AI & data · sourced statistics
Machine Learning Statistics 2026
This page brings together measured figures on machine learning: how many models exist and who builds them, the compute and money behind training, the open-source ecosystem, how developers use AI tools, the talent market and adoption by businesses. It covers global data with a focus on the United States and the European Union, mostly from 2025 and 2026. Figures come from Epoch AI, the Stanford AI Index, Hugging Face, the Stack Overflow Developer Survey, GitHub, ML Contests, the U.S. Bureau of Labor Statistics, Eurostat, the OECD and others. Every figure carries the exact sentence it was taken from and a link to the source.
Compiled by the FruitPunch research desk · Updated November 15, 2026
- figures
- 53
- sources
- 21
- publishers
- 13
The short answer
Epoch AI tracks more than 3,700 notable machine learning models, and training compute for frontier models has grown about 5 times a year since 2020. Hugging Face hosted 2.96 million public models by August 2026, 80% of Stack Overflow's 2026 respondents used AI at least an hour a day, and the BLS projects 35% job growth for U.S. data scientists from 2025 to 2035. In business, 88% of surveyed organizations use AI (Stanford AI Index), but official surveys put adoption near one firm in five.
Quote these
Key findings
Each finding names who published it. Copy it with the source attached, or jump to it in context.
- 3,700#1
Epoch AI's public database tracks more than 3,700 machine learning models released from 1950 to 2026.
Source: Epoch AI, Nov 2026
In context - 90%#2
Industry produced more than 90% of notable AI models in 2025, according to the Stanford AI Index.
Source: Stanford HAI, Apr 2026
In context - 5×#3
Training compute for frontier language models has grown about 5 times per year since 2020.
Source: Epoch AI, 2026
In context - 17.1 million#4
Global AI compute capacity has grown 3.3 times per year since 2022 and reached 17.1 million H100-equivalents.
Source: Stanford HAI, Apr 2026
In context - 2.96 million#5
Public model repositories on Hugging Face grew from 2.43 million to 2.96 million between January and August 2026.
Source: Hugging Face, Aug 2026
In context - 80%#6
In the 2026 Stack Overflow Developer Survey, 80% of respondents used AI at least an hour per day in their work.
Source: Stack Overflow, 2026
In context - 35 percent#7
U.S. employment of data scientists is projected to grow 35% from 2025 to 2035, much faster than the average for all occupations.
Source: U.S. Bureau of Labor Statistics, 2026
In context - 88%#8
88% of organizations surveyed in 2025 reported using AI in at least some part of the business, according to the Stanford AI Index.
Source: Stanford HAI, Apr 2026
In context
Reading the numbers
The frontier of machine learning is narrow and expensive. Epoch AI counts more than 3,700 notable models in its database, but only 33 publicly disclosed models had passed the GPT-4 compute scale by mid-2025. Training compute for frontier language models has grown about five times a year since 2020 and training costs 3.5 times a year, so the Stanford AI Index finds industry behind more than 90% of notable models in 2025. The United States produced 59 of them and China 35.
Hardware and power are where that growth shows up. Stanford HAI estimates global AI compute at 17.1 million H100-equivalents, with NVIDIA supplying more than 60%, and NVIDIA's own data center revenue reached $89.0 billion in a single quarter of 2026. Epoch AI reports that five hyperscalers own more than two-thirds of AI compute. The International Energy Agency puts data center electricity use at 415 TWh in 2024 and projects about 945 TWh by 2030.
The open ecosystem is broad but concentrated. Hugging Face grew to 2.96 million public model repositories by August 2026, yet 85.6% of models have fewer than 200 downloads and small models under 1 billion parameters take 83% of downloads. Chinese models, led by Qwen with 151,448 derivatives, took 41% of downloads in the year to spring 2026. GitHub counts more than 1.1 million public repositories using an LLM SDK.
For people, machine learning is now a daily work tool and a growing career, with caveats. 80% of 2026 Stack Overflow respondents use AI at least an hour a day, but only 16% trust it for most tasks. The BLS projects 35% growth in data scientist jobs to 2035 at a median wage of $120,230, while Stanford HAI reports that employment for software developers aged 22 to 25 fell nearly 20%. Among businesses, a lack of expertise is the top reason EU firms give for not adopting AI (Eurostat, 70.89%).
How many machine learning models exist and who builds them
- 13,700
Epoch AI's public database tracks more than 3,700 machine learning models released from 1950 to 2026.
Source: Epoch AI, Nov 2026
Exact wording at the source
Our public database, the largest of its kind, tracks over 3700 machine learning models from 1950 to today.
- 233
As of June 2025, Epoch AI had identified 33 publicly disclosed models trained with at least 10^25 floating-point operations, the scale of GPT-4.
Source: Epoch AI, Jun 2025
Exact wording at the source
We have identified 33 publicly disclosed models estimated to have been trained with at least 10 25 FLOP.
- 390%
Industry produced more than 90% of notable AI models in 2025, according to the Stanford AI Index.
Source: Stanford HAI, Apr 2026
Exact wording at the source
Industry produced over 90% of notable AI models in 2025, but the most capable models are now the least transparent.
- 459
U.S. organizations produced 59 notable AI models in 2025, compared with 35 from China.
Source: Stanford HAI, Apr 2026
Exact wording at the source
retains higher-impact patents and produced 59 notable models in 2025 to China's 35.
- 51 trillion
Parameter counts of the largest models have stayed near 1 trillion for three years, and frontier labs have stopped reporting them.
Source: Stanford HAI, Apr 2026
Exact wording at the source
Parameter counts have stayed near 1 trillion for three years, though reporting from frontier labs has stopped.
- 63.3%
In March 2026, the best closed model led the best open-weight model by 3.3%, up from 0.5% in August 2024.
Source: Stanford HAI, Apr 2026
Exact wording at the source
As of March 2026, the top closed model leads the top open model by 3.3%, up from 0.5% in August 2024.
- 737%
Industry's share of model development on Hugging Face fell from about 70% before 2022 to roughly 37% in 2025.
Source: Hugging Face, 2026
Exact wording at the source
Industry's share of overall development fell from around 70% before 2022 to roughly 37% in 2025.
- 841%
Chinese models accounted for 41% of downloads on Hugging Face over the year to spring 2026, the largest share of any country.
Source: Hugging Face, 2026
Exact wording at the source
In the past year, Chinese models quickly accounted for the plurality or 41% of downloads.
Chart
Arena Elo rating of the top model from each leading developer, March 2026
Human preference ratings on the Arena leaderboard as reported in the Stanford AI Index 2026.
- Anthropic1,503
- xAI1,495
- Google1,494
- OpenAI1,481
- Alibaba1,449
- DeepSeek1,424
Source: Stanford HAI, Technical Performance | The 2026 AI Index Report
Compute, training costs and hardware
- 95×
Training compute for frontier language models has grown about 5 times per year since 2020.
Source: Epoch AI, 2026
Exact wording at the source
Training compute for frontier language models has been growing at 5× per year since 2020.
- 103.5×
The cost to train frontier language models has risen about 3.5 times per year since 2020.
Source: Epoch AI, 2026
Exact wording at the source
The cost to train frontier language models has increased by 3.5×/year since 2020.
- 1149%
AI chip performance per dollar has improved by about 49% per year since 2023, in constant 2025 dollars.
Source: Epoch AI, 2026
Exact wording at the source
AI chip performance per dollar has grown by 49% per year since 2023.
- 121.1 million
The largest known AI data center, Colossus 2 in Memphis, has an estimated capacity of 1.1 million NVIDIA H100-equivalent chips.
Source: Epoch AI, 2026
Exact wording at the source
SpaceXAI’s Colossus 2 in Memphis, Tennessee has an estimated computing capacity of 1.1 million H100-equivalents, making it the largest known AI data center.
- 13two-thirds
Five hyperscale cloud companies own more than two-thirds of global AI compute, according to Epoch AI.
Source: Epoch AI, 2026
Exact wording at the source
Five hyperscalers own over two-thirds of global AI compute.
- 1417.1 million
Global AI compute capacity has grown 3.3 times per year since 2022 and reached 17.1 million H100-equivalents.
Source: Stanford HAI, Apr 2026
Exact wording at the source
Global AI compute capacity grew 3.3x per year since 2022, reaching 17.1 million H100-equivalents.
- 1560%
NVIDIA chips account for more than 60% of total AI compute capacity worldwide.
Source: Stanford HAI, Apr 2026
Exact wording at the source
Nvidia accounts for over 60% of total compute, with Google and Amazon supplying much of the remainder and Huawei holding a small but growing share.
- 16$89.0 billion
NVIDIA reported data center revenue of $89.0 billion for the quarter ended July 2026, up 117% from a year earlier.
Source: NVIDIA, Aug 2026
Exact wording at the source
Second-quarter revenue was $89.0 billion, up 18% from the previous quarter and up 117% from a year ago.
- 17415
Data centers of all kinds used about 415 TWh of electricity in 2024, around 1.5% of world consumption.
Source: International Energy Agency, Apr 2025
Exact wording at the source
Data centres accounted for around 1.5% of the world’s electricity consumption in 2024, or 415 terawatt-hours (TWh).
- 18945
The IEA projects data center electricity use will more than double to around 945 TWh by 2030.
Source: International Energy Agency, Apr 2025
Exact wording at the source
Data centre electricity consumption is set to more than double to around 945 TWh by 2030.
Open-source models, datasets and code
- 192.96 million
Public model repositories on Hugging Face grew from 2.43 million to 2.96 million between January and August 2026.
Source: Hugging Face, Aug 2026
Exact wording at the source
Public model repositories grew from 2.43 to 2.96 million over the period, datasets from 711,000 to 1 million, Spaces from 1.00 to 1.44 million.
- 2085.6%
About 85.6% of models on Hugging Face have fewer than 200 lifetime downloads, while 1.5% of repositories account for 99.2% of all downloads.
Source: Hugging Face, Aug 2026
Exact wording at the source
The distribution underneath stays extreme, roughly 85.6% of models have fewer than 200 lifetime downloads, and 1.5% of repositories account for 99.2% of all downloads.
- 2183%
Among models that declare a size, those under 1 billion parameters take 83% of all-time downloads on Hugging Face, and models above 100 billion take 1%.
Source: Hugging Face, Aug 2026
Exact wording at the source
Among models that declare a parameter count, those under 1B take 83% of all-time downloads and everything above 100B takes 1%.
- 22151,448
Qwen-based models account for 151,448 derivative models on Hugging Face, 2.6 times Meta's total.
Source: Hugging Face, Aug 2026
Exact wording at the source
Qwen-based models now account for 151,448 derivatives on the Hub, 2.6× Meta’s total footprint and 4.7× the Llama repositories specifically.
- 2313 million
Hugging Face reached 13 million users in 2025, with more than 2 million public models and over 500,000 public datasets.
Source: Hugging Face, 2026
Exact wording at the source
In 2025, Hugging Face grew to 13 million users, more than 2 million public models, and over 500,000 public datasets.
- 245.6 million
There are 5.6 million AI projects on GitHub, and uploads to Hugging Face have tripled since 2023.
Source: Stanford HAI, Apr 2026
Exact wording at the source
Open-source AI development continues to scale, with 5.6 million projects on GitHub and Hugging Face uploads tripling since 2023.
- 251.1 million
More than 1.1 million public GitHub repositories used an LLM SDK in 2025, including 693,867 created in the previous 12 months.
Source: GitHub, Oct 2025
Exact wording at the source
More than 1.1 million public repositories now use an LLM SDK with 693,867 of these projects created in just the past 12 months alone
- 262.4 million
In 2025, 2.4 million GitHub repositories used Jupyter-style notebooks, up 75% from a year earlier.
Source: GitHub, Oct 2025
Exact wording at the source
In 2025, 2.4 million repositories used Notebooks (+75% YoY) and 1.9 million used Dockerfiles (+120% YoY).
Chart
Public repositories on the Hugging Face Hub, August 2026 (millions)
Public repositories only, end of the January to August 2026 period.
- Models2.96 million
- Spaces1.44 million
- Datasets1 million
Source: Hugging Face, State of Open Models: Summer 2026
How developers use AI and machine learning tools
- 2780%
In the 2026 Stack Overflow Developer Survey, 80% of respondents used AI at least an hour per day in their work.
Source: Stack Overflow, 2026
Exact wording at the source
Devs are using AI more than ever before in their daily work—80% of respondents from this year’s survey used AI at least an hour per day.
- 2831%
Among developers who use AI daily, 31% use it for more than four hours a day.
Source: Stack Overflow, 2026
Exact wording at the source
Of the over 70% of respondents who use AI daily, 31% use it over four hours per day.
- 2966%
Coding assistants and agents are the most common use of AI among 2026 survey respondents at 66%, ahead of general chatbots at 63%.
Source: Stack Overflow, 2026
Exact wording at the source
Coding assistants and agents have pulled ahead as the dominant use case for AI (66%), outpacing even general-purpose LLM-powered chatbots (63%).
- 3066%
Claude Code (66%) and GitHub Copilot (59%) were the most used coding agents among 2026 Stack Overflow respondents.
Source: Stack Overflow, 2026
Exact wording at the source
Claude Code (66%) and GitHub Copilot (59%) are the leaders for coding agents.
- 3148%
48% of 2026 respondents trust AI when they can easily validate its answers, and only 16% trust it for most tasks outside important work decisions.
Source: Stack Overflow, 2026
Exact wording at the source
When asked when it’s acceptable to trust AI, 48% indicate they trust AI when they can easily validate the answers and only 16% extend that trust to most tasks that are not important work decisions.
- 3230%
Among developers building AI workflows, 30% use LangChain and 24% use the OpenAI Agents SDK for orchestration.
Source: Stack Overflow, 2026
Exact wording at the source
Respondents are using LangChain (30%) and OpenAI Agents SDK (24%) for orchestration and Make (39%) and n8n (33%) for automation.
- 3384%
In 2025, 84% of Stack Overflow survey respondents used or planned to use AI tools in their development process, up from 76% in 2024.
Source: Stack Overflow, Jul 2025
Exact wording at the source
84% of respondents are using or planning to use AI tools in their development process, an increase over last year (76%).
- 3487%
87% of 2025 respondents said they were concerned about the accuracy of AI agents, and 81% about data security and privacy.
Source: Stack Overflow, Jul 2025
Exact wording at the source
87% of all respondents agree they are concerned about the accuracy, and 81% agree they have concerns about the security and privacy of data.
Chart
Hours per day spent using AI, 2026 Stack Overflow respondents
All respondents to the question, n = 10,926, 2026 Stack Overflow Developer Survey.
- 4+ hours a day31%
- 2 to 4 hours a day24%
- 1 to 2 hours a day25%
- Under 1 hour, every day12%
- Under 1 hour, not every day7%
- None1%
Source: Stack Overflow, 2026 Stack Overflow Developer Survey: AI
Machine learning talent, jobs and competitions
- 35275,600
Data scientists held about 275,600 jobs in the United States in 2025.
Source: U.S. Bureau of Labor Statistics, 2026
Exact wording at the source
Data scientists held about 275,600 jobs in 2025.
- 3635 percent
U.S. employment of data scientists is projected to grow 35% from 2025 to 2035, much faster than the average for all occupations.
Source: U.S. Bureau of Labor Statistics, 2026
Exact wording at the source
Employment of data scientists is projected to grow 35 percent from 2025 to 2035, much faster than the average for all occupations.
- 37$120,230
The median annual wage for U.S. data scientists was $120,230 in May 2025.
Source: U.S. Bureau of Labor Statistics, 2026
Exact wording at the source
The median annual wage for data scientists was $120,230 in May 2025.
- 3822%
The number of new AI PhDs in the United States and Canada rose 22% from 2022 to 2024.
Source: Stanford HAI, Apr 2026
Exact wording at the source
The number of new AI PhDs in the U.S. and Canada increased 22% from 2022 to 2024
- 3989%
The number of AI researchers and developers moving to the United States has fallen 89% since 2017.
Source: Stanford HAI, Apr 2026
Exact wording at the source
The number of AI researchers and developers moving to the United States has dropped 89% since 2017.
- 4020%
U.S. employment of software developers aged 22 to 25 has fallen nearly 20% from 2024.
Source: Stanford HAI, Apr 2026
Exact wording at the source
Employment for software developers ages 22 to 25 has fallen nearly 20% from 2024.
- 41390
More than 390 machine learning competitions ran in 2025 across 30-plus platforms, with total prize money above $16 million.
Source: ML Contests, 2026
Exact wording at the source
More than 390 machine learning competitions took place in 2025, across 30+ platforms, with a total prize pool of over $16m.
- 42seven million
Kaggle added about seven million users in 2025, according to ML Contests.
Source: ML Contests, 2026
Exact wording at the source
As well as growing by another seven million users, Kaggle’s product saw more significant features added in 2025 than in previous years.
- 4344
Of the 2025 competition winners ML Contests analyzed, 44 used PyTorch and only 1 used TensorFlow.
Source: ML Contests, 2026
Exact wording at the source
PyTorch remains the dominant deep learning framework; we found 44 winning solutions using PyTorch and only 1 using TensorFlow.
- 441.5 million
Kaggle's free five-day AI Agents Intensive course with Google drew more than 1.5 million learners in November 2025.
Source: Google, Dec 2025
Exact wording at the source
Our partnership with Kaggle to host the 5-Day AI Agents Intensive Course surpassed all expectations, welcoming over 1.5 million learners from around the world.
Chart
Median annual wage, United States, May 2025
Occupational Employment and Wage Statistics data, May 2025.
- Data scientists$120,230
- Mathematical science occupations$107,570
- All occupations$50,980
Source: U.S. Bureau of Labor Statistics, Data Scientists: Occupational Outlook Handbook
Machine learning in business
- 4588%
88% of organizations surveyed in 2025 reported using AI in at least some part of the business, according to the Stanford AI Index.
Source: Stanford HAI, Apr 2026
Exact wording at the source
Organizational AI adoption continued to rise in 2025, up to 88% of surveyed organizations, though AI agent use remains early.
- 4620.2%
In 2025, 20.2% of firms across OECD countries used AI, up from 8.7% in 2023.
Source: OECD, Jan 2026
Exact wording at the source
In 2025, 20.2% of firms reported using AI, up from 14.2% in 2024 and 8.7% in 2023, meaning adoption has more than doubled over the past two years.
- 4757.3%
57.3% of information and communication technology firms in OECD countries used AI in 2025, the highest of any industry.
Source: OECD, Jan 2026
Exact wording at the source
In 2025, 57.3% of ICT firms across the OECD used AI, the highest share of any industry, followed by professional and scientific services (36.8%).
- 4855.03%
55.03% of large EU enterprises used AI technologies in 2025.
Source: Eurostat, Dec 2025
Exact wording at the source
In 2025, 55.03% of large EU enterprises used AI technologies.
- 4962.52%
62.52% of EU enterprises in the information and communication sector used AI in 2025.
Source: Eurostat, Dec 2025
Exact wording at the source
In 2025, the information and communication sector (with 62.52%) and professional, scientific and technical service activities (with 40.43%) stood out with the highest share of enterprises that used AI.
- 5070.89%
Among EU enterprises that considered AI but did not use it in 2025, 70.89% cited a lack of relevant expertise.
Source: Eurostat, Dec 2025
Exact wording at the source
In 2025, among EU enterprises who had ever considered using AI technologies, the most common reason for not using AI technologies was the lack of relevant expertise (70.89%), followed by lack of clarity about the legal consequences (52.52%), and concerns regarding violation of data protection and privacy (48.83%).
- 5137%
37% of U.S. firms with at least 250 employees reported using AI in their operations in spring 2026.
Source: U.S. Census Bureau, May 2026
Exact wording at the source
For example, 37% of firms with at least 250 employees reported using AI in their business operations.
- 5239.7%
In the U.S. Information sector, 39.7% of businesses used AI as of May 2026, about twice the national rate of 19.8%.
Source: U.S. Census Bureau, May 2026
Exact wording at the source
As of May 3, 2026, the AI use rates in the Information (39.7%) and Finance and Insurance (33.9%) sectors were both higher than the national rate (19.8%) but neither reported significant shifts since December.
- 5330%
More than 30% of Fortune 500 companies maintain verified accounts on Hugging Face.
Source: Hugging Face, 2026
Exact wording at the source
Over 30% of the Fortune 500 now maintain verified accounts on Hugging Face.
Chart
Share of EU enterprises using AI by size, 2025
Enterprises with 10 or more employees in the EU, 2025.
- Small17%
- Medium30.4%
- Large55%
Source: Eurostat, Use of artificial intelligence in enterprises
FAQ
Questions readers ask
- How many machine learning models are there?
- It depends on the definition. Epoch AI tracks more than 3,700 notable models released since 1950, while Hugging Face hosted 2.96 million public model repositories in August 2026, most of them fine-tunes, conversions and experiments. Only 33 publicly disclosed models had been trained at GPT-4 scale by June 2025, according to Epoch AI.
- How fast is the compute used to train AI models growing?
- Epoch AI estimates training compute for frontier language models has grown about 5 times per year since 2020, while training costs have grown 3.5 times per year. Stanford HAI reports global AI compute capacity grew 3.3 times per year since 2022 to 17.1 million H100-equivalents.
- Is machine learning a good career?
- U.S. projections are strong. The Bureau of Labor Statistics expects data scientist employment to grow 35% from 2025 to 2035, with a median wage of $120,230 in May 2025. Entry is harder for young workers: Stanford HAI reports employment for software developers aged 22 to 25 fell nearly 20% from 2024.
- Which deep learning framework is most popular?
- PyTorch. Among 2025 machine learning competition winners analyzed by ML Contests, 44 used PyTorch and only 1 used TensorFlow.
- How many companies use machine learning?
- Executive surveys summarized by the Stanford AI Index put organizational AI adoption at 88% in 2025. Official statistics are much lower because they count every small firm: 20.2% of firms across the OECD and 55.03% of large EU enterprises used AI in 2025.
Sources
We chose figures from the organizations that collected or modeled the data: research groups that track models and compute (Epoch AI, Stanford HAI), platforms reporting on their own users (Hugging Face, GitHub, Stack Overflow, Google for Kaggle), company filings (NVIDIA), and statistical agencies and international bodies (U.S. Bureau of Labor Statistics, U.S. Census Bureau, Eurostat, OECD, International Energy Agency). Definitions differ: Epoch AI's notable model list is curated and not exhaustive, compute and cost trends are estimates with confidence intervals, Hugging Face counts repositories rather than distinct models, and developer surveys are self-selected. Official business surveys count every firm equally, while executive surveys overrepresent large organizations. We checked every quote against the live source page before publishing. We store the sentence each number appears in, and a script reloads the publisher's page to confirm it is still there. The methodology has the details.
- Epoch AI. Data on AI Models (Nov 2026).
- Epoch AI. Over 30 AI models have been trained at the scale of GPT-4 (Jun 2025).
- Epoch AI. Trends in AI (2026).
- Eurostat. Use of artificial intelligence in enterprises (Dec 2025).
- GitHub. Octoverse: A new developer joins GitHub every second as AI leads TypeScript to #1 (Oct 2025).
- Google. Inside Kaggle's AI Agents intensive course with Google (Dec 2025).
- Hugging Face. State of Open Source on Hugging Face: Spring 2026 (2026).
- Hugging Face. State of Open Models: Summer 2026 (Aug 2026).
- International Energy Agency. Energy and AI: Executive summary (Apr 2025).
- ML Contests. The State of Machine Learning Competitions 2025 (2026).
- NVIDIA. NVIDIA Announces Financial Results for Second Quarter Fiscal 2027 (Aug 2026).
- OECD. AI use by individuals surges across the OECD as adoption by firms continues to expand (Jan 2026).
- Stack Overflow. 2026 Stack Overflow Developer Survey (2026).
- Stack Overflow. 2026 Stack Overflow Developer Survey: AI (2026).
- Stack Overflow. 2025 Stack Overflow Developer Survey: AI (Jul 2025).
- Stanford HAI. Research and Development | The 2026 AI Index Report (Apr 2026).
- Stanford HAI. Technical Performance | The 2026 AI Index Report (Apr 2026).
- Stanford HAI. The 2026 AI Index Report (Apr 2026).
- Stanford HAI. Economy | The 2026 AI Index Report (Apr 2026).
- U.S. Bureau of Labor Statistics. Data Scientists: Occupational Outlook Handbook (2026).
- U.S. Census Bureau. Large Firms With at Least 20 Employees Biggest AI Users (May 2026).
Cite this page
Link to this page or to a single numbered figure. If you reuse a number, name the organization that first published it.
FruitPunch. (2026, November 15). Machine Learning Statistics 2026. https://fruitpunch.ai/machine-learning-statistics
Found a number that changed or a source that moved? Email editorial@fruitpunch.ai. We plan to check this page again by January 14, 2027.
Related rankings
- Best AI Chatbots · top pick ChatGPT
Keep reading