh4152025IEA estimate, 2025: 485 TWh485
View the data
IEA estimateIEA projection
2024415
2025485
2030 (base case)950
Source: IEA, Energy and AI (April 2025) and Key Questions on Energy and AI (April 2026) MEDIUM
Notes: Estimates and projection are separate series. The 2030 projection is 950 TWh in the 2026 edition (945 TWh in the 2025 edition). 2025 growth is CONTESTED: about 17% in secondary coverage, ‘almost 20%’ in IEA slides. No IEA estimate for 2026; Gartner separately forecasts +26% in 2026 (June 2026).REPORTAI report-ai.org

Power shapes spending in two ways. It limits where capacity can be built, and it adds to the cost of each site. The 10-gigawatt project planned at a former uranium-enrichment plant in Ohio includes a 9.2-gigawatt gas-fired power station of its own, because no grid connection of that size is available; independent estimates put generation at about $37 billion of a roughly $500 billion total. Gas-turbine makers’ order books stretch years ahead. These constraints help explain why IDC’s first quarter of 2026 was flat and why some of Microsoft’s third-quarter spending slipped. GE Vernova’s gas-turbine backlog reached about 116 GW in the second quarter of 2026 against manufacturing capacity of roughly 20 GW a year. Money cannot buy delivery slots that do not exist.

The energy figures should be read with care. The IEA’s numbers are modelled estimates, not metered totals, and the share attributable to AI rather than to conventional cloud and storage is itself an estimate. Per-query energy figures published by AI companies are small, about 0.2–0.3 watt-hours for a text prompt, but they exclude training and idle capacity. The fleet total is what matters for grids and spending, and it is the fleet total that is growing fastest. Water use follows the same pattern: per-site consumption depends on the cooling method, which announcements rarely specify. We therefore quote the IEA’s fleet estimates, with their dates, and treat per-query figures as indicative. Neither tells us how much electricity goes to training models and how much to serving them.

Estimates, projections and forecasts

An estimate describes a year that has ended but has not been fully measured, such as the IEA’s 485 TWh for 2025. A projection describes a scenario, such as the IEA’s 950 TWh base case for 2030, which depends on stated assumptions about efficiency and demand. A forecast, such as Gartner’s +26% for 2026, is a single expected outcome. We plot estimates and projections as separate series and never join them with a line. When an organisation revises an estimate, we keep both values and mark the later one RESTATED.

4. Enterprise spending

Enterprise spending on AI models is growing faster than adoption, and one provider now takes most of it

Menlo Ventures’ annual survey of US enterprise decision-makers is the most cited measure of company spending on generative AI. It estimated $37 billion of spending in 2025, 3.2 times its figure for 2024, split roughly evenly between applications (about $19 billion) and infrastructure (about $18 billion). Coding tools were the largest application category, at about $4 billion. Menlo’s numbers have been restated as its method changed: it first reported $13.8 billion for 2024, then restated that to $11.5 billion after excluding inference. We keep both values. The survey covers about 500 US enterprise decision-makers combined with a bottom-up model of the market, so it describes large US buyers rather than all companies worldwide. Menlo invests in some of the companies whose revenue it estimates, which it discloses.

In November 2026 Menlo moved to a different measure: enterprise AI deployment spending as an annualised run-rate. It estimated that rate at $176.8 billion in September 2026, of which $99.5 billion was infrastructure and $77.3 billion applications, 5.4 times a restated $33 billion a year earlier. The figure excludes model training. It is not money spent in 2026 and it is not a successor to the $37 billion, so a growth rate from one to the other would be meaningless. SaaStr’s reading of the report is that only 22% of the spending came out of software budgets. The run-rate is useful for one thing: it shows the level of spending at the end of the period, which is higher than the average across the year when spending is growing quickly.

Exhibit 6
Anthropic more than doubled its share of enterprise API spending in two years while Google’s fell back
Share of US enterprise LLM API spending by provider, %, 2024–2026
2024202520260204060%Anthropic2024, Anthropic: 24 %242025, Anthropic: 40 %402026, Anthropic: 55 %55OpenAI2024, OpenAI: 34 %342025, OpenAI: 27 %272026, OpenAI: 31 %31Google2024, Google: 12 %122025, Google: 21 %212026, Google: 9 %9
View the data
202420252026
Anthropic244055
OpenAI342731
Google12219
Source: Menlo Ventures, State of Generative AI in the Enterprise (Nov 2024, Dec 2025) and 2026: The State of Enterprise AI (3 Nov 2026) MEDIUM
Notes: Google 2024 (~12%) is a Report AI interpolation between Menlo’s 2023 (7%) and 2025 (21%) values (DERIVED). Survey-based estimates of US enterprises (~400–600 respondents per edition). Menlo Ventures is an investor in Anthropic.REPORTAI report-ai.org

The split by provider has changed faster than the totals. Menlo estimated that Anthropic took 24% of US enterprise LLM API spending in 2024, 40% in 2025 and 55% in 2026, while OpenAI went from 34% to 27% and then 31%. Google rose to 21% in 2025 and then fell to 9%. API spending itself ran at about $3.5 billion annualised in November 2024 and $8.4 billion by mid-2025; we found no dollar total for 2026. Menlo is an investor in Anthropic, and these are survey estimates, so we rate them MEDIUM. Together, Anthropic and OpenAI took 86% of enterprise API spending in 2026, up from 67% a year earlier; xAI and open-weight Chinese models together took just under 5%. Concentration rose even as the number of usable models grew.

Spending on AI agents is the most inflated category in public discussion. Gartner’s widely quoted $206.5 billion for 2026 counts all spending on software that has agentic features built in, much of it existing enterprise applications. Its line for standalone AI agents and assistants is $29.2 billion for 2026, up from $16.5 billion in 2025. Research firms that size a standalone agents market put it at about $7–8 billion in 2025, with a median 2030 forecast of about $49 billion. Adoption data support the lower figures: Gartner’s 2026 survey found 17% of organisations had deployed agents. McKinsey’s 2025 survey found 23% of organisations scaling an agentic system somewhere and no more than 10% in any single function, and Gartner expects over 40% of agentic AI projects to be cancelled by the end of 2027.

Exhibit 7
‘AI agent spending’ ranges from $7 billion to $1.3 trillion depending on what is counted
Published measures of spending on AI agents, US$ billion, by definition
Measure20252026Later year
Standalone agents market, research firms (median of 20)~7–8—~49 (2030)
Gartner: standalone AI agents and assistants16.529.2 (fcst)65.5 (2027)
Gartner: all software with agentic features86.4206.5 (fcst)376.3 (2027); ~985 (2030)
IDC: total AI spending, ‘driven by agentic AI’——~1,300 (2029)
Source: Gartner (5 May 2026; 16 Sep 2026; abstract 17 Feb 2026); IDC (26 Aug 2025); Report AI comparison of 30 forecasts (November 2026) MEDIUM
Notes: These are different markets, not competing estimates of one market. The research-firm median is DERIVED (each forecast converted to 2030 at its own CAGR). IDC’s $1.3T is all AI spending. All figures MEDIUM; Gartner and IDC primary pages could not be opened.REPORTAI report-ai.org
Why Menlo’s figures cannot be chained

Menlo Ventures has published four different bases in three years. Its November 2024 report gave $13.8 billion for 2024, including inference. Its December 2025 report gave $37 billion for 2025 and restated 2024 to $11.5 billion and 2023 to $1.7 billion, excluding inference. Its November 2026 report measures deployment spending as an annualised run-rate, $176.8 billion in September 2026 against a restated $33 billion a year earlier, excluding model training. Each figure is useful within its own edition. Growth rates should only be computed within one edition, and every quotation should name the edition and the measure.

5. Industries and budgets

AI is taking a growing share of technology spending, but industry rankings depend on what is measured

Comparing Gartner’s AI spending forecast with its own forecast of worldwide IT spending gives a rough measure of AI’s weight in technology budgets. On our arithmetic, AI spending was about 19% of IT spending in 2024, about 32% in 2025 and is forecast at about 42% in 2026, when Gartner expects IT spending of $6.39 trillion. The ratio assumes that Gartner’s AI categories sit inside its IT total, which Gartner does not state, and part of the rise reflects Gartner widening its AI definition. It describes market spending, not the budget of a typical company. The same calculation using Gartner’s earlier editions would give different values, so the ratio should be quoted with its inputs and years. It is a rough indicator, not a statistic.

Surveys of buyers point in the same direction without giving a clean share. Futurum’s September 2026 survey of 1,636 enterprise decision-makers found 46.9% reporting AI spending over budget in the second half of 2026, and 38.7% of those said they had moved money from elsewhere in the IT budget to cover it. RBC’s December 2025 survey of 117 IT leaders found 90% planned to increase AI spending in 2026. Neither survey gives the share of IT budgets spent on AI. Both surveys suggest AI is being funded partly by moving money, not only by adding it. BCG’s May 2026 survey of 423 IT leaders found they expected overall IT spending to grow 5.8% in 2026, far less than any AI spending series, and BCG reported that buyers were pulling back in other areas to pay for AI.

Exhibit 8
Healthcare takes the largest share of vertical generative-AI spending even though banking leads total AI spending
US enterprise vertical generative-AI software spending by industry, 2025, US$ million
US$ millionHealthcareHealthcare: 1,500 US$ million1,500Other verticalsOther verticals: 990 US$ million990LegalLegal: 650 US$ million650Creator toolsCreator tools: 360 US$ million360
View the data
Vertical GenAI spend, 2025
Healthcare1,500
Other verticals990
Legal650
Creator tools360
Source: Menlo Ventures, State of Generative AI in the Enterprise 2025 (n≈495), as shown on Report AI page 376 MEDIUM
Notes: Covers the ~$3.5B ‘vertical AI’ slice of enterprise GenAI spending only; healthcare is 43% of it. Not comparable with IDC’s all-AI industry figures, where banking leads at ~$31.3B in 2024 (>20% of worldwide AI spending) and is 12.6% of European AI spending in 2026 (IDC, 17 Sep 2026). ‘Other verticals’ includes financial services.REPORTAI report-ai.org

By industry, the ranking depends on the measure. On IDC’s all-AI measure, banking was the largest spender worldwide in 2024, with about $31.3 billion and more than 20% of the total; banking, software and information services, and retail together accounted for about $89.6 billion. In Europe, IDC puts banking at 12.6% of AI spending in 2026 and financial services as a whole at 19.2%. IDC’s worldwide industry tables for 2025 and 2026 are not public. IDC expects the top three verticals to reach about $222 billion by 2028. Financial services’ weight reflects long-standing use of machine learning in fraud detection, credit scoring and trading as well as newer generative tools; JPMorgan alone has said it spends about $2 billion a year on AI within an $18 billion technology budget.

On Menlo’s narrower measure of vertical generative-AI software, healthcare led in 2025 with $1.5 billion, 43% of about $3.5 billion, followed by legal at $650 million. The two rankings are not in conflict. Banks spend heavily across all forms of AI, including fraud detection and risk models built over many years, while healthcare providers are buying new generative tools, such as clinical documentation, faster than other sectors. Any statement that one industry ‘spends the most on AI’ should name the measure. Healthcare organisations’ adoption of domain-specific AI tools reached 22% in 2025, about ten times the 2023 level, according to Menlo. ‘Other verticals’, including financial services, made up about $990 million. All of these sums are small beside capex.

How we derived AI’s share of IT spending

Share = Gartner worldwide AI spending ÷ Gartner worldwide IT spending for the same year. 2024: $987.9 billion ÷ $5,114.8 billion = 19.3%, from the September 2025 AI table and the January 2025 IT table. 2025: $1,786.7 billion ÷ $5,581.7 billion = 32.0%, where 2025 IT spending is Gartner’s 2026 total of $6,391 billion divided by its stated growth of 14.5%. 2026: $2,670.5 billion ÷ $6,391 billion = 41.8%. All inputs are Gartner’s; the ratio is ours and is marked DERIVED. It is not a survey of company IT budgets.

6. Governments

Government AI spending is small next to corporate capex, and its biggest headlines are not budgets

Public AI money is measured in single-digit billions a year where corporate capex is measured in hundreds of billions. The US requested $3.3 billion of non-defence federal AI research and development for fiscal 2025, up from $3.1 billion for fiscal 2024; we did not find the fiscal 2026 total. Defence is larger: the Pentagon’s fiscal 2026 request included $13.4 billion for AI and autonomy. Brookings counted $7.2 billion of federal AI contract obligations in 2026 to May, almost all for the Department of Defense, though its 2024 base differs between accounts ($355 million or $675 million). These are budget requests and contract records, not outlays, and they use different definitions. GAO counted $5.6 billion of federal AI contract obligations over fiscal 2022–24 using another method, so neither series can be extended with the other.

Elsewhere the largest commitments are one-off vehicles. China registered the ¥344 billion (about $47.5 billion) semiconductor ‘Big Fund III’ in May 2024 and a ¥60.1 billion (about $8.2 billion) national AI fund in January 2025. The EU’s verifiable public commitment is more than €2.7 billion for EuroHPC AI Factories over 2024–27; its €200 billion InvestAI is a mobilisation total that includes hoped-for private capital. The $500 billion Stargate venture contains no US federal funds. Both Chinese funds are equity vehicles registered once and invested over many years, not annual budgets. China also announced in March 2025 a national venture-capital guidance fund intended to mobilise about ¥1 trillion over 20 years. InvestAI includes a planned €20 billion public-private fund for AI gigafactories.

Exhibit 9
Public AI commitments since 2024 have mostly been one-off funds announced on single dates, not recurring budgets
Selected government and sovereign AI commitments, announcement or programme period, 2024–2030
ChinaEUUKUSGulf2024202520262027202820292030China Big Fund IIIChina Big Fund III: 2024-05China national AI fundChina national AI fund: 2025-01EU AI FactoriesEU AI Factories: 2024-12 to 2027-12EU InvestAI (headline)EU InvestAI (headline): 2025-02UK £2B AI envelopeUK £2B AI envelope: 2025-06 to 2030-03UK Sovereign AI FundUK Sovereign AI Fund: 2026-04US AI R&D (FY2025)US AI R&D (FY2025): 2024-10 to 2025-09HUMAIN launchedHUMAIN launched: 2025-05MGX Fund I closeMGX Fund I close: 2026-07
View the data
EventGroupStartEnd
China Big Fund IIIChina2024-05
China national AI fundChina2025-01
EU AI FactoriesEU2024-122027-12
EU InvestAI (headline)EU2025-02
UK £2B AI envelopeUK2025-062030-03
UK Sovereign AI FundUK2026-04
US AI R&D (FY2025)US2024-102025-09
HUMAIN launchedGulf2025-05
MGX Fund I closeGulf2026-07
Source: Report AI pages 374 and 955; DSIT; Forbes; NITRD MEDIUM
Notes: Amounts: China ¥344B (~$47.5B) and ¥60.1B (~$8.2B); EU >€2.7B AI Factories (2024–27; start month approximate), €200B InvestAI is a mobilisation headline, not a budget; UK £2B (2025–30) and £500M fund (+£60M in Oct 2026); US $3.3B non-defence AI R&D request for FY2025 (Oct 2024–Sep 2025); MGX $49B. HUMAIN’s capital was not disclosed. Never sum across these types.REPORTAI report-ai.org

The UK set a £2 billion AI envelope for 2025–30 in its June 2025 Spending Review, about half for sovereign compute, launched a £500 million Sovereign AI Fund in April 2026 and added £60 million in the October 2026 Budget. Reports of a ‘£700 million sovereign AI fund’ describe the Budget’s wider AI package, whose size is reported as £600 million or £700 million. In the Gulf, Abu Dhabi’s MGX closed a $49 billion first fund in July 2026, and Saudi Arabia’s HUMAIN, launched in May 2025, has not disclosed its capital. MGX invests commercially for return and holds stakes in OpenAI, Anthropic and xAI, so it is better compared with a sovereign-wealth fund than with a budget. Its stated goal is more than $100 billion of assets.

These figures cannot be added together. An annual appropriation, a fund registered once and invested over years, a sovereign-wealth vehicle that invests for return and a mobilisation headline are four different things. By verifiable annual budget, Japan’s ¥1.23 trillion (about $7.9 billion) for chips and AI in fiscal 2026 and South Korea’s ₩10.1 trillion (about $7.4 billion) for 2025 remain larger than the US non-defence AI R&D line. None is close to a single quarter of one hyperscaler’s capex. India’s IndiaAI Mission, at about $1.25 billion from 2024, and Singapore’s more than S$1 billion are smaller still. Governments shape where AI capacity is built through permits, power and export rules far more than through their own spending.

Four kinds of public AI money

Appropriation: money a legislature or ministry has allocated for a year, such as US federal AI R&D; the only kind that can be compared across countries. Equity vehicle: capital registered once and invested over years, such as China’s national AI fund. Sovereign vehicle: public capital invested commercially for return, such as MGX or HUMAIN. Mobilisation headline: a total that bundles public money with private capital a government hopes to attract, such as InvestAI’s €200 billion. We label every government figure by type and never sum across types.

Comparison

Between 2024 and 2026 every measured AI spending series at least doubled, but the measures moved further apart

On the measures that are counted rather than modelled, AI spending roughly doubled each year. Big-four capex went from about $245 billion in 2024 to about $413 billion in 2025, and reached about $508 billion in the first nine months of 2026. IDC’s AI infrastructure series went from $153.9 billion to $318.7 billion, with $521 billion forecast for 2026. Enterprise API spending, enterprise generative-AI budgets and data-centre electricity all rose, though electricity grew far more slowly, by about 17–20% in 2025, because efficiency gains absorb part of the hardware growth.

Across companies, Amazon spent the most in each year, but Alphabet grew fastest: its January–September 2026 capex of $132.0 billion was 2.5 times its full-year 2024 figure. Microsoft’s 2026 spending to September matched its whole 2025 total. Among model providers, Anthropic’s share of enterprise API spending more than doubled from 2024 to 2026, while Google’s rose and then fell. Meta’s 2026 guidance is about twice its 2025 total, and Amazon’s nine-month spending in 2026 exceeded every company’s full-year figure for 2025.

The comparison also shows how unstable the measures are. Gartner’s estimate for 2025 moved by more than $300 billion between editions, Menlo changed its definition twice, and IDC revised its 2026 infrastructure forecast three times. A reader comparing years should use one source and one edition at a time, and treat any cross-source growth rate as unreliable.

Exhibit 10
On every measured series, AI spending roughly doubled between 2024 and 2025 and is on course to rise again in 2026
Key AI spending measures by year, with status of each figure
Measure202420252026
Big-four capex (US$B)~245~413~508 Jan–Sep; ~730 guided
IDC AI infrastructure (US$B)153.9318.7521 (forecast)
Gartner AI spending (US$T)0.99 (est.)1.79 (est.)2.67 (forecast)
Menlo enterprise GenAI (US$B)11.5 (restated)37n/a: new measure
Anthropic API share (%)244055
Data-centre power (TWh)~415~485not yet estimated
AI share of IT spend (%)193242 (forecast)
Source: data/spending.json (Gartner, IDC, Menlo Ventures, IEA, company results); Report AI analysis DERIVED
Notes: Big-four capex and AI share of IT spending are DERIVED. Gartner’s 2024 and 2025 values come from different editions with different categories. Menlo’s 2026 report uses an annualised run-rate ($176.8B) that is not a successor to the $37B. The 2026 column mixes year-to-date actuals and forecasts; read each cell’s label.REPORTAI report-ai.org

Year over year, every comparable measure

Measure202420252026 (to date)2025→26
Worldwide AI spending (Gartner)$988B$1.79T$2.67T (forecast)+49%
AI infrastructure spending (IDC)$153.9B$318.7B$521B (forecast)+63%
AI spending as a share of worldwide IT spending (Gartner on Gartner)19%32%42% (forecast)+31%
Anthropic share of enterprise LLM API spending (Menlo Ventures)24%40%55%+38%
OpenAI share of enterprise LLM API spending (Menlo Ventures)34%27%31%+15%
Google share of enterprise LLM API spending (Menlo Ventures)~12%21%9%-57%
Microsoft capital expenditure, calendar year (incl. finance leases)$75.6B$118.0B$118.1B (Jan–Sep, to date)+0%
Alphabet capital expenditure, calendar year$52.5B$91.4B$132.0B (Jan–Sep, to date)+44%
Amazon capital expenditure, calendar year (cash, net of proceeds and incentives)$77.7B$131.8B$172.2B (Jan–Sep, to date)+31%
Meta capital expenditure, calendar year (incl. finance-lease principal)$39.2B$72.2B~$85.8B (Jan–Sep, to date)+19%
Oracle capital expenditure, fiscal year ending 31 May (gross)$21.2B (FY2025, Jun 2024–May 2025)$55.7B (FY2026, Jun 2025–May 2026)$28.5B (FY2027 Q1, Jun–Aug 2026, to date)-49%
Combined capex of Microsoft, Alphabet, Amazon and Meta, calendar year~$245B~$413B~$508B (Jan–Sep, to date)+23%
  • Big-tech capex roughly doubled in two years: about $245 billion in 2024 and $413 billion in 2025 for Microsoft, Alphabet, Amazon and Meta, and $508 billion in the first nine months of 2026 alone (Report AI sums of reported figures).
  • IDC’s measured AI infrastructure spending went from $153.9 billion (2024) to $318.7 billion (2025), with $521 billion forecast for 2026; the first two are actuals, the third is a forecast.
  • Gartner’s worldwide AI spending series rose from about $988 billion (2024 estimate) to $1.79 trillion (2025 estimate) and a $2.67 trillion forecast for 2026, but category changes between editions mean part of each step is definitional.
  • In enterprise LLM APIs the leader changed and then pulled away: Anthropic’s share of enterprise API spend went 24% (2024), 40% (2025), 55% (2026), while OpenAI went 34%, 27%, 31% (Menlo Ventures, an Anthropic investor).

All eleven themes side by side: State of AI, year over year.

Data appendix

The numbers, 2024 to 2026

Every measure in this report with its value for each year, as-of date, source and confidence. Change compares 2025 with 2026 to date where both are numeric.

Worldwide AI spending

Measure202420252026 (to date)Change
Worldwide AI spending (Gartner)
US$ per year, all AI categories
$988B
Sep 2025 (estimate for a completed year), Gartner, AI spending forecast, 17 Sep 2025 MEDIUM
$1.79T
16 Sep 2026 (estimate for a completed year), Gartner, AI spending forecast, 16 Sep 2026 MEDIUM
$2.67T (forecast)
16 Sep 2026, Gartner, AI spending forecast, 16 Sep 2026 MEDIUM
+49%
AI infrastructure spending (IDC)
US$ per year, AI-optimised servers and storage
$153.9B
Nov 2026 (actual), IDC Worldwide Quarterly AI Infrastructure Tracker MEDIUM
$318.7B
Apr 2026 (actual), IDC Worldwide Quarterly AI Infrastructure Tracker, 16 Apr 2026 MEDIUM
$521B (forecast)
6 Nov 2026, IDC Worldwide Quarterly AI Infrastructure Tracker, 6 Nov 2026 MEDIUM
+63%

Budgets

Measure202420252026 (to date)Change
AI spending as a share of worldwide IT spending (Gartner on Gartner)
% of worldwide IT spending
19%
Jan 2025 / Sep 2025 tables, Report AI calculation from Gartner DERIVED
32%
Sep–Oct 2026 tables, Report AI calculation from Gartner DERIVED
42% (forecast)
Sep–Oct 2026 tables, Report AI calculation from Gartner DERIVED
+31%

Enterprise spending

Measure202420252026 (to date)Change
US enterprise generative-AI spending (Menlo Ventures)
US$ per year
$11.5B (restated; first published as $13.8B)
Dec 2025, Menlo Ventures, State of Generative AI in the Enterprise 2025 MEDIUM
$37B
9 Dec 2025, Menlo Ventures, State of Generative AI in the Enterprise 2025 MEDIUM
—
Enterprise AI deployment spending, annualised run-rate (Menlo Ventures)
US$, annualised run-rate
—$33B
Sep 2025 run-rate (published 3 Nov 2026), Menlo Ventures, 2026: The State of Enterprise AI MEDIUM
$176.8B
Sep 2026 run-rate (published 3 Nov 2026), Menlo Ventures, 2026: The State of Enterprise AI MEDIUM
+436%
Enterprise LLM API spending (Menlo Ventures)
US$, annualised
$3.5B
Nov 2024, Menlo Ventures, Mid-Year LLM Market Update, 31 Jul 2025 MEDIUM
$8.4B
Mid-2025, Menlo Ventures, Mid-Year LLM Market Update, 31 Jul 2025 MEDIUM
—
Anthropic share of enterprise LLM API spending (Menlo Ventures)
% of enterprise LLM API spend
24%
Nov 2024, Menlo Ventures, State of Generative AI in the Enterprise 2024 MEDIUM
40%
Dec 2025, Menlo Ventures, State of Generative AI in the Enterprise 2025 MEDIUM
55%
3 Nov 2026, Menlo Ventures, 2026: The State of Enterprise AI MEDIUM
+38%
OpenAI share of enterprise LLM API spending (Menlo Ventures)
% of enterprise LLM API spend
34%
Nov 2024, Menlo Ventures, State of Generative AI in the Enterprise 2024 MEDIUM
27%
Dec 2025, Menlo Ventures, State of Generative AI in the Enterprise 2025 MEDIUM
31%
3 Nov 2026, Menlo Ventures, 2026: The State of Enterprise AI MEDIUM
+15%
Google share of enterprise LLM API spending (Menlo Ventures)
% of enterprise LLM API spend
~12%
2025 (interpolated), Report AI interpolation between Menlo’s 2023 (7%) and 2025 (21%) values DERIVED
21%
Dec 2025, Menlo Ventures, State of Generative AI in the Enterprise 2025 MEDIUM
9%
3 Nov 2026, Menlo Ventures, 2026: The State of Enterprise AI MEDIUM
-57%

Big-tech capex

Measure202420252026 (to date)Change
Microsoft capital expenditure, calendar year (incl. finance leases)
US$ per calendar year
$75.6B
Jan 2025, Report AI sum of Microsoft quarterly results DERIVED
$118.0B
Jan 2026, Report AI sum of Microsoft quarterly results DERIVED
$118.1B (Jan–Sep, to date)
28 Oct 2026, Report AI sum of Microsoft quarterly results DERIVED
+0%
Alphabet capital expenditure, calendar year
US$ per calendar year (purchases of property and equipment)
$52.5B
Feb 2025, Alphabet results, as reported MEDIUM
$91.4B
Feb 2026, Alphabet Q4 2025 results, as reported MEDIUM
$132.0B (Jan–Sep, to date)
28 Oct 2026, Report AI sum of Alphabet quarterly results DERIVED
+44%
Amazon capital expenditure, calendar year (cash, net of proceeds and incentives)
US$ per calendar year
$77.7B
Feb 2025, Amazon results, as reported MEDIUM
$131.8B
Feb 2026, Amazon results, as compiled by Sherwood News MEDIUM
$172.2B (Jan–Sep, to date)
29 Oct 2026, Amazon Q3 2026 results, as reported MEDIUM
+31%
Meta capital expenditure, calendar year (incl. finance-lease principal)
US$ per calendar year
$39.2B
Jan 2025, Meta FY2024 results MEDIUM
$72.2B
Jan 2026, Meta results, as compiled by Sherwood News MEDIUM
~$85.8B (Jan–Sep, to date)
21 Oct 2026, Report AI sum of Meta quarterly results DERIVED
+19%
Oracle capital expenditure, fiscal year ending 31 May (gross)
US$ per fiscal year; year key = calendar year holding most of the fiscal year
$21.2B (FY2025, Jun 2024–May 2025)
Jun 2025, Oracle FY2025 results, as reported MEDIUM
$55.7B (FY2026, Jun 2025–May 2026)
10 Jun 2026, Oracle FY2026 results and call, as reported MEDIUM
$28.5B (FY2027 Q1, Jun–Aug 2026, to date)
Sep 2026, Oracle FY2027 Q1 results, via Zacks MEDIUM
-49%
Combined capex of Microsoft, Alphabet, Amazon and Meta, calendar year
US$ per calendar year, sum of each company’s reported basis
~$245B
Feb 2025, Report AI sum DERIVED
~$413B
Feb 2026, Report AI sum DERIVED
~$508B (Jan–Sep, to date)
29 Oct 2026, Report AI sum DERIVED
+23%
2026 full-year capex guidance, Microsoft + Alphabet + Amazon + Meta
US$, company guidance
——~$720–740B (guidance)
29 Oct 2026, Company guidance, summed by Report AI DERIVED

Energy and power

Measure202420252026 (to date)Change
Global data-centre electricity consumption (IEA)
TWh per year
~415 TWh
Apr 2025, IEA, Energy and AI (2025) MEDIUM
~485 TWh
Apr 2026, IEA, Key Questions on Energy and AI (Apr 2026) CONTESTED
—

AI agents

Measure202420252026 (to date)Change
Spending on software with agentic AI capabilities (Gartner)
US$ per year, embedded measure
—$86.4B
5 May 2026, Gartner press release, 5 May 2026 MEDIUM
$206.5B (forecast)
5 May 2026, Gartner press release, 5 May 2026 MEDIUM
+139%
Spending on standalone AI agents and assistants (Gartner)
US$ per year
—$16.5B
16 Sep 2026, Gartner AI spending forecast, 16 Sep 2026 MEDIUM
$29.2B (forecast)
16 Sep 2026, Gartner AI spending forecast, 16 Sep 2026 MEDIUM
+77%

Government spending

Measure202420252026 (to date)Change
US federal non-defence AI R&D, budget request (NITRD)
US$ per fiscal year
$3.09B (FY2024 request)
2023, NITRD supplement to the President’s FY2024 budget MEDIUM
$3.32B (FY2025 request)
Dec 2024, NITRD/NAIIO supplement to the President’s FY2025 budget MEDIUM
—
US federal AI contract obligations (Brookings)
US$ obligated
$355M or $675M
18 May 2026, Brookings, ‘Where does federal AI spending stand in 2026?’ CONTESTED
—$7.2B (to date)
18 May 2026, Brookings, ‘Where does federal AI spending stand in 2026?’ MEDIUM
EU public AI compute commitment: EuroHPC AI Factories
EUR, multi-year co-investment (EU + member states)
—>€2.7B
2025, EuroHPC Joint Undertaking HIGH
—
China state AI and chip investment vehicles (registered capital)
CNY, one-time equity vehicles (US$ in num)
¥344B (~$47.5B), Big Fund III
May 2024, China Daily / Caixin MEDIUM
¥60.1B (~$8.2B), National AI Industry Investment Fund
Jan 2025, MIIT / Ministry of Finance via China Daily MEDIUM
—
UK public AI commitments
GBP
—£2B (2025–30 envelope)
Jun 2025, GOV.UK, Spending Review 2025 HIGH
£560M Sovereign AI Fund (£500M + £60M)
28 Oct 2026, DSIT; Autumn Budget 2026 coverage (ITPro, CMS) CONTESTED
Gulf sovereign AI capital: MGX (Abu Dhabi) fund close
US$, fund commitments
——$49B (Fund I close)
Jul 2026, Forbes / Bloomberg MEDIUM

Spending by industry

Measure202420252026 (to date)Change
Banking AI spending, worldwide (IDC)
US$ per year
~$31.3B
2025, IDC Worldwide AI and GenAI Spending Guide MEDIUM
——
Fact check

Numbers to stop quoting

Widely repeatedWhat the evidence supports
“Gartner: worldwide AI spending will be $2.59 trillion in 2026 (or $2.5 trillion).”Superseded. Gartner’s 16 September 2026 forecast is about $2.67 trillion (rounded to $2.7 trillion), up 49.5%. The $2.52 trillion (15 Jan 2026) and $2.59 trillion (19 May 2026) figures were earlier editions. · detail
“$2.59 trillion in 2026 is up 47% from about $1.5 trillion in 2025.”The arithmetic does not work: $1.5 trillion plus 47% is about $2.2 trillion. Gartner had restated its 2025 base to about $1.76 trillion by January 2026; its September 2026 table gives $1.79 trillion. The $1.5 trillion was its September 2025 forecast for 2025. · detail
“IDC forecasts $487 billion (or $497 billion) of AI infrastructure spending in 2026.”Superseded. IDC raised its 2026 forecast to $521 billion on 6 November 2026, after $487 billion in April and $497 billion in July. · detail
“Hyperscaler capex will top $600 billion in 2026.”Out of date. Microsoft, Alphabet, Amazon and Meta alone had spent about $508 billion by September 2026, and their full-year guidance sums to roughly $720–740 billion (Report AI sum, October 2026). · detail
“The AI agents market is worth $206.5 billion in 2026 (Gartner).”Gartner’s $206.5 billion counts spending on all software with agentic features built in. Its own line for standalone AI agents and assistants is $29.2 billion for 2026. · detail
“IDC: agentic AI spending will reach $1.3 trillion by 2029.”IDC’s $1.3 trillion for 2029 (26 Aug 2025) is total AI spending, with agentic AI described as its main driver; it is not an agentic-AI market size. · detail
“Enterprises spent $177 billion on AI in 2026 (Menlo Ventures).”Menlo’s $176.8 billion is an annualised run-rate of enterprise AI deployment spending as of September 2026, on a new definition. It is not money spent in 2026 and is not comparable with Menlo’s $37 billion for 2025.
“The EU is investing €200 billion in AI; Stargate is $500 billion of US government AI spending.”InvestAI’s €200 billion is a mobilisation total including hoped-for private capital; the verifiable public commitment is >€2.7 billion for EuroHPC AI Factories. Stargate is a private venture with no federal funds. · detail
“The UK’s October 2026 Budget created a £700 million sovereign AI fund.”The Sovereign AI Fund is £500 million (launched 16 April 2026) plus a £60 million top-up in the Budget; £600–700 million is the reported size of the Budget’s wider AI package.
“Anthropic has 40% of enterprise LLM API spend.”True for Menlo’s 2025 report; its November 2026 report puts Anthropic at 55%, OpenAI 31% and Google 9%. · detail
Outlook

In 2027 the question is whether spending keeps outrunning cash flow, and the first test is fourth-quarter capex

Every forecast available in November 2026 expects AI spending to rise again in 2027: Gartner projects about $3.64 trillion of worldwide AI spending, IDC expects AI infrastructure to pass $1 trillion by 2029, and Alphabet and Amazon have said their capex will increase. The risk to that view is financial rather than technical. Three of the four largest spenders had negative free cash flow in the third quarter of 2026, and the gap is being filled with bonds, new shares and off-balance-sheet structures whose cost is rising, with some data-centre debt marketed at 11% or more.

Two things would change our view. A cut to 2027 capex plans by any of the four, or a first downward revision to Gartner’s or IDC’s forecasts, would signal that the build-out has peaked. Conversely, enterprise spending on models and agents growing fast enough to cover the depreciation on this capex would show the spending is being earned back; on current figures, measured enterprise revenue streams are an order of magnitude smaller than capex.

What to watch in 2027

IndicatorWhere it standsWhat would change the picture
Big-four capex, full year 2026~$508B for Jan–Sep 2026; guidance ~$720–740B (late Oct 2026)A full-year total below ~$700B would mean guidance was missed; first 2027 guidance below the 2026 level would mark a peak.
Big-four free cash flowMeta −$2.12B (Q3 2026), Alphabet ≈ −$9.1B (Q3 2026), Amazon −$42.4B (12 months to Sep 2026)A second straight quarter of negative free cash flow at all three, with further equity raises, would make financing, not demand, the binding constraint.
Gartner worldwide AI spending forecast$2.67T for 2026, $3.64T for 2027 (16 Sep 2026)A downward revision in the next edition would be the first of the cycle.
IDC quarterly AI infrastructure spending$121.4B in Q2 2026 (6 Nov 2026)Two consecutive quarters of sequential decline would contradict the $521B full-year forecast.
Standalone AI agent spending$29.2B forecast for 2026 (Gartner, 16 Sep 2026); 17% of organisations deployed agents (Gartner 2026)Gartner’s forecast that up to 40% of enterprise apps will include task-specific agents by end-2026 resolves in January 2027; a result far below would argue for the lower agent-spending figures.
Data-centre electricity~485 TWh in 2025 (IEA, April 2026)An IEA estimate for 2026 above ~610 TWh (+26%, Gartner’s forecast rate) would put grids, not chips, at the centre of 2027 spending plans.
Definitions
Capital expenditure (capex)
Spending on long-lived assets such as data centres, servers and land, capitalised and depreciated over years. Reported quarterly by listed companies; each company uses its own basis.
AI infrastructure (IDC)
Spending on AI-optimised servers and storage, measured by IDC’s quarterly tracker from vendor revenue. Hardware only.
Worldwide AI spending (Gartner)
Gartner’s total of spending on AI infrastructure, software, services, models, security and AI-enabled devices. Values for completed years are estimates that Gartner revises.
Enterprise generative-AI spending (Menlo)
Menlo Ventures’ survey-based estimate of US enterprise spending on generative-AI applications and infrastructure; definition has changed between editions.
Annualised run-rate
Spending in a recent period scaled to a full year. Not the same as money spent in a calendar year.
Standalone AI agents vs agentic software
Standalone: products sold as AI agents or assistants. Agentic software: all software with agent features built in, including existing applications. Gartner’s $29.2B and $206.5B for 2026 measure these two things.
Appropriation
Money a legislature or ministry has allocated for a period. Distinct from equity funds, sovereign-wealth vehicles and mobilisation headlines.
TWh
Terawatt-hour, one billion kilowatt-hours. Global electricity demand is roughly 30,000 TWh a year.

Frequently asked

How much will the world spend on AI in 2026?

Gartner forecast on 16 September 2026 that worldwide AI spending will reach about $2.67 trillion in 2026, up 49.5% from its estimate of about $1.79 trillion for 2025. About $1.48 trillion of the 2026 figure, roughly 56%, is AI infrastructure. It is a forecast that Gartner has raised twice this year, from $2.52 trillion in January and $2.59 trillion in May. IDC’s narrower measure, AI infrastructure hardware only, is forecast at $521 billion for 2026 (6 November 2026).

How much are Microsoft, Alphabet, Amazon and Meta spending on capex in 2026?

They spent about $508 billion on capital expenditure from January to September 2026, by Report AI’s sum of each company’s quarterly results: Amazon $172.2 billion, Alphabet $132.0 billion, Microsoft $118.1 billion and Meta about $85.8 billion. Their full-year 2026 guidance, given in late October 2026, adds up to roughly $720–740 billion: Amazon about $220 billion, Alphabet $195–205 billion, Microsoft about $170 billion (calendar or fiscal year is disputed) and Meta $135–145 billion. The companies report on slightly different bases, so totals are approximate.

How much do enterprises spend on generative AI?

Menlo Ventures estimated that US enterprises spent $37 billion on generative AI in 2025, 3.2 times the $11.5 billion it now estimates for 2024. Its November 2026 report uses a different measure: enterprise AI deployment spending running at an annualised $176.8 billion in September 2026, up from a restated $33 billion a year earlier. That is a run-rate on a new definition, not money spent in 2026, so it should not be compared directly with the $37 billion.

How much is being spent on AI agents in 2026?

It depends on the definition. Gartner forecasts $29.2 billion of spending on standalone AI agents and assistants in 2026 (16 September 2026 forecast), up from $16.5 billion in 2025. Its better-known $206.5 billion figure for 2026 (5 May 2026) counts all spending on software that has agentic features built in, including existing enterprise applications. Research firms that size a standalone agents market put it at about $7–8 billion in 2025 (Report AI comparison, November 2026).

How much do governments spend on AI?

Far less than companies, and much of it is one-off. The US requested $3.3 billion of non-defence federal AI R&D for fiscal 2025, and Brookings counted $7.2 billion of federal AI contract obligations so far in 2026 (May 2026), almost all for defence. The EU’s verifiable public commitment is the >€2.7 billion EuroHPC AI Factories programme, not the €200 billion InvestAI headline. The UK launched a £500 million Sovereign AI Fund in April 2026, topped up by £60 million in October, and Abu Dhabi’s MGX closed a $49 billion fund in July 2026.

Methodology & sources

Figures are compiled from company disclosures, regulators, research firms and reputable reporting, and cross-checked against Report AI’s own index pages. Each value carries its as-of date, source and a confidence rating: HIGH where the primary source was read, MEDIUM where it rests on consistent reputable secondary reporting (most figures, because many primary pages cannot be opened from our research environment), DERIVED where it is our own arithmetic, and CONTESTED where credible sources disagree. A year means the calendar year the figure describes; 2026 values are the latest available and marked “to date”. Weekly and monthly user counts, and forecasts and outcomes, are never compared as if they were the same measure. Exhibits plot only dated, sourced readings; lines are not interpolated between them, and values with different definitions are never placed on one axis. Known gaps in this edition: No primary page could be opened (WebFetch failed on gartner.com and idc.com), and the session’s web-search budget ran out before every item was checked; nothing in this file is rated HIGH from new research. HIGH ratings that appear are carried over from page 374.; Gartner 2024 AI spending on its 2026 category basis: not found. The 2024 value ($988B) is from the September 2025 edition, so the 2024-to-2025 step mixes editions.; Microsoft calendar-2024 and 2025 capex and Meta 2024 capex rest partly on quarterly inputs not re-verified in this pass; Microsoft’s Dec 2025 quarter ($37.5B) and Meta’s Q2 2026 (~$31B) are single-source.; Microsoft’s ~$170B guidance: calendar 2026 or fiscal 2027 (CONTESTED); FY27 Q1 components do not add up to the reported $45.2B.; Menlo Ventures 2026: no dollar total for enterprise LLM API spend found; Google’s fall to 9% share rests on secondary coverage; survey size reported as ~400 or 495.; IDC’s full-stack Worldwide AI and GenAI Spending Guide: totals for 2024–2026 not verified (2024 reported as $235B in the Aug 2024 edition and $315.9B in a later citation; a ‘$940B in 2026’ figure could not be traced). Worldwide industry splits for 2025–2026 are paywalled.; IEA: no 2026 data-centre electricity estimate; 2025 growth is 17% in secondary coverage and ‘almost 20%’ in IEA slides. No US data-centre power demand in GW was gathered.; US federal non-defence AI R&D request for FY2026 not found; Brookings’ 2024 contract-obligation base is CONTESTED ($355M vs $675M) and its 2026 period is not stated in what we saw.; No 2026 EU appropriation, no 2026 China AI vehicle, and no disclosed HUMAIN capital figure found. Japan (¥1.23T, FY2026) and South Korea (₩10.1T, 2025) budgets are on page 374 but not carried as metrics here.; No survey gives a clean ‘share of IT budget spent on AI’ for 2024–2026; the share metric here is Gartner AI spending divided by Gartner IT spending (DERIVED).; Site pages carrying stale spending figures: 376, 967, 362 ($2.59T/$2.5T); 93, 376 ($497B); 73, 320, 354, 318 ($487B); 73, 354, 349, 320 (‘$600B+ hyperscaler capex 2026’); 321 (Anthropic 40%, and its 2024 shares are labelled interpolated although Menlo published 24% and 34%).. These reports compile and verify secondary data with original analysis; they are not original surveys. See our editorial standards, corrections log and fact-check register.