Artificial intelligence is expanding across businesses, research, medicine, education, and public policy, but adoption remains uneven. The statistics below distinguish observed use from forecasts, estimates, benchmark scores, reported incidents, and survey responses across clearly defined populations and periods.
Key Artificial Intelligence Statistics
The following figures summarize the most useful measures across adoption, investment, technical progress, applications, and governance:
- In the United States, 5.4% of employer businesses excluding farms reported current AI use in February 2024, up from about 3.7% in September 2023.
- U.S. employer businesses expected AI use to reach about 6.6% by fall 2024, according to a February 2024 forward-looking survey response.
- 18.1% of U.S. information-sector firms reported current AI use across pooled December 2023–February 2024 reference periods.
- 13.9% of enterprises in the OECD area implemented AI in 2024, with enterprise coverage generally beginning at 10 employees.
- 13.5% of EU27 enterprises with at least 10 employees used AI in 2024, up from 8.0% in 2023.
- Large OECD firms had a 39% AI-use rate in 2024, compared with 12% for small firms.
- Global corporate AI investment reached an estimated $252.3 billion in 2024.
- Global private AI investment increased an estimated 44.5% in 2024 from the prior year.
- Private investment in generative AI reached an estimated $33.9 billion globally in 2024.
- Estimated GPT-4 training cost was about $79 million in 2023, based on cloud-compute rental prices.
- The cost of querying a model at GPT-3.5-equivalent MMLU accuracy fell from $20 to $0.07 per million tokens between November 2022 and October 2024.
- AI systems’ SWE-bench coding success rate rose from 4.4% in 2023 to 71.7% in 2024 on that benchmark.
- OpenAI o1 scored 96.0% on the MedQA clinical-knowledge benchmark in 2024, which is not a measure of clinical-care performance.
- FDA-approved AI-enabled medical devices increased from six in 2015 to 223 in 2023 in the cited count.
- The AI Incidents Database recorded 233 AI-related incidents globally in 2024, while noting that its record is not comprehensive.
- China had 83% of surveyed respondents say AI products and services offer more benefits than drawbacks in 2024, compared with 39% in the United States.
Contents
- U.S. artificial intelligence adoption statistics
- Global enterprise AI adoption data
- AI investment and model-training costs
- AI model performance and efficiency trends
- Artificial intelligence in science, medicine, and education
- AI governance, safety, and public attitudes
U.S. Artificial Intelligence Adoption Statistics
The U.S. Census Bureau’s Business Trends and Outlook Survey (BTOS) measures employer businesses excluding farms, defining current use as AI used in producing goods or services during the prior two weeks. Its firm-weighted and employment-weighted measures describe different forms of exposure.
| Measure | Result | Period and population |
|---|---|---|
| Current AI use | 3.7% to 5.4% | September 2023–February 2024, U.S. employer businesses |
| Expected AI use | About 6.6% | Fall 2024 expectation reported February 2024 |
| Employment at firms currently using AI | About 4.5% to nearly 9% | September 2023–February 2024, U.S. employment |
| Employment at firms expecting AI use | About 10% to 12% | September 2023–February 2024 expectations |
The information sector reported the highest current-use rate in the cited pooled BTOS snapshot: 18.1% from December 3, 2023, through February 11, 2024. Construction reported 1.4% over the same pooled reference periods, while six-month expectations were 21.5% for information and 2.2% for construction. Tracking Firm Use of AI in Real Time: A Snapshot from the Business Trends and Outlook Survey
About one in seven U.S. AI-using firms did not expect to continue using AI in the near future during December 2023–February 2024. That result indicates a continuation expectation, not permanent de-adoption, and the 6.6% figure is a forecast rather than an observed fall-2024 rate.
Global Enterprise AI Adoption Data
International business-use measures show rising adoption alongside wide diffusion gaps. OECD and Eurostat enterprise statistics generally cover organizations with at least 10 employees, so they should not be treated as population-wide firm counts.
| Geography or group | AI adoption | Period |
|---|---|---|
| OECD area | 13.9% of enterprises | 2024 |
| EU27 enterprises | 13.5%, up from 8.0% | 2023–2024 |
| OECD information and communication services | 44% average | 2024 |
| OECD professional science and technology services | 26% average | 2024 |
| Large OECD firms | 39% | 2024 |
| Small OECD firms | 12% | 2024 |
Within the EU, Denmark recorded the highest cited enterprise AI-use share at 27.6% in 2024, while Romania recorded 3.1%, the lowest among the cited countries. Sweden had the largest cited increase, gaining 14.7 percentage points from 2023 to 2024; text-mining AI alone reached 6.9% of EU enterprises in 2024 after a 4.0-percentage-point year-over-year increase. Emerging divides in the transition to artificial intelligence and Usage of AI technologies increasing in EU enterprises
The OECD country adoption range widened to 24 percentage points in 2024, spanning roughly 4% to 28%. This is a country-rate range, not a population-weighted global average.
AI Investment and Model-Training Costs
Capital flows and compute requirements show why artificial intelligence is both a business technology and an infrastructure-intensive field. The following investment figures are estimates from the AI Index and should not be read as identical accounting categories.
| Investment measure | 2024 result | Geography or scope |
|---|---|---|
| Corporate AI investment | $252.3 billion | Global |
| Private AI investment | Up 44.5% year over year | Global, 2023–2024 |
| Generative-AI private investment | $33.9 billion | Global |
| U.S. private AI investment | $109.1 billion | United States |
| China private AI investment | $9.3 billion | China |
| United Kingdom private AI investment | $4.5 billion | United Kingdom |
Global AI mergers and acquisitions increased 12.1% in 2024 from 2023, while generative-AI private investment grew 18.7%. Corporate investment is broader than private investment alone, and the country figures are private-investment estimates. Economy | The 2025 AI Index Report
Model-training costs are also estimates based on inflation-adjusted cloud-compute rental prices, not disclosed invoices. Estimated training costs were about $79 million for OpenAI GPT-4 in 2023, $170 million for Meta Llama 3.1-405B in 2024, $192 million for Google Gemini 1.0 Ultra in 2024, and $41 million for Mistral Large in 2024.
For historical scale, the original Transformer model’s estimated training cost was about $670 in 2017. Artificial Intelligence Index Report 2025
AI Model Performance and Efficiency Trends
Benchmark results show rapid gains, but each score applies to a particular task, test set, and evaluation design. They are not universal measures of workplace productivity or general intelligence.
The cost of querying a model matching GPT-3.5-equivalent MMLU accuracy fell from $20 to $0.07 per million tokens between November 2022 and October 2024. In 2024, U.S.-based institutions produced 40 notable AI models, compared with 15 from China and 3 from Europe. AI Index 2025: State of AI in 10 Charts
| Benchmark or measure | Earlier result | Later result |
|---|---|---|
| MMMU performance | 2023 baseline | Up 18.8 percentage points in 2024 |
| GPQA performance | 2023 baseline | Up 48.9 percentage points in 2024 |
| SWE-bench coding success | 4.4% in 2023 | 71.7% in 2024 |
| Smallest model above 60% on MMLU | PaLM, 540 billion parameters in 2022 | Phi-3-mini, 3.8 billion in 2024 |
On RE-Bench with a two-hour budget, top AI systems scored four times higher than human experts. This result is limited to that agent benchmark and time horizon, while the SWE-bench rate is not a general software-engineering productivity measure. Technical Performance | The 2025 AI Index Report
Artificial Intelligence in Science, Medicine, and Education
Medical and educational statistics show expanding institutional use and research attention, but they do not automatically establish clinical effectiveness or learning outcomes.
OpenAI o1 scored 96.0% on MedQA in 2024, and the leading score was 5.8 percentage points above the best 2023 score. MedQA performance improved 28.4 percentage points from late 2022 to 2024, although benchmark saturation can limit interpretation.
FDA-approved AI-enabled medical devices increased from six in 2015 to 223 in 2023. Medical-AI ethics publications rose from 288 in 2020 to 1,031 in 2024; publication counts measure literature volume, not research quality or clinical outcomes. Science and Medicine | The 2025 AI Index Report
The number of U.S. graduates earning a master’s degree in AI increased by nearly 100% between 2022 and 2023, a near-doubling rather than an exact percentage. In a 2024 survey, 81% of U.S. high-school computer-science teachers agreed that AI use and AI learning belong in foundational computer-science education. Education | The 2025 AI Index Report
AI Governance, Safety, and Public Attitudes
Governance activity and reported safety events increased alongside adoption. These measures capture policy attention, database reporting, conference output, or public opinion rather than the total prevalence or effectiveness of AI regulation.
The AI Incidents Database recorded 233 AI-related incidents in 2024, a 56.4% increase from 2023. The database is not comprehensive, and incident reporting can change over time.
Responsible-AI papers accepted at leading AI conferences rose from 992 in 2023 to 1,278 in 2024. Responsible AI | The 2025 AI Index Report
| Policy measure | Earlier | Later |
|---|---|---|
| U.S. federal AI-related regulations introduced | 25 in 2023 | 59 in 2024 |
| Agencies issuing those regulations | 21 in 2023 | 42 in 2024 |
| U.S. state-level AI-related laws | 1 in 2016; 49 in 2023 | 131 in 2024 |
| Legislative AI mentions across 75 major countries | 1,557 in 2023 | 1,889 in 2024 |
The federal regulation count covers regulations introduced, not necessarily enacted rules or regulatory impact. Across 75 major countries, legislative mentions measure attention rather than policy effectiveness, while the state-law classification follows the Stanford tracker. Policy and Governance | The 2025 AI Index Report
Public attitudes also differ by geography. In a 2024 survey, 83% of respondents in China said AI-powered products and services offer more benefits than drawbacks, compared with 39% of respondents in the United States; wording and sample design are source-specific. AI Index 2025: State of AI in 10 Charts