AI Statistics 2026

44%

44% of US adults had used ChatGPT by early 2026

44% of US adults reported using ChatGPT in a Pew Research Center survey of 5,119 adults conducted February 17-23, 2026, up from 34% a year earlier. Pew found about half of US adults now use AI chatbots of some kind. Nationally representative survey.

Source: Pew Research Center · 2026-06

20M+

Microsoft 365 Copilot passed 20 million paid seats

Microsoft said Microsoft 365 Copilot crossed 20 million paid enterprise seats as of its Q3 FY2026 earnings (April 2026), up from about 15 million a quarter earlier, with more than 90% of the Fortune 500 using Copilot. Microsoft does not publish a single unified consumer Copilot user metric. Company-disclosed figure.

Source: Microsoft (Q3 FY2026 earnings) · 2026-04

~117M

xAI's Grok reached an estimated 117 million monthly users

Grok, xAI's assistant integrated into X, was reported at about 117 million monthly active users in a March 2026 SpaceX IPO filing, up from roughly 35 million in December 2025 and ~64 million cited by xAI at year-end 2025. Grok's tight integration with the X feed inflates reach versus standalone apps. Figure drawn from a regulatory filing rather than a standalone xAI usage disclosure.

Source: SpaceX IPO filing (via reports) · 2026-03

900M

ChatGPT has passed 900 million weekly active users

OpenAI said ChatGPT reached about 900 million weekly active users, announced in its February 2026 funding-round disclosure. That extends a steep climb the company has reported itself: roughly 100 million weekly users in early 2023, 400 million in February 2025, 700 million in July 2025 and about 800 million at DevDay in October 2025. OpenAI's own figures are point-in-time disclosures, not audited metrics.

Source: OpenAI · 2026-02

50M+

ChatGPT has more than 50 million paying subscribers

OpenAI disclosed more than 50 million paying ChatGPT subscribers in its February 2026 funding-round update, adding that January and February 2026 were on track to be its largest months ever for new subscribers. This is OpenAI's own figure.

Source: OpenAI · 2026-02

54%

54% of US teens use AI chatbots for schoolwork

54% of US teens said they use AI chatbots such as ChatGPT, Copilot or Character.AI to help with schoolwork, per a Pew Research Center survey conducted September 25 to October 9, 2025 (published 2026). 57% said they use chatbots to search for information. Nationally representative survey of teens.

Source: Pew Research Center · 2026-02

~220M (est.)

Anthropic's Claude is estimated at 200M+ monthly users

Anthropic does not publish a consolidated consumer monthly-active-user number for Claude; it discloses business-customer counts (300,000+ business customers). Third-party estimates put Claude.ai at roughly 220 million monthly active users across web and app in early 2026. Treat this as an estimate, not an official figure.

Source: Third-party estimate (Sensor Tower / analytics) · 2026

~48%

AI Overviews appear on roughly half of US Google searches

AI Overviews appear on an estimated 48-50% of US Google Search queries in 2026, up from under 7% in January 2025, according to third-party SEO tracking. These are external estimates from search-visibility tools, not Google-disclosed figures.

Source: Third-party SEO tracking (estimate) · 2026

800M

ChatGPT reached ~800 million weekly users at OpenAI DevDay 2025

Sam Altman said ChatGPT had about 800 million weekly active users at OpenAI's DevDay in October 2025 — close to 10% of the world's adult population, and double the 400 million OpenAI reported in February 2025. This is OpenAI's own point-in-time disclosure.

Source: OpenAI (Sam Altman, DevDay) · 2025-10

650M+

Google's Gemini app passed 650 million monthly users

Alphabet CEO Sundar Pichai said the Gemini app had more than 650 million monthly active users on the Q3 2025 earnings call (October 2025), up from 350 million disclosed in court filings for March 2025. Google later reported the Gemini app surpassed 750 million monthly users at its Q4 2025 earnings (February 2026). Company-disclosed figures.

Source: Google / Alphabet (Sundar Pichai) · 2025-10

~77%

Three tasks make up about 77% of ChatGPT use

OpenAI's own study of consumer ChatGPT usage (with economists via NBER, September 2025, based on ~1.5 million messages) found three categories account for roughly 77% of all messages: practical guidance (how-to advice, tutoring, ideas), seeking information, and writing. Writing was the single most common work-related use, and about two-thirds of writing requests were edits to existing text rather than generation from scratch.

Source: OpenAI / NBER (Chatterji, Deming et al.) · 2025-09

73%

About 73% of ChatGPT messages are non-work related

OpenAI's usage study found non-work messages rose to about 73% of ChatGPT traffic by mid-2025, up from roughly 53% a year earlier — evidence that consumer, everyday use has outgrown workplace use. Based on OpenAI's own message sample.

Source: OpenAI / NBER · 2025-09

~5.8B/mo

ChatGPT drew about 5.8 billion website visits a month in 2025

ChatGPT.com reached roughly 5.8 billion monthly visits by August 2025, up from about 2.6 billion a year earlier, capturing an estimated 69% of all traffic to AI tools, per Similarweb. Similarweb figures are modelled traffic estimates.

Source: Similarweb · 2025-08

2B

Google AI Overviews reached 2 billion monthly users

Google said its AI Overviews in Search reached 2 billion monthly users as of July 2025, up from 1.5 billion in May 2025, and that its dedicated AI Mode had passed 100 million monthly users in the US and India. Company-disclosed figures (Sundar Pichai, Alphabet earnings).

Source: Google / Alphabet (Sundar Pichai) · 2025-07

$10B ARR

OpenAI passed $10 billion in annualized revenue in mid-2025

OpenAI told CNBC it reached about $10 billion in annualized revenue (ARR) as of June 2025, driven by ChatGPT growth — roughly double the ~$5.5 billion of late 2024. Later reporting put OpenAI's run-rate around $20 billion by the end of 2025, a figure attributed to its CFO but not independently audited.

Source: OpenAI (via CNBC) · 2025-06

34%

34% of US adults had used ChatGPT in early 2025

34% of US adults reported having used ChatGPT in a Pew Research Center survey conducted February 24 to March 2, 2025 — roughly double the share from 2023. Based on a nationally representative survey.

Source: Pew Research Center · 2025-06

58%

58% of US adults under 30 had used ChatGPT in early 2025

58% of US adults under 30 reported having used ChatGPT in Pew Research Center's early-2025 survey, up from 43% in 2024 and 33% in 2023 — the clearest sign that AI adoption skews young. Nationally representative survey.

Source: Pew Research Center · 2025-06

51% vs 18%

ChatGPT use is far higher among college graduates

About half of US adults with a bachelor's degree (51%) or postgraduate degree (52%) had used ChatGPT in Pew's early-2025 survey, compared with 33% of those with some college and just 18% of those with a high school education or less. Nationally representative survey.

Source: Pew Research Center · 2025-06

34%

34% of people worldwide use generative AI weekly

Weekly use of generative AI nearly doubled from 18% in 2024 to 34% in 2025 across markets surveyed, per the Reuters Institute Digital News Report 2025 (survey of ~97,000 people in 48 markets). ChatGPT was the single most-used tool, at 22% weekly. Based on a large cross-country survey.

Source: Reuters Institute Digital News Report 2025 · 2025-06

59% vs 20%

59% of 18-24s used generative AI in the past week vs 20% of over-55s

In the Reuters Institute Digital News Report 2025, 59% of 18-24-year-olds had used any generative AI in the last week, versus 20% of those aged 55 and over — a roughly threefold age gap. Cross-country survey.

Source: Reuters Institute Digital News Report 2025 · 2025-06

~7%

Only about 7% use AI chatbots for news each week

Despite fast-growing general use, only about 7% of people use AI chatbots to get news in a given week (15% among under-25s), and just 4% used ChatGPT specifically for news, per the Reuters Institute Digital News Report 2025 — a gap showing AI adoption is running ahead of trust in AI for news. Cross-country survey.

Source: Reuters Institute Digital News Report 2025 · 2025-06

1B

Meta AI reached 1 billion monthly active users

Mark Zuckerberg said Meta AI passed 1 billion monthly active users, announced at Meta's shareholder meeting in May 2025. Meta AI is distributed inside WhatsApp, Instagram, Facebook and Messenger, so its reach is not directly comparable to a standalone app like ChatGPT. Company-disclosed figure.

Source: Meta (Mark Zuckerberg) · 2025-05

84%

84% of US high schoolers had used generative AI for schoolwork by May 2025

The share of US high school students using generative AI tools for schoolwork rose from 79% to 84% between January and May 2025, with 69% specifically using ChatGPT for assignments, according to College Board research. Based on College Board's student surveys (higher than Pew's teen figures because it measures ever-use among high schoolers specifically).

Source: College Board · 2025-05

~22M DAU

DeepSeek reached an estimated 22 million daily users in January 2025

Chinese assistant DeepSeek surged to an estimated 22 million daily active users in late January 2025 after its R1 model launch, briefly topping the US App Store free chart. This is a third-party analytics estimate, not a company disclosure; DeepSeek does not publish official user counts.

Source: Third-party analytics estimate · 2025-01

26%

26% of US teens use ChatGPT for schoolwork

26% of US teens (ages 13-17) said they use ChatGPT for schoolwork, double the 13% share of 2023, per a Pew Research Center survey published January 2025. Nationally representative survey of teens.

Source: Pew Research Center · 2025-01

#2

ChatGPT was the second most-downloaded app worldwide in 2025

ChatGPT was the second most-downloaded app globally in 2025, behind only TikTok, with downloads up about 148% year over year, per Sensor Tower's State of Mobile 2026 report. Sensor Tower figures are modelled estimates from app-store data.

Source: Sensor Tower (State of Mobile 2026) · 2025

20M+

GitHub Copilot has passed 20 million users

GitHub Copilot surpassed 20 million all-time users by mid-2025, Microsoft CEO Satya Nadella said on the company's FY2025 fourth-quarter earnings call, making AI coding assistants one of the fastest-adopted developer tools ever.

Source: Microsoft / GitHub (FY25 Q4 earnings) · 2025

90%

90% of the Fortune 100 use GitHub Copilot

Around 90% of Fortune 100 companies now use GitHub Copilot, per GitHub/Microsoft, evidence that AI coding assistants have moved from pilots to standard enterprise development infrastructure.

Source: Microsoft / GitHub · 2025

~30%

Up to 30% of Microsoft's code is now AI-generated

Microsoft CEO Satya Nadella said in 2025 that as much as 20-30% of code in some Microsoft repositories is now written by AI, an early signal of how far generative AI is reshaping software development.

Source: Microsoft (Satya Nadella, LlamaCon 2025) · 2025

45%

45% of US workers now use AI on the job

The share of US employees using AI at work at least a few times a year rose from 40% to 45% between Q2 and Q3 2025, with frequent (weekly-plus) use climbing to 23%, according to Gallup's Workforce survey of 23,068 workers.

Source: Gallup · 2025

100M

Google's AI Mode reached 100 million monthly users

Google's conversational AI Mode in Search passed 100 million monthly active users in the US and India by mid-2025, while AI Overviews reached 2 billion monthly users — showing how fast AI answers are spreading through search, per Alphabet's Q2 2025 earnings.

Source: Alphabet / Google (Q2 2025 earnings) · 2025

~2 months

ChatGPT hit 100 million users in about two months

ChatGPT is estimated to have reached 100 million monthly active users by January 2023, roughly two months after its November 2022 launch, which a UBS analyst note (reported by Reuters, citing Similarweb data) called the fastest ramp for a consumer internet app on record. For comparison, TikTok took about nine months and Instagram about 2.5 years. The 100M figure is a third-party estimate (Similarweb/UBS), not an OpenAI disclosure.

Source: UBS / Similarweb (via Reuters) · 2023-02

19.9%

About 20% of EU enterprises used at least one AI technology in 2025

Denmark (42%), Finland (37.8%) and Sweden (35%) lead Europe; the EU27 average is 19.9% (Eurostat).

Source: Eurostat (via Affärslivet) · 2025

$3.5T

The global AI market could hit $3.5 trillion by 2033

In its 2026 update, Grand View Research forecasts the worldwide AI market to reach USD 3.5 trillion by 2033, powered by a 31.5% compound annual growth rate.

Source: Grand View Research · 2026

$324.7B

The generative-AI market is forecast to grow from $22bn in 2025 to $325bn by 2033

Grand View Research values the global generative-AI market at USD 22.2 billion in 2025 and projects it to reach USD 324.68 billion by 2033, a 40.8% CAGR — one of the fastest-growing technology segments tracked.

Source: Grand View Research · 2026

$285.9B

US private AI investment reached $285.9 billion in 2025

Stanford HAI's 2026 AI Index reports US private investment in AI hit USD 285.9 billion in 2025, dwarfing China's USD 12.4 billion and underscoring the concentration of AI capital in the United States.

Source: Stanford HAI — AI Index 2026 · 2026

~20%

Around 1 in 5 US businesses were using AI in early 2026

The US Census Bureau's Business Trends and Outlook Survey found overall AI use among US businesses hovered between 17% and 20% from December 2025 to May 2026, up from 4.6% in production-focused terms in early 2024.

Source: US Census Bureau (Business Trends and Outlook Survey) · 2026

37% vs <20%

37% of large US firms use AI versus under 20% of the smallest firms

US Census Bureau data (through May 2026) show about 37% of firms with 250+ employees used AI, versus under 20% of firms with fewer than 20 employees — and the gap widened over the prior six months.

Source: US Census Bureau (Business Trends and Outlook Survey) · 2026

$1.81T

The global AI market is projected to reach roughly $1.8 trillion by 2030

Grand View Research projects the total artificial-intelligence market (hardware, software and services) to reach USD 1,811.75 billion by 2030, growing at a 36.6% CAGR from 2025.

Source: Grand View Research · 2025

$19.8B

The enterprise generative-AI market is set to reach nearly $20bn by 2030

Grand View Research estimates the enterprise generative-AI market at USD 2.9 billion in 2024, rising to USD 19.8 billion by 2030 at a 38.4% CAGR, with software making up over 72% of revenue.

Source: Grand View Research · 2025

$50.3B

The AI agents market is projected to reach $50bn by 2030

Grand View Research forecasts the market for AI agents (autonomous, task-completing software) to reach USD 50.31 billion by 2030, growing at a 45.8% CAGR — the fastest-growing AI sub-segment it tracks.

Source: Grand View Research · 2025

$252.3B

Global corporate AI investment reached $252.3 billion in 2024

Stanford HAI's 2025 AI Index reports total global corporate investment in AI reached USD 252.3 billion in 2024, with private investment up 44.5% year over year.

Source: Stanford HAI — AI Index 2025 · 2025

$33.9B

Generative AI attracted nearly $34 billion in private investment in 2024

Stanford HAI's 2025 AI Index reports generative AI drew USD 33.9 billion in global private investment in 2024, and the number of newly funded generative-AI startups nearly tripled.

Source: Stanford HAI — AI Index 2025 · 2025

88%

88% of organizations now use AI in at least one business function

McKinsey's State of AI 2025 survey of 1,993 respondents across 105 countries found 88% say their organizations regularly use AI in at least one business function — up sharply from prior years.

Source: McKinsey — The State of AI 2025 · 2025

78%

78% of organizations reported using AI in 2024, up from 55% a year earlier

Stanford HAI's 2025 AI Index found 78% of organizations reported using AI in at least one business function in 2024, up from 55% in 2023 — evidence of a rapid acceleration in adoption.

Source: Stanford HAI — AI Index 2025 · 2025

71%

Enterprise generative-AI use more than doubled to 71% in 2024

Stanford HAI's 2025 AI Index reports the share of organizations using generative AI in at least one business function more than doubled — from 33% in 2023 to 71% in 2024.

Source: Stanford HAI — AI Index 2025 · 2025

20%

20% of EU enterprises used AI in 2025, up from 13.5% in 2024

Eurostat reports 20% of EU enterprises with 10+ employees used AI technologies in 2025, up from 13.5% in 2024 and 8.0% in 2023. Adoption is highest in Denmark, Sweden and Belgium.

Source: Eurostat · 2025

13.5%

In 2024, 13.5% of EU enterprises used AI — but over 27% in Denmark

Eurostat's 2024 data show 13.5% of EU enterprises with 10+ employees used AI, ranging from 27.6% in Denmark and 25.1% in Sweden to 3.1% in Romania — a wide adoption gap across member states.

Source: Eurostat · 2025

10-20%

Software and IT teams report 10-20% cost cuts from AI; marketing sees 10%+ revenue lift

McKinsey's State of AI 2025 found software engineering and IT report 10-20% cost reductions from AI, while marketing and product development report revenue uplift above 10% tied to AI initiatives.

Source: McKinsey — The State of AI 2025 · 2025

33%

33% of enterprise software will include agentic AI by 2028

Gartner predicts 33% of enterprise software applications will include agentic AI by 2028, up from less than 1% in 2024 — a signal of how fast autonomous AI is expected to enter business tools.

Source: Gartner · 2025

15%

By 2028, 15% of day-to-day work decisions will be made autonomously by agentic AI

Gartner predicts at least 15% of day-to-day work decisions will be made autonomously through agentic AI by 2028, up from 0% in 2024.

Source: Gartner · 2025

40%+

Over 40% of agentic AI projects will be scrapped by the end of 2027

Gartner predicts more than 40% of agentic-AI projects will be canceled by the end of 2027 due to escalating costs, unclear business value or inadequate risk controls — a caution against agentic hype.

Source: Gartner · 2025

23%

23% of enterprises are already scaling AI agents in at least one function

McKinsey's State of AI 2025 found 23% of enterprises are scaling AI agents in at least one function — led by IT, knowledge management and engineering — though no more than 10% do so in any single function.

Source: McKinsey — The State of AI 2025 · 2025

$632B

Worldwide spending on AI is forecast to reach $632 billion in 2028

IDC projects total worldwide spending on AI — including AI-enabled applications, infrastructure and related IT/business services — to more than double to USD 632 billion by 2028, a 29% CAGR from 2024.

Source: IDC (Worldwide AI and Generative AI Spending Guide) · 2024

$202B

Generative-AI spending is expected to hit $202 billion by 2028

IDC expects generative-AI spending to reach USD 202 billion by 2028 — about 32% of all AI spending — growing at a 59.2% five-year CAGR, faster than the overall AI market.

Source: IDC (Worldwide AI and Generative AI Spending Guide) · 2024

75%

75% of generative AI's value lands in four business functions

McKinsey estimates about 75% of the value generative AI could deliver falls across four functions: customer operations, marketing and sales, software engineering, and R&D.

Source: McKinsey Global Institute · 2023

$2.6-4.4T

Generative AI could add $2.6-4.4 trillion in value to the global economy annually

McKinsey estimates generative AI could add the equivalent of USD 2.6 trillion to USD 4.4 trillion annually across 63 use cases in 16 business functions — comparable to the entire GDP of the United Kingdom.

Source: McKinsey Global Institute · 2023

$15.7T

AI could add $15.7 trillion to the global economy by 2030

PwC's 'Sizing the Prize' study estimates AI could contribute USD 15.7 trillion to the global economy by 2030 — a 14% GDP boost — split between USD 6.6T from productivity and USD 9.1T from consumption effects.

Source: PwC · 2017

93%

The best AI models now score over 90% on graduate-level science questions (GPQA Diamond)

Leading models exceed the ~65% human-expert benchmark on GPQA Diamond — but capability leadership is split across labs.

Source: Epoch AI Benchmarking Hub (via Affärslivet) · 2026

~32%

Even the best AI models solve only about a third of research-level maths problems (FrontierMath)

FrontierMath, built with mathematicians to resist AI, still humbles every model — hard mathematical reasoning is far from solved.

Source: Epoch AI Benchmarking Hub (via Affärslivet) · 2026

CN > US

China releases far more of its AI models as open-weight than the United States

Openness is a national strategy: China and Europe release a much higher share of their models with open weights than the US.

Source: Affärslivet Open-Source AI Index · 2026

1,040

Epoch AI tracks more than 1,000 notable AI models ever built

Epoch AI's Notable AI Models database listed 1,040 models judged historically significant across the whole history of the field, of which 585 were released since 2020. It is the most complete public census of consequential models.

Source: Epoch AI — Notable AI Models · 2026-07

123

About 123 models qualify as frontier by Epoch's compute bar

Epoch AI classes 123 of its catalogued models as 'frontier' — trained with compute within roughly a factor of ten of the largest run at their time. The US accounts for 36 of them, China 3.

Source: Epoch AI — Notable AI Models · 2026-07

262 open / 313 closed

Closed models still outnumber open-weight ones since 2020

Of models released since 2020, Epoch AI counts 262 with open weights against 313 closed. Open-weight releases are a large minority of significant models rather than the majority.

Source: Epoch AI — Notable AI Models · 2026-07

386 vs 149

The US produced far more notable models than China

The United States accounts for 386 of Epoch's catalogued notable models and China 149 — the two dominant producers. China's output skews more open-weight (59% of its models) than America's (40%).

Source: Epoch AI — Notable AI Models · 2026-07

338 problems

FrontierMath is the benchmark built to stay hard for AI

Epoch AI's FrontierMath comprises 338 original research-level maths problems (43 in the hardest Tier 4), designed to resist the saturation that has overtaken older tests like MMLU as models improve.

Source: Epoch AI — Benchmarks · 2026-06

87%

Top models now score around 87% on the hard GPQA science benchmark

On Epoch AI's evaluation of GPQA Diamond — graduate-level science questions where PhD experts score ~65-74% — the leading result was 87% (Grok 4, July 2025), showing benchmark saturation at the frontier.

Source: Epoch AI — Benchmarks · 2025-07

10M tokens

The largest published context window is 10 million tokens

Meta's open-weight Llama 4 Scout advertises a 10-million-token context window (~15,000 pages), ahead of Gemini's 2 million — though independent tests show reliable reasoning degrading well before those limits.

Source: Meta (Llama 4 announcement) · 2025-04

~7 months

The task length AI can do solo is doubling about every 7 months

METR finds the length of software tasks frontier agents can complete unsupervised at 50% reliability — measured in human hours — has doubled roughly every 7 months for six years, a key gauge of agentic capability.

Source: METR · 2025-03

8.0% to 1.7%

Open-weight models nearly closed the gap with closed ones

The Stanford AI Index reports the leading closed model's edge over the best open-weight model on Chatbot Arena shrank from 8.0% (Jan 2024) to 1.7% (Feb 2025) — open models are now roughly at parity.

Source: Stanford HAI AI Index 2025 · 2025

40 vs 15

US labs shipped 40 notable models in 2024, China 15

In 2024 US organisations released 40 notable models, China 15 and Europe just 3, per the Stanford AI Index — a snapshot of where frontier model production concentrates.

Source: Stanford HAI AI Index 2025 · 2024

90%

Nearly all notable models now come from industry, not academia

90% of notable models in 2024 came from industry, up from 60% in 2023, while academia produced none, per the Stanford AI Index — the compute cost of frontier training has pushed universities out.

Source: Stanford HAI AI Index 2025 · 2024

4.4% to 71.7%

AI coding scores leapt from 4% to 72% on SWE-bench in one year

The Stanford AI Index records SWE-bench Verified performance rising from 4.4% of real software issues solved in 2023 to 71.7% in 2024 — one of the fastest capability jumps on any benchmark.

Source: Stanford HAI AI Index 2025 · 2024

~75%

The United States holds roughly three-quarters of the world's AI computing power

On Epoch AI's adjusted estimate the US commands about 75% of global AI compute and several times China's (roughly 5×); raw tracked-cluster sums, which under-count undisclosed Chinese compute, imply an even larger gap.

Source: Epoch AI / Affärslivet AI Compute Index · 2026

92%

The top 5 economies hold about 92% of the world's AI compute

AI compute is one of the most concentrated resources in technology — an HHI far above the 'highly concentrated' threshold.

Source: Affärslivet AI Compute Index · 2026

~70%

Nvidia has shipped about 70% of all AI compute ever deployed

By cumulative H100-equivalents shipped, Nvidia dominates AI accelerators — more than every rival combined.

Source: Epoch AI (via Affärslivet) · 2026

946 MW

The world's largest AI data centre draws about 946 MW of power

xAI's Colossus 2 is the largest tracked AI data centre by power — enough for a mid-sized city.

Source: Epoch AI (via Affärslivet) · 2026

~70%

TSMC manufactures about 70% of the world's chips and over 90% of leading-edge AI chips

TSMC is the single manufacturing chokepoint of the AI supply chain; ASML holds a monopoly on the EUV machines it depends on.

Source: Affärslivet — AI Chips & Semiconductors · 2025

~1.11M H100e

xAI's Colossus 2 is the largest AI data centre by installed compute

Epoch AI's data-centre tracker puts xAI's Colossus 2 in Memphis at about 1.11 million H100-equivalent chips drawing roughly 946 MW — the single largest AI compute site it tracks.

Source: Epoch AI — Data Centers · 2026-07

18.1M H100e

Nvidia accounts for over 18 million H100-equivalents of AI compute

Across the clusters Epoch AI tracks, Nvidia-designed chips total about 18.1 million H100-equivalents, dwarfing Google's TPUs (~4.8M) and AMD (~1.3M) — a measure of Nvidia's grip on AI compute.

Source: Epoch AI · 2026-07

~120-130K wpm

TSMC is racing to expand the CoWoS packaging AI chips depend on

TSMC's CoWoS advanced-packaging capacity — the bottleneck for Nvidia/AMD AI accelerators — is set to roughly double from ~75-80k wafers/month in late 2025 toward 120-130k by end-2026, per TrendForce.

Source: TrendForce · 2026-06

~20%

HBM memory is so scarce that suppliers are raising prices ~20%

TrendForce reports Samsung and SK hynix plan ~20% HBM3E price hikes for 2026 as AI-accelerator demand outstrips supply — the high-bandwidth memory stacked beside every leading GPU is now a gating resource.

Source: TrendForce · 2025-12

~$490M

xAI's Grok 4 training run cost an estimated $490 million

Epoch AI estimates Grok 4 took about 246 million H100-hours at xAI's Memphis Colossus site, a median cost near $490 million — among the most expensive disclosed training runs to date.

Source: Epoch AI · 2025-09

275,796 H100e

xAI's Colossus Memphis cluster reached ~276,000 H100-equivalents

Epoch AI records xAI's Colossus Memphis Phase 3 cluster at 275,796 H100-equivalent chips and 352 MW — the largest single interconnected training cluster in its database.

Source: Epoch AI — GPU Clusters · 2025-07

33

At least 33 models have been trained at GPT-4 scale or beyond

As of June 2025 Epoch AI counted 33 publicly disclosed models trained with at least 10^25 FLOP, from 12 different developers — averaging roughly two new GPT-4-scale models per month through 2024.

Source: Epoch AI · 2025-06

4.6e26 FLOP

Grok-3 is the largest known training run at about 4.6e26 FLOP

Epoch AI estimates xAI's Grok-3 (Feb 2025) used about 4.6 x 10^26 FLOP, the most compute of any identified model and the first believed to exceed 10^26 FLOP.

Source: Epoch AI · 2025-06

$115.2B

Nvidia's data-centre revenue hit a record $115 billion in FY2025

Nvidia reported record full-year data-centre revenue of $115.2 billion for fiscal 2025, up 142%, on demand for its Hopper GPUs used in LLM training and inference.

Source: Nvidia (Q4/FY2025 results) · 2025-02

~2x/yr

The power needed to train frontier models is doubling every year

Epoch AI finds the electrical power required for the largest training runs has been doubling annually, driven mainly by ever-larger GPU clusters — a key reason AI is now a grid-planning problem.

Source: Epoch AI · 2025

$791.7B

Global chip sales hit a record $791.7 billion in 2025

The Semiconductor Industry Association reports global chip sales rose 25.6% to a record $791.7 billion in 2025, led by AI: logic chips alone grew 39.9% to $301.9 billion. SIA projects ~$1 trillion in 2026.

Source: Semiconductor Industry Association (SIA) · 2025

100%

ASML is the sole maker of the EUV machines that print AI chips

Every leading-edge AI chip is patterned on ASML's extreme-ultraviolet lithography tools, of which it is the only supplier worldwide; ASML posted about EUR 32.7 billion in 2025 revenue on surging AI demand.

Source: ASML / Counterpoint Research · 2025

4.1x/yr

AI training compute has grown roughly 4x per year since 2010

The compute used to train notable AI models grew about 4.1x per year (90% CI 3.7-4.6x) from 2010 to 2024, per Epoch AI — an order of magnitude every ~1.5 years, far faster than Moore's Law.

Source: Epoch AI · 2024

5x/yr

Frontier-model training compute doubles about every five months

For frontier language models specifically, training compute has grown about 5x per year since 2020 — a doubling roughly every 5.2 months, per Epoch AI.

Source: Epoch AI · 2024

~$191M

Google's Gemini Ultra cost roughly $191 million to train

Epoch AI estimates the compute for Gemini Ultra's training run at about $191 million, the priciest single model in the Stanford AI Index cost table.

Source: Epoch AI (via Stanford HAI AI Index 2025) · 2024

~$170M

Meta's Llama 3.1 405B cost about $170 million to train

Epoch AI puts the training compute for Meta's open-weight Llama 3.1 405B at roughly $170 million — showing frontier-scale spend is not limited to closed models.

Source: Epoch AI (via Stanford HAI AI Index 2025) · 2024

2.4x/yr

The cost of the largest training runs grows about 2.4x a year

Epoch AI finds the amortised hardware and energy cost of frontier final training runs has grown about 2.4x per year since 2016, and projects the largest runs will exceed $1 billion by 2027.

Source: Epoch AI · 2024

280x cheaper

The cost of AI inference fell over 280-fold in about 18 months

The Stanford AI Index reports inference cost for GPT-3.5-level performance fell from $20 per million tokens in Nov 2022 to $0.07 by Oct 2024 — a ~280x reduction as training costs rose.

Source: Stanford HAI AI Index 2025 · 2024

~$78M

Training GPT-4 cost an estimated $78 million in compute

Epoch AI's amortised-hardware estimate, reproduced in the Stanford AI Index, puts GPT-4's final training run at about $78 million in compute. Sam Altman has separately said total cost exceeded $100M.

Source: Epoch AI (via Stanford HAI AI Index 2025) · 2023

$183B

OpenAI is the most-funded AI company, having raised about $183 billion

OpenAI has raised more equity than any AI company; Anthropic is second.

Source: Epoch AI (via Affärslivet) · 2026

2 labs

Just two labs — OpenAI and Anthropic — hold most of all AI-lab funding

Frontier-AI capital is astonishingly concentrated in a handful of US labs.

Source: Affärslivet — The Largest AI Companies · 2026

2x all of 2025

Foundational-AI startup funding in Q1 2026 doubled all of 2025

Venture funding to foundational AI startups (OpenAI, Anthropic, xAI and peers) in Q1 2026 was roughly double the total raised across all of 2025, per Crunchbase.

Source: Crunchbase · 2026-Q1

$965B

Anthropic reached a $965B valuation in May 2026, overtaking OpenAI

Anthropic announced roughly $65 billion in fresh funding at a $965 billion valuation on May 28, 2026, briefly making it the most valuable AI startup ahead of OpenAI.

Source: CNBC · 2026-05-28

$122B

OpenAI raised $122B at an $852B valuation in early 2026

OpenAI's reported $122 billion round closed around March 31, 2026 at an $852 billion post-money valuation, up from $500 billion in October 2025. Flag: sourced to VC reporting, not an OpenAI press release.

Source: Value Add VC · 2026-03

$230B

xAI reached a ~$230B valuation in January 2026

xAI raised a roughly $20 billion Series E lifting its standalone valuation to about $230 billion in January 2026, before SpaceX moved to fold it into a combined entity in early 2026. Flag: valuation sourced to secondary reporting.

Source: Secondary reporting (BearBull / SentiSight) · 2026-01

$725B

Big Tech capex is projected to hit $725B in 2026, up 77%

Alphabet, Amazon, Microsoft and Meta collectively guided toward about $725 billion of capex in 2026 — up roughly 77% from 2025 — with most earmarked for AI infrastructure (Amazon ~$200B, Alphabet ~$175-185B, Meta ~$115-135B, Microsoft ~$120B+).

Source: Statista / Tom's Hardware · 2026

$500B

OpenAI hit a $500B valuation in October 2025

A secondary share sale in October 2025 valued OpenAI at about $500 billion, up from $300 billion in March 2025.

Source: Value Add VC (secondary-sale reporting) · 2025-10

$183B

Anthropic raised $13B at a $183B valuation in September 2025

Anthropic announced a $13 billion Series F on September 2, 2025 at a $183 billion post-money valuation, up from $61.5 billion in March 2025.

Source: Anthropic · 2025-09-02

$14B

Mistral AI hit a $14B (EUR 12B) valuation in September 2025

France's Mistral AI raised about EUR 2 billion in a Series C led by ASML in September 2025, valuing the company at EUR 12 billion (roughly $14 billion) — Europe's most valuable AI lab.

Source: Bloomberg · 2025-09

498

There were 498 AI unicorns worth $2.7T by mid-2025

As of August 2025 there were 498 private AI companies valued at $1 billion or more, worth a combined $2.7 trillion, per CB Insights.

Source: CB Insights (via Fortune) · 2025-08

$14.3B

Meta invested $14.3B in Scale AI in June 2025

Meta invested $14.3 billion for a 49% non-voting stake in Scale AI in June 2025, hiring founder Alexandr Wang to lead a superintelligence team — one of the largest AI deals on record.

Source: CNBC · 2025-06

177 deals

AI M&A hit a record 177 deals in Q2 2025

AI merger and acquisition activity reached a quarterly record of 177 deals in Q2 2025, roughly double the quarterly average since 2020, per CB Insights.

Source: CB Insights · 2025-06

$32B

Safe Superintelligence hit a $32B valuation in April 2025

Ilya Sutskever's Safe Superintelligence raised about $2 billion in April 2025 at a $32 billion valuation, despite having no public product or revenue — one of the highest ever pre-product valuations.

Source: TechCrunch · 2025-04

$40B

OpenAI raised $40B in March 2025 — the largest private tech round ever

OpenAI closed a $40 billion round led by SoftBank on March 31, 2025 at a $300 billion post-money valuation — the largest private tech fundraise on record per PitchBook.

Source: OpenAI / CNBC · 2025-03-31

$581.7B

Global corporate AI investment reached $581.7B in 2025

Total global corporate investment in AI (private investment, M&A, minority stakes and public offerings) hit $581.7 billion in 2025, up roughly 130% from about $253 billion in 2024, per the Stanford HAI AI Index 2026.

Source: Stanford HAI AI Index 2026 · 2025

$344.7B

Global private AI investment reached $344.7B in 2025

Private AI investment (venture and growth funding to AI startups) totalled $344.7 billion globally in 2025, up about 127.5% year over year and roughly 60% of all corporate AI investment, per the Stanford HAI AI Index 2026.

Source: Stanford HAI AI Index 2026 · 2025

$12.4B

China's private AI investment was $12.4B in 2025

China recorded $12.4 billion in private AI investment in 2025 per the Stanford HAI AI Index 2026, which cautions that private figures understate China because state guidance funds channel large sums outside standard venture tracking.

Source: Stanford HAI AI Index 2026 · 2025

23x

US private AI investment was 23x China's in 2025

US private AI investment ($285.9B) was more than 23 times China's ($12.4B) in 2025, per the Stanford HAI AI Index 2026 — a gap of roughly $273 billion, though the report notes China's true spend is higher via government funds.

Source: Stanford HAI AI Index 2026 · 2025

>200% growth

Generative AI investment grew over 200% and took nearly half of private AI funding in 2025

Private investment in generative AI grew more than 200% in 2025 and captured close to half of all private AI funding, per the Stanford HAI AI Index 2026 (generative AI drew $33.9 billion globally in 2024 for reference).

Source: Stanford HAI AI Index 2026 · 2025

1,953

The US had 1,953 newly funded AI companies in 2025

1,953 US AI companies received new funding in 2025 — more than 10 times the next-closest country — per the Stanford HAI AI Index 2026.

Source: Stanford HAI AI Index 2026 · 2025

~50%

AI took about half of all global venture funding in 2025

Roughly half of all global venture funding in 2025 went to AI-related companies, per Crunchbase — the highest concentration of capital in a single sector on record.

Source: Crunchbase (via Venture Capital Journal) · 2025

$211B

Global venture funding to AI reached $211B in 2025

Venture funding to AI companies reached about $211 billion in 2025, up roughly 85% year over year from about $114 billion in 2024, per Crunchbase.

Source: Crunchbase · 2025

53%

AI startups took 53% of global VC dollars in H1 2025

AI startups received 53% of all global venture capital dollars in the first half of 2025 — about $104 billion of $205 billion total — per PitchBook.

Source: PitchBook (via SiliconANGLE) · 2025

$381B

Big Tech spent about $381B on capex in 2025

Microsoft, Alphabet, Amazon and Meta together spent roughly $381 billion in capital expenditure in 2025, the bulk of it on AI data centers, chips and networking.

Source: CNBC · 2025

~25%

1 in 4 jobs worldwide is exposed to generative AI

A refined ILO–NASK global index finds roughly a quarter of jobs worldwide are potentially exposed to generative AI. The ILO stresses transformation, not replacement, is the most likely outcome — most occupations mix automatable tasks with tasks that still require human judgment.

Source: ILO / NASK (Generative AI and Jobs: A Refined Global Index of Occupational Exposure, Working Paper 140) · 2025

18.3% vs 5.5%

In high-income countries, ~18% of jobs face augmentation vs ~6% automation

The ILO splits exposure into augmentation and automation. In high-income countries about 18.3% of employment could see significant augmentation (AI assists the worker) while only about 5.5% is potentially exposed to automation (AI does most of the task) — evidence the augmentation case dominates.

Source: ILO / NASK (Generative AI and Jobs, Working Paper 140) · 2025

Clerical roles

Clerical work is the most exposed occupation to automation

The ILO index identifies clerical and administrative support roles as the most automation-exposed occupational group, and finds women face nearly double the highest-level exposure of men — partly because clerical work is female-dominated in many economies. Manual, physical and care roles remain least exposed.

Source: ILO / NASK (Generative AI and Jobs, Working Paper 140) · 2025

170M

170 million new jobs projected to be created by 2030

The WEF Future of Jobs Report 2025 forecasts that technology, the green transition, demographics and economic shifts will create about 170 million new roles by 2030, based on a survey of over 1,000 employers across 55 economies. This is an employer-expectation forecast, not observed data.

Source: World Economic Forum (Future of Jobs Report 2025) · 2025

92M

92 million jobs projected to be displaced by 2030

Against 170 million new roles, the WEF projects about 92 million existing jobs will be displaced by 2030 as tasks are automated and business models shift. Total structural job churn is estimated at 22% of today's jobs by 2030.

Source: World Economic Forum (Future of Jobs Report 2025) · 2025

+78M

Net gain of 78 million jobs projected by 2030

The WEF's central forecast is a net increase of 78 million jobs by 2030 (170 million created minus 92 million displaced). The headline is positive, but the report stresses the transition depends heavily on large-scale reskilling to move displaced workers into growing roles.

Source: World Economic Forum (Future of Jobs Report 2025) · 2025

39%

39% of workers' core skills will change by 2030

Employers expect 39% of the key skills workers need will change by 2030 — a large reskilling burden, though notably down from the 44% projected in the WEF's 2023 report, suggesting some stabilization as AI skills diffuse. This is an employer-expectation forecast.

Source: World Economic Forum (Future of Jobs Report 2025, Skills Outlook) · 2025

77%

77% of employers plan to reskill and upskill workers for AI

77% of surveyed employers plan to reskill and upskill their existing workforce between 2025 and 2030 to work alongside AI, and 85% say they will prioritize upskilling. Framed as a workforce complement to automation of about 59 of every 100 workers needing training by 2030.

Source: World Economic Forum (Future of Jobs Report 2025) · 2025

41%

41% of employers expect to reduce headcount where AI can automate tasks

41% of employers surveyed by the WEF anticipate reducing their workforce by 2030 in areas where AI can automate tasks. This sits alongside the same survey's net-positive jobs forecast — the two coexist because AI displaces some roles while other roles and firms grow.

Source: World Economic Forum (Future of Jobs Report 2025) · 2025

27% (from 7%)

Productivity growth nearly quadrupled in the most AI-exposed industries

PwC's analysis of nearly a billion job ads found productivity growth in the industries most exposed to AI rose from 7% (2018-2022) to 27% (2018-2024), while the least-exposed industries stagnated near 9%. Correlational evidence that AI-intensive sectors are pulling ahead.

Source: PwC (2025 Global AI Jobs Barometer) · 2025

+56%

Jobs requiring AI skills carry a 56% wage premium

PwC found roles demanding AI skills pay a 56% wage premium over otherwise-similar roles, up from 25% a year earlier — and jobs are still growing even in highly automatable occupations. Points to augmentation and rising returns to AI-complementary skills rather than pure displacement.

Source: PwC (2025 Global AI Jobs Barometer) · 2025

78%

78% of organizations reported using AI in 2024

The Stanford HAI AI Index 2025 reports 78% of organizations used AI in 2024, up from 55% the year before, with generative AI adoption rising fastest. The 2026 Index puts organizational adoption at about 88%, with generative AI in at least one business function at roughly 70% of organizations.

Source: Stanford HAI (AI Index Report 2025, Economy chapter) · 2025

~39%

About 39% of employed workers report using AI on the job

A November 2025 Federal Reserve survey found 39% of currently employed respondents had used AI tools in their job in the past 12 months; Gallup's Q3 2025 data similarly put US workers using AI at least a few times a year at 45%. Daily use remains far lower — PwC's survey found only about 14% use generative AI daily.

Source: Federal Reserve Bank of New York (Liberty Street Economics, Survey of Consumer Expectations) · 2025

57% vs 43%

57% of AI work tasks are augmentation, 43% automation

Analyzing millions of anonymized Claude conversations, the Anthropic Economic Index found 57% of task interactions were augmentative (AI collaborates with the human) versus 43% automative (AI performs the task directly) — real-world usage tilting toward augmentation. Use concentrates in mid-to-high-wage knowledge work.

Source: Anthropic (Anthropic Economic Index) · 2025

~40%

AI could affect ~40% of jobs worldwide

The IMF estimates that about 40% of jobs globally are exposed to AI. Exposure means AI can perform or reshape some of the job's tasks — not that the job disappears. Roughly half of exposed jobs may benefit from AI-driven productivity, the other half see tasks substituted.

Source: IMF (Gen-AI: Artificial Intelligence and the Future of Work, Staff Discussion Note) · 2024

~60%

~60% of jobs in advanced economies are exposed to AI

Advanced economies face higher exposure because of their large share of cognitive, high-skill occupations. The IMF projects roughly 60% of jobs in advanced economies are exposed — about half gaining from AI augmentation, half facing task substitution and possible wage or hiring pressure.

Source: IMF (Gen-AI: Artificial Intelligence and the Future of Work) · 2024

~26%

Emerging markets ~40% and low-income economies ~26% exposed

Exposure is lower where employment is concentrated in manual and agricultural work. The IMF puts emerging-market exposure near 40% and low-income-country exposure around 26%, meaning less near-term disruption but also a risk of falling behind on AI-driven productivity gains.

Source: IMF (Gen-AI: Artificial Intelligence and the Future of Work) · 2024

300M

AI could expose the equivalent of 300 million full-time jobs to automation

Goldman Sachs estimated that generative AI could expose the equivalent of 300 million full-time jobs across the US and Europe to some degree of automation. This is a forecast of task exposure, not net job losses — Goldman also argues automation historically spawns new roles that offset displacement.

Source: Goldman Sachs (The Potentially Large Effects of Artificial Intelligence on Economic Growth, Briggs & Kodnani) · 2023

~66%

Two-thirds of US and European jobs are exposed to some AI automation

Goldman Sachs found about two-thirds of current US and European jobs are exposed to some degree of AI automation, and that of those exposed roughly a quarter to a half of the workload could be automated. Exposure is uneven — office and administrative work is far more exposed than physically intensive roles.

Source: Goldman Sachs (The Potentially Large Effects of Artificial Intelligence on Economic Growth) · 2023

+7%

Generative AI could raise global GDP by ~7% over a decade

Goldman Sachs projects widespread generative-AI adoption could raise annual global GDP by about 7% — nearly $7 trillion — and lift productivity growth over a roughly 10-year period. This is a long-run forecast contingent on broad adoption, not a realized figure.

Source: Goldman Sachs (Generative AI could raise global GDP by 7%) · 2023

+14%

Generative AI raised call-center productivity by 14% on average

A study of 5,179 customer-support agents found access to a generative-AI assistant raised issues resolved per hour by 14% on average — and by 34% for novice and lower-skilled workers, with little effect on the most experienced. Evidence that AI can augment and level up less-experienced staff.

Source: Brynjolfsson, Li & Raymond, Generative AI at Work (NBER / Quarterly Journal of Economics) · 2023

+34%

Novice workers gained 34% in productivity from AI assistance

In the same call-center study, the least-experienced agents saw a 34% productivity gain while top performers saw little change — because the AI spread the best workers' tactics to everyone. This skill-compression effect is a recurring finding in AI-at-work studies.

Source: Brynjolfsson, Li & Raymond, Generative AI at Work (NBER / QJE) · 2023

55% faster

Developers with GitHub Copilot completed a coding task 55% faster

In a controlled experiment, developers using GitHub Copilot finished a programming task 55.8% faster than a control group (about 1h11m vs 2h41m), with a wide 95% confidence interval of 21-89%. A task-level lab result; real-world gains across full workflows are typically smaller.

Source: GitHub / Microsoft Research (The Impact of AI on Developer Productivity: Evidence from GitHub Copilot) · 2023

$2.6-4.4T

Generative AI could add $2.6-4.4 trillion in value annually

McKinsey estimates 63 generative-AI use cases across 16 business functions could add $2.6 trillion to $4.4 trillion in value annually across the global economy. A modeled potential, dependent on adoption pace, not a realized figure.

Source: McKinsey Global Institute (The economic potential of generative AI) · 2023

60-70%

AI could automate work absorbing 60-70% of employees' time

McKinsey finds generative AI plus other technologies have the theoretical potential to automate activities that currently take up 60-70% of employees' working time — up from earlier estimates, largely because language models can handle knowledge work. Technical potential, not a prediction of actual displacement.

Source: McKinsey Global Institute (The economic potential of generative AI) · 2023

~945 TWh

Data-centre power use is projected to double by 2030

Projection: the IEA base case sees data-centre electricity demand more than doubling to around 945 TWh by 2030 — just under 3% of global electricity. Growth averages ~15% per year over 2024–2030.

Source: IEA, Energy and AI (2025) — projection · 2030

~30%/yr

AI servers drive nearly half of data-centre demand growth

Projection: electricity use by accelerated (mainly AI) servers is projected to grow about 30% per year and account for almost half of the net increase in global data-centre electricity consumption to 2030.

Source: IEA, Energy and AI (2025) — projection · 2030

310 GWh

Training Grok 4 used about 310 gigawatt-hours of electricity

Epoch AI estimates Grok 4's training consumed roughly 310 GWh — enough to power a town of about 4,000 US homes for a year — plus around 750 million litres of water for cooling.

Source: Epoch AI · 2025-09

415 TWh

Data centres used about 415 TWh of electricity in 2024

The IEA estimates data centres consumed around 415 TWh in 2024, roughly 1.5% of global electricity demand.

Source: IEA, Energy and AI (2025) · 2024

180 Mt

Data centres emitted about 180 Mt of CO2 in 2024

The IEA estimates data centres produced around 180 million tonnes of CO2 in 2024, rising to about 300 Mt by 2035 in the base case (up to 500 Mt in a high-growth case).

Source: IEA, Energy and AI (2025) · 2024

#1 US

The United States ranks #1 on overall national AI power; Sweden ranks #12

Across seven dimensions — models, compute, research, R&D, talent, high-tech and digital adoption — the US leads 49 economies.

Source: Affärslivet National AI Power Index · 2026

US vs China

The US produces the most notable and frontier AI models; China leads on volume of research

China publishes more and releases more open-weight models, but the US ships more frontier systems.

Source: Affärslivet — Countries Leading in AI · 2026

0 frontier

The Nordics lead Europe on enterprise AI adoption but build almost no frontier models

Denmark, Finland and Sweden top EU AI adoption, yet the five Nordic countries have built zero frontier AI models.

Source: Affärslivet — Sweden & Nordic AI · 2026

116k

China leads the world in AI research output with about 116,000 publications

China publishes the most AI research, ahead of India and the US — though the US builds more frontier models.

Source: Our World in Data / OpenAlex (via Affärslivet) · 2025

$665M

AMD acquired Finland's Silo AI — Europe's largest private AI lab — for about $665 million

The 2024 all-cash deal was the largest European private-AI-lab acquisition, underscoring Nordic deep-tech AI strength.

Source: AMD (via Affärslivet) · 2024-07

2.7%

The top US model now leads China's best by just 2.7%

The performance gap between the leading US and Chinese models has effectively closed: as of March 2026 the top US model led the top Chinese model by only 2.7%, down from double-digit margins a year earlier, per the Stanford AI Index 2026.

Source: Stanford HAI AI Index 2026 · 2026

50

The United States released the most notable AI models in 2025

US-based organisations released 50 notable AI models in 2025, per the Stanford HAI AI Index 2026, keeping the United States ahead of every other country in frontier model output.

Source: Stanford HAI AI Index 2026 · 2025

35

China released 35 notable AI models in 2025, closing the gap on the US

China produced 35 notable AI models in 2025 versus 50 for the United States, according to the Stanford AI Index 2026 — a narrowing gap after US and Chinese systems traded the performance lead repeatedly through the year.

Source: Stanford HAI AI Index 2026 · 2025

2

Europe produced just 2 notable AI models in 2025

Europe accounted for only 2 notable AI models in 2025, dwarfed by the US (50) and China (35), underscoring how far the frontier race has consolidated into two countries, per the Stanford AI Index 2026.

Source: Stanford HAI AI Index 2026 · 2025

#1

China leads the world in AI publication volume and citations

China leads globally in AI publication volume, citations and patent grants, according to the Stanford AI Index 2026, while the United States still produces more of the highest-impact, most-cited work.

Source: Stanford HAI AI Index 2026 · 2025

#1

South Korea leads the world in AI patents per capita

While China leads in total AI patent grants, South Korea has the highest number of AI patents per capita, according to the Stanford AI Index 2026 — a reminder that the patent race looks different once population is accounted for.

Source: Stanford HAI AI Index 2026 · 2025

>50%

China installs more industrial robots than the rest of the world combined

China now accounts for more than half of all annual industrial-robot installations worldwide — more than the rest of the world combined — extending its lead in physical AI and automation, per the Stanford AI Index 2026.

Source: Stanford HAI AI Index 2026 · 2025

23x

US private AI investment is 23 times China's

US private AI investment reached $285.9 billion in 2025 — more than 23 times China's $12.4 billion — per the Stanford AI Index 2026, though the figure understates China's total once government guidance funds are counted.

Source: Stanford HAI AI Index 2026 · 2025

$471B

The US has raised more AI capital than the rest of the world combined

Over the 12 years to 2025, US firms raised $471 billion in private AI funding — more than the rest of the world combined — ahead of China ($119 billion) and the UK ($28 billion), according to the Stanford AI Index 2026.

Source: Stanford HAI AI Index 2026 · 2025

1,953

The US funded more than 10 times as many new AI companies as any rival

The United States saw 1,953 newly funded AI companies in 2025 — more than ten times the next-closest country — underlining the depth of its startup ecosystem, per the Stanford AI Index 2026.

Source: Stanford HAI AI Index 2026 · 2025

$17.5B

Microsoft alone committed $17.5bn to AI in India

Microsoft announced a $17.5 billion investment in India in December 2025 to build AI and cloud infrastructure and drive population-scale AI diffusion, one of several multibillion-dollar commitments (Google, AWS) targeting India as a strategic AI market.

Source: Microsoft · 2025

60+

Over 60 countries have adopted a national AI strategy

More than 60 countries have launched national AI strategies, and OECD.AI's live policy repository tracks over 1,000 AI policy initiatives across 70+ jurisdictions and the EU.

Source: OECD.AI · 2025

41 of 100

China now authors 41 of the 100 most-cited AI papers

China's share of the 100 most-cited AI papers rose from 33 in 2021 to 41 in 2024, per the Stanford AI Index 2026, evidence that Chinese research is gaining influence, not just volume.

Source: Stanford HAI AI Index 2026 · 2024

≈12%

India has become the leading Asian contributor to AI research after China

China leads global AI publications with roughly 36% of the total, followed by India (~12%) and the United States (~12%), according to a comparative analysis published in Human Behavior and Emerging Technologies (2024).

Source: Al-Marzouqi et al. (2024), Human Behavior and Emerging Technologies · 2024

38,000+

China has filed six times more generative-AI patents than the US

China-based inventors filed more than 38,000 generative-AI patent inventions between 2014 and 2023 — roughly six times the United States, the second-placed country — per WIPO's Patent Landscape Report on Generative AI. Tencent, Ping An and Baidu top the list of applicants.

Source: WIPO (Patent Landscape Report: Generative AI) · 2024

1,889

AI mentions in legislatures rose 21% across 75 countries

Across 75 countries, AI mentions in legislative proceedings grew 21.3% in 2024 — from 1,557 to 1,889 — signalling how quickly AI has climbed the policy agenda worldwide, per the Stanford AI Index 2026.

Source: Stanford HAI AI Index 2026 · 2024

45%

45% of consumers now use AI during their shopping journey

An IBM–NRF global consumer study found almost half (45%) of shoppers turn to AI for help during their buying journey — for product research, comparisons and recommendations — reshaping how decisions are made before shoppers reach a retailer.

Source: IBM / NRF · 2026

1,200+

The FDA has authorized more than 1,200 AI-enabled medical devices

The US Food and Drug Administration's official list of AI/ML-enabled medical devices passed roughly 950 cleared devices in August 2024 (as tracked by the Stanford HAI AI Index) and exceeded 1,200 through 2025, up from just six in 2015. Radiology accounts for roughly three-quarters of all authorizations.

Source: US FDA / Stanford HAI AI Index · 2025

+98 mL

The first AI-designed drug improved lung function in a Phase IIa trial

Insilico Medicine's rentosertib — a TNIK inhibitor whose target and molecule were designed with generative AI — improved mean forced vital capacity by 98.4 mL at 60 mg versus a 20.3 mL decline on placebo in a 71-patient Phase IIa idiopathic-pulmonary-fibrosis trial. Results were published in Nature Medicine (June 2025).

Source: Nature Medicine / Insilico Medicine · 2025

240 hrs

AI is set to free up around 240 hours a year for lawyers

Thomson Reuters' Future of Professionals 2025 report projects AI will free roughly 240 hours per legal professional per year (up from 200 in 2024) — about five hours a week — worth an estimated $19,000 each in reclaimed time.

Source: Thomson Reuters Institute · 2025

80%

80% of legal and tax professionals expect AI to be transformational

In Thomson Reuters' Future of Professionals 2025 survey, 80% of respondents said generative AI would have a high or transformational impact on their work over the next five years, yet only 22% said their organisation had a visible, defined AI strategy.

Source: Thomson Reuters Institute · 2025

189

189 factories make up the WEF Global Lighthouse Network

The World Economic Forum's Global Lighthouse Network — factories recognised for deploying AI, IoT and advanced analytics at scale — grew to 189 sites across more than 30 countries in January 2025, deploying over 1,000 use cases with documented productivity and sustainability gains.

Source: World Economic Forum · 2025

88%

Around 9 in 10 UK students use generative AI for their studies

The Higher Education Policy Institute's 2025 student survey found 88% of UK undergraduates use generative-AI tools such as ChatGPT for assessments, up from 53% a year earlier — one of the fastest technology-adoption curves ever recorded in education.

Source: Higher Education Policy Institute (HEPI) · 2025

60%

Six in ten US teachers used AI in the 2024-25 school year

A Gallup–Walton Family Foundation survey of 2,232 US public-school teachers found 60% used an AI tool during the 2024-25 school year; those using AI weekly saved an average of 5.9 hours per week — roughly six weeks over the year.

Source: Gallup / Walton Family Foundation · 2025

2/3

Two-thirds of countries offer or plan K-12 computer-science education

The Stanford HAI AI Index 2025 reports two-thirds of countries now offer or plan to offer K-12 computer-science education — twice as many as in 2019 — though in the US fewer than half of CS teachers feel equipped to teach AI itself.

Source: Stanford HAI AI Index 2025 · 2025

2×

AI-tutored students learned about twice as much

A randomized controlled trial of 194 Harvard physics students (2023) found learning gains for the AI-tutored group were roughly double those of students in an active-learning classroom, in less time. Published in Scientific Reports (2025).

Source: Kestin et al., Scientific Reports / Harvard (2025) · 2025

200M+

AlphaFold has predicted structures for over 200 million proteins

Google DeepMind's AlphaFold, working with EMBL's European Bioinformatics Institute, released predicted 3D structures for more than 200 million proteins — nearly every catalogued protein known to science — used by millions of researchers to accelerate drug discovery and biology.

Source: Google DeepMind / EMBL-EBI · 2024

66%

Two in three US physicians now use AI in their practice

The American Medical Association's 2024 physician survey found 66% of doctors used AI in their practice, up from 38% in 2023 — a 78% year-on-year jump. Documentation, translation and summarising medical evidence were the most common uses.

Source: American Medical Association (AMA) · 2024

75%

About 75% of UK financial firms now use AI

The Bank of England and Financial Conduct Authority's third survey of AI in UK financial services found 75% of firms already use AI, with a further 10% planning adoption within three years — up from 58% two years earlier.

Source: Bank of England / FCA · 2024

95%

AI adoption reaches 95% among UK insurers

In the same Bank of England / FCA survey, insurance had the highest AI adoption of any financial sub-sector at 95%, followed by international banks at 94%; financial-market-infrastructure firms were lowest at 57%.

Source: Bank of England / FCA · 2024

+20%

Generative AI lifted Mastercard's fraud detection by up to 300%

Mastercard's Decision Intelligence Pro, a proprietary generative-AI model scanning around 125 billion annual transactions, improves fraud-detection rates by an average of 20% — and by as much as 300% in some cases — while cutting false positives by more than 85%.

Source: Mastercard · 2024

17-33%

Leading legal-AI tools hallucinate 17-33% of the time

A Stanford RegLab / HAI study — the first preregistered evaluation of commercial legal-AI tools — found LexisNexis's Lexis+ AI and Thomson Reuters' Westlaw AI-Assisted Research each produced incorrect or unsupported answers between 17% and 33% of the time, despite marketing claims of hallucination-free research.

Source: Stanford RegLab / HAI · 2024

55%

55% of retailers are ready to deploy generative AI now

A Google Cloud survey released for NRF 2024 found 81% of retail decision-makers feel urgency to adopt generative AI and 55% are ready to deploy it immediately, with store analytics, pricing and conversational AI the top use cases.

Source: Google Cloud / NRF · 2024

1,012

South Korea has 1,012 robots per 10,000 factory workers

The International Federation of Robotics' World Robotics 2024 report ranks South Korea the world's most automated manufacturing economy at 1,012 robots per 10,000 employees — more than six times the global average of 162 — ahead of Singapore (770), China (470) and Germany (429).

Source: International Federation of Robotics (IFR) · 2024

162

Global factory robot density has doubled in seven years to 162

The IFR reports global manufacturing robot density reached a record 162 units per 10,000 employees in 2023 — double the 74 recorded seven years earlier — as automation adoption accelerated across Asia, Europe and the Americas.

Source: International Federation of Robotics (IFR) · 2024

-50%

AI predictive maintenance can cut equipment downtime by up to 50%

McKinsey estimates analytics- and AI-driven predictive maintenance can reduce machine downtime by up to 50%, extend machine life by 20-40% and lower maintenance costs by 10-40% by flagging failures before they happen.

Source: McKinsey & Company · 2024

700

Klarna's AI assistant does the work of 700 support agents

Klarna's OpenAI-powered assistant handled 2.3 million customer-service conversations in its first month — two-thirds of all chats, equivalent to about 700 full-time agents — resolving errands in under 2 minutes versus 11 minutes previously. Klarna later rebalanced by rehiring humans for complex cases.

Source: Klarna · 2024

2x

Gen-AI use in marketing and sales more than doubled in a year

McKinsey's State of AI 2024 survey found the largest jump in generative-AI adoption came in marketing and sales, where reported use more than doubled year on year; revenue increases from AI were most commonly reported in this function.

Source: McKinsey & Company · 2024

2,133

US federal agencies reported 2,133 AI use cases in 2024

The 2024 US Federal AI Use Case Inventory, consolidated by the Office of Management and Budget, listed 2,133 publicly reportable AI use cases across 41 agencies — roughly triple the ~710 reported in 2023.

Source: OMB — 2024 Federal AI Use Case Inventory · 2024

+29%

AI-supported mammography detected 29% more breast cancers

The MASAI randomised controlled trial in Sweden (over 100,000 women) found AI-supported screen reading raised the cancer detection rate by 29% while cutting radiologist reading workload by 44% versus standard double reading. Results were published in The Lancet Oncology (Lang et al., 2023).

Source: The Lancet Oncology (MASAI trial) · 2023

$200-340B

Generative AI could add $200-340bn a year to banking

McKinsey estimates generative AI could add $200 billion to $340 billion in value annually to the global banking industry — equivalent to 9-15% of operating profits — mainly through productivity in customer operations, marketing, software and risk.

Source: McKinsey & Company · 2023

$240-390B

Generative AI could add $240-390bn a year to retail

McKinsey estimates generative AI could create $240 billion to $390 billion in additional value each year across retail — a potential 1.2-2.0 percentage-point lift in margins — led by customer service, marketing personalisation and merchandising.

Source: McKinsey & Company · 2023

+30-45%

Gen AI can lift customer-service productivity by up to 45%

McKinsey estimates generative AI could increase productivity in customer-care functions by 30-45% of current costs, by drafting agent responses, summarising calls and resolving routine queries automatically.

Source: McKinsey & Company · 2023

70%

70% of large US crop farms now use GPS auto-guidance

USDA Economic Research Service data show guidance/autosteering systems were used on 70% of large-scale crop-producing farms and 52% of midsize farms in 2023, up from single-digit adoption in the early 2000s.

Source: USDA Economic Research Service (2023 ARMS) · 2023

$0.8–1.2T

Gen AI could add up to $1.2 trillion a year to sales and marketing

McKinsey estimates generative AI could add $0.8–1.2 trillion in incremental annual productivity across sales and marketing — one of the highest-value functions among the 63 use cases it analysed.

Source: McKinsey — Economic Potential of Generative AI (2023) · 2023

14%

AI raised customer-support productivity by 14% on average

A study of 5,179 customer-support agents by Brynjolfsson, Li and Raymond found access to a generative-AI assistant raised productivity (issues resolved per hour) by 14% on average. Published in NBER and the Quarterly Journal of Economics.

Source: Brynjolfsson et al., NBER WP 31161 · 2023

34%

The least-experienced support agents gained 34% from AI

The same NBER study found the AI assistant's benefit concentrated among less-experienced staff: novice and low-skilled agents improved 34%, while experienced agents saw minimal change, as the tool spread expert practices.

Source: Brynjolfsson et al., NBER WP 31161 · 2023

Dec 2027

The EU delayed its high-risk AI rules to December 2027

Under the Digital Omnibus package (provisionally agreed May 2026), the EU deferred obligations for high-risk (Annex III) AI systems from August 2026 to 2 December 2027, easing the compliance timeline for businesses.

Source: European Commission (Digital Omnibus) · 2026

362

Reported AI incidents jumped 55% in a single year

Documented AI incidents rose to 362 in 2025, up 55% from 233 the year before, per the AI Incident Database as reported in the Stanford AI Index.

Source: Stanford HAI AI Index 2026 (AI Incident Database) · 2025

11%

Fewer organisations operate with no responsible-AI policy

The share of organisations reporting no responsible-AI policies fell from 24% in 2024 to 11% in 2025, according to the Stanford AI Index.

Source: Stanford HAI AI Index 2026 · 2025

53%

Global trust in AI companies has slid to 53%

Trust in AI companies fell from 61% to 53% globally over five years, per the 2025 Edelman Trust Barometer (28-country survey of over 32,000 respondents).

Source: 2025 Edelman Trust Barometer · 2025

35%

Only about a third of Americans trust AI companies

US trust in AI companies dropped 15 points, from 50% to 35%, over five years, per the 2025 Edelman Trust Barometer.

Source: 2025 Edelman Trust Barometer · 2025

50%

Half of Americans are more concerned than excited about AI

50% of US adults say the increased use of AI in daily life makes them more concerned than excited, versus just 10% more excited (Pew survey of 5,023 US adults, 9–15 June 2025).

Source: Pew Research Center · 2025

57%

A majority rate AI's societal risks as high

57% of Americans rate the societal risks of AI as high, per Pew's June 2025 survey of 5,023 US adults.

Source: Pew Research Center · 2025

~60%

Most Americans want more control over how AI is used

About six-in-ten US adults say they would like more control over how AI is used in their lives, per Pew's June 2025 survey (n=5,023).

Source: Pew Research Center · 2025

~60%

Americans fear regulators won't go far enough on AI

About six-in-ten US adults are more concerned that government will not go far enough in regulating AI than that it will go too far (Pew survey of US public and AI experts, published April 2025).

Source: Pew Research Center · 2025

62%

Few Americans trust government to regulate AI well

62% of US adults have not too much or no confidence that government will regulate AI effectively, per Pew (published April 2025).

Source: Pew Research Center · 2025

34%

Concern about AI outweighs excitement worldwide

Across 25 countries, a median of 34% are more concerned than excited about AI; the US is highest at 50%, per Pew's global survey (published October 2025).

Source: Pew Research Center · 2025

€35M / 7%

Banned AI uses can trigger €35M fines under the EU AI Act

Breaches of the EU AI Act's prohibited-practices rules can draw fines up to €35 million or 7% of global annual turnover, whichever is higher (Article 99). Prohibitions have applied since 2 February 2025.

Source: EU Artificial Intelligence Act, Article 99 (Penalties) · 2025

2 Aug 2025

General-purpose AI rules took effect in August 2025

Obligations for providers of general-purpose AI models under the EU AI Act became applicable on 2 August 2025.

Source: EU Artificial Intelligence Act (implementation timeline) · 2025

1,208

US state legislatures flooded 2025 with AI bills

In 2025 lawmakers in all 50 states introduced AI-related bills — 1,208 in total, of which 145 were enacted, per the National Conference of State Legislatures.

Source: National Conference of State Legislatures (NCSL), 2025 AI legislation tracker · 2025

47

Most US states now have deepfake laws on the books

As of mid-2025, 47 states had enacted deepfake legislation; only Alaska, Missouri and Ohio had none, per MultiState's tracker.

Source: MultiState (deepfake legislation tracker) · 2025

May 2025

The first US federal AI-content law targets deepfakes

The TAKE IT DOWN Act (P.L. 119-12), signed in May 2025, is the first US federal law directly regulating AI-generated content, targeting nonconsensual intimate deepfakes.

Source: US Public Law 119-12 (TAKE IT DOWN Act) · 2025

2 Feb 2025

The EU has banned 'unacceptable-risk' AI since February 2025

The EU AI Act's bans on 'unacceptable-risk' uses — social scoring, untargeted facial-image scraping and certain manipulative or biometric systems — have been legally enforceable since 2 February 2025, the first tranche of the Act to take effect.

Source: European Commission · 2025

90

Trump's AI Action Plan sets out 90 federal policy actions

Following Executive Order 14179, the White House released 'Winning the Race: America's AI Action Plan' on 23 July 2025, outlining roughly 90 federal policy actions across accelerating innovation, building AI infrastructure and international AI diplomacy.

Source: The White House · 2025

Dec 2025

A December 2025 executive order pushes a single national AI framework

On 11 December 2025 President Trump signed an executive order, 'Ensuring a National Policy Framework for Artificial Intelligence', pressing a deregulatory, federally-led approach and challenging the patchwork of state AI laws.

Source: The White House · 2025

1 Sep 2025

China now requires all AI-generated content to be labelled

China's Measures for Labelling AI-Generated and Synthetic Content took effect on 1 September 2025, requiring both explicit and implicit (metadata) labels on AI-generated text, images, audio, video and virtual scenes.

Source: Cyberspace Administration of China · 2025

233

AI incidents reached a new record in 2024

The AI Incident Database logged 233 reported AI incidents in 2024, the highest annual total to that point. Figures are compiled in the Stanford HAI AI Index.

Source: Stanford HAI AI Index 2026 (AI Incident Database) · 2024

2024

Colorado passed the first comprehensive US state AI law

Colorado's SB 24-205, signed in May 2024, became the first comprehensive US state law governing high-risk AI systems.

Source: Colorado General Assembly (SB 24-205) · 2024

4x

Deepfakes detected worldwide quadrupled in a year

Sumsub recorded a fourfold increase in deepfakes detected worldwide from 2023 to 2024, when deepfakes made up about 7% of all fraud attempts.

Source: Sumsub (Identity Fraud research, 2024) · 2024

303%

Deepfakes surged before the 2024 US elections

Sumsub found deepfake incidents in the US jumped 303% in the run-up to the 2024 elections, part of a broad surge in election-year markets.

Source: Sumsub (Identity Fraud research, Q1 2024) · 2024

1st

The EU AI Act is the world's first comprehensive AI law

The EU AI Act, which entered into force on 1 August 2024, is the world's first comprehensive, risk-based horizontal regulation of artificial intelligence, applying rules by risk tier across all sectors.

Source: European Commission · 2024

5%

AI researchers put a 5% median chance on human extinction

In a survey of 2,778 published AI researchers, the median estimate for an extremely bad outcome such as human extinction was 5% (mean 9%).

Source: AI Impacts / Grace et al., 'Thousands of AI Authors on the Future of AI' (2024, surveyed 2023) · 2023

37.8–51.4%

Many AI researchers give double-digit odds to catastrophe

Between 37.8% and 51.4% of surveyed AI researchers gave at least a 10% chance that advanced AI leads to outcomes as bad as human extinction (n=2,778).

Source: AI Impacts / Grace et al., 'Thousands of AI Authors on the Future of AI' (2024, surveyed 2023) · 2023

Aug 2023

China was first to regulate generative AI at national level

China's Interim Measures for the Management of Generative AI Services took effect on 15 August 2023, among the world's first binding national rules for public generative-AI services, requiring content controls and security assessments.

Source: Cyberspace Administration of China · 2023