The Three-Way AI Race in Numbers: United States, China and the EU
The United States has the money and the chips. China has the factories, the power and a model gap that has all but closed. The European Union has the rulebook and the trust. A data briefing on who actually leads what.
Who is winning the AI race? On the evidence available in late 2026, nobody is winning all of it. The United States leads on money, frontier models and computing power. China leads on research volume, patents, factory robots and electricity, and its best models now trail the best American ones by a margin too small to matter. The European Union trails both on almost every measure of scale, and leads on the one thing that cannot be bought: more people trust it to set the rules.
This briefing puts the main published numbers side by side. Every figure is linked to its source, and a section near the end sets out what these numbers cannot tell you.
- 23×US private AI investment relative to China's in 2025
- 2.7%lead of the top US model over the top Chinese model, March 2026
- 45%US share of the world's data centre electricity use in 2024
- 53%median share of people in 25 countries who trust the EU to regulate AI
Who leads what?
| Dimension | United States | China | European Union | Leader |
|---|---|---|---|---|
| Private AI investment, 2025 | $285.9 bn | $12.4 bn | $20.9 bn (all of Europe) | United States |
| Notable AI models released, 2025 | 59 | 35 | No country in the top three | United States |
| Best-model performance | Leads by 2.7% | Level, within 2.7% | Not at the frontier | Effectively tied |
| AI chips | Designs the leading chips | About 5% of Nvidia's computing output | Depends on imports | United States |
| Data centre electricity, 2024 | 45% of world total | 25% | 15% (Europe) | United States |
| Research papers and patents | Higher-impact patents | Most papers, citations and patents | Behind both | China |
| Public AI contracts and grants, 2013 to 2024 | $20.4 bn | Not comparable (see below) | $3.7 bn (all of Europe) | United States |
| Industrial robots installed, 2024 | 34,000 | 295,000, or 54% of the world total | 27,000 (Germany) | China |
| Binding AI law | No comprehensive federal law | Sector rules and state plans | AI Act, in force since 2024 | European Union |
| Trusted to regulate AI, median of 25 countries | 37% | 27% | 53% | European Union |
Sources for each row are given in the sections below. The pattern is the finding: the three blocs are not running the same race.
Who is spending the most?
The United States, by a distance that is hard to overstate. The 2026 AI Index from Stanford counts $285.9 billion of private AI investment in the US in 2025. Europe as a whole attracted $20.9 billion and China $12.4 billion, according to the report's economy chapter. Global private investment reached $344.7 billion, up 127.5% in a year, so the US accounted for roughly five of every six dollars. Counting mergers and acquisitions as well, total corporate AI investment hit a record $581.7 billion.
Europe is the whole continent, not only the EU. Source: Stanford HAI, 2026 AI Index Report, chapter 4.
The US also founded 1,953 newly funded AI companies in 2025, more than ten times the next country.
Two cautions belong beside that chart. First, the Index itself warns that private investment figures likely understate what China spends. It estimates that Chinese government guidance funds put about $184 billion into AI firms between 2000 and 2023, money this measure does not capture. Second, venture investment is now the smaller part of the story. Goldman Sachs Research estimates that total AI-related investment worldwide will reach about $1 trillion in 2026, with $581 billion of it in the United States, equal to 1.8% of US GDP. The better-known figure, $794 billion of capital spending by the largest US technology companies, measures something narrower.
Public money follows a different pattern. Democratic governments spend little directly; China's state spends a great deal through channels that are hard to compare.
| Public funding for AI | Amount | Period | What it counts |
|---|---|---|---|
| United States | $20.4 bn | 2013 to 2024 | Federal AI-related contracts and grants |
| Europe | $3.7 bn | 2013 to 2024 | Public AI contracts, of which the UK $1.6 bn, Germany $0.5 bn, France $0.3 bn |
| China | $184 bn (estimate) | 2000 to 2023 | Government guidance funds invested in AI firms |
These rows are not like for like, and all three come from the AI Index's economy and policy chapters. They are shown together to make one point: the US lead is a private-sector lead.
The EU's answer is public money used as a lever. Its InvestAI initiative, launched in February 2025, aims to mobilise €200 billion, including a €20 billion fund for AI gigafactories. An informal call for interest drew 77 proposals from 16 member states. The formal call for tenders opened on 30 July 2026 for up to seven gigafactories, with bids due on 12 November 2026, selection in early 2027 and operations expected within 18 months of that. The EU hopes its role as anchor customer will unlock more than €20 billion of private investment. Nineteen smaller AI factories are already being set up.
These are plans and procurement rounds. The American figure is money already spent.
Who builds the best models?
Here the picture changed fastest. For years the US led on every measure of model quality. The AI Index now describes the performance gap between US and Chinese models as effectively closed: the two countries' models have traded the lead several times since early 2025, and in March 2026 the top American model led the top Chinese one by 2.7%.
The Index's technical performance chapter lists the laboratories in the top tier of the Arena leaderboard, where models are rated against one another by human votes. Four are American and two are Chinese. None is European.
Ratings are relative scores from head-to-head human votes, so the bars show closeness, not absolute ability. Source: Stanford HAI, 2026 AI Index Report, chapter 2.
On volume, the US is still ahead. It released 59 notable AI models in 2025 against China's 35. The third country on the list is South Korea, with 8. No European country is in the top three.
Counts depend on how a 'notable' model is defined. Source: Stanford HAI, 2026 AI Index Report, chapter 1.
The United States spent 23 times as much as China and bought a lead of 2.7%.
That is the most striking ratio in the data, and it should be read carefully. It does not show that American money was wasted. Much of that spending buys computing capacity for serving customers rather than a higher benchmark score. But it does show that money alone no longer guarantees a durable lead in model quality.
Who has the chips?
The United States, and this is where its advantage is deepest. The AI Index estimates that the world's AI computing capacity reached 17.1 million H100-equivalents, a standard unit based on one Nvidia chip, having grown 3.3 times a year since 2022. Nvidia alone accounts for more than 60% of it, with Google and Amazon supplying much of the rest and Huawei holding what the Index calls a small but growing share. US export controls limit which American chips China may buy.
How much that matters is disputed. A Council on Foreign Relations analysis compared the two national champions. Nvidia's chief executive has said the company would make four to five million AI chips in 2025. Public estimates for Huawei's output that year range from 200,000, the US Commerce Department's figure, to 400,000. The analysis, published in December 2025, concludes that even on aggressive assumptions Huawei produces only about 5% of Nvidia's total AI computing power, and under 3% on its central assumptions. It also finds that the best US chips are roughly five times more powerful than Huawei's.
| Nvidia (US) | Huawei (China) | |
|---|---|---|
| AI chips made in 2025 | 4 to 5 million | 200,000 to 400,000 (estimates) |
| Total AI computing power produced | 100% (baseline) | Under 3% of Nvidia's on central assumptions, about 5% on aggressive ones |
| Manufacturing process | Leading-edge, made by TSMC in Taiwan | 7 nm, made by SMIC in China |
Three things complicate the simple reading.
- The American lead runs through Taiwan. The AI Index notes that the US hosts 5,427 data centres, more than ten times any other country, and that a single company, TSMC, fabricates almost every leading AI chip. A TSMC expansion in the US began operating in 2025.
- The controls leak, and they move. The CFR paper was written after the US administration announced in December 2025 that it would allow sales of Nvidia's H200 chip to China. A CNAS analysis cites an estimate that Chinese firms have remotely accessed around one million H100-equivalents of computing power through cloud providers in third countries, which it puts at one-twentieth of the world's total.
- China is building around the gap. The same CNAS piece cites a Goldman Sachs estimate that China can produce only around 8% of the advanced AI chips it needs. CSIS has documented how much effort is going into closing that gap, and how DeepSeek found Huawei's chips usable for running models even where they fall short for training them.
The EU is largely a customer in this market. No European company appears in either column of that table.
Who has the power?
AI runs on electricity, and here the ranking starts to shift. The International Energy Agency estimates that data centres used about 415 terawatt-hours in 2024, around 1.5% of the world's electricity. The United States accounted for 45% of that, China 25% and Europe 15%. The AI Index puts the power capacity of AI data centres alone at 29.6 gigawatts, comparable to the peak demand of New York state.
The IEA's base case has global data centre demand more than doubling to about 945 TWh by 2030, just under 3% of world electricity use. Almost 80% of the growth comes from two countries.
US demand rises by about 130%, China's by 170%, Europe's by 70%. Source: IEA, Energy and AI (2025), base case.
Demand is one half of the question. Supply is the other, and on supply China is in a category of its own. According to Ember's Global Electricity Review 2026, China's clean electricity generation grew by 561 TWh in 2025 while its total demand grew by 503 TWh. The solar capacity China installed in that single year, 378 gigawatts, exceeded the entire installed solar capacity of the United States.
Put the two sources together and one comparison stands out: China added more clean generation in one year than every data centre on Earth consumed in 2024. That is our arithmetic on two separate datasets, not a claim either source makes. It is still a useful measure of scale. In the US, by contrast, the IEA expects data centres to account for nearly half of all growth in electricity demand to 2030.
Who has the people and the ideas?
China produces more of the raw material of research. The AI Index reports that China leads the world in the volume of AI publications, in citations and in patent grants, while the US still produces more top-tier models and higher-impact patents. Quality is following quantity: China's share of the 100 most-cited AI papers rose from 33 in 2021 to 41 in 2024.
The more consequential number may be about people. The Index finds that the number of AI researchers and developers moving to the United States has fallen by 89% since 2017, with an 80% drop in the last year alone. The US is still home to more AI talent than any other country, but it is attracting new talent at the lowest rate in over a decade. American AI was built in large part by people who came from somewhere else. A country can lead on capital and chips and still lose ground if the talent stops arriving.
Who is actually using AI?
Building AI and using it are different contests, and the second is where China's industrial base shows.
Source: Stanford HAI, 2026 AI Index Report, chapter 4.
China installed about 295,000 industrial robots in 2024, which the Index puts at 54% of the world total: more than every other country combined. Its government intends to keep going: the State Council's "AI Plus" guideline of August 2025 sets a target for intelligent terminals and AI agents to reach more than 70% penetration by 2027 and more than 90% by 2030.
Among individuals, the richest countries are not the heaviest users. The AI Index puts generative AI adoption at 28.3% of the population in the US, which ranks 24th, behind countries such as Singapore at 61%. At work, 58% of employees worldwide say they use AI regularly or semi-regularly, and the share is above 80% in China, according to the Index's public opinion chapter.
Europe offers a lesson in how much the choice of measure matters. In a survey of organisations reported in the AI Index, 91% of European respondents said their organisation uses AI, the highest of any region. Eurostat, which counts all enterprises with ten or more employees, found that 20.0% of EU enterprises used AI technologies in 2025, up from 13.5% in 2024, ranging from 42.0% in Denmark to 5.2% in Romania. Both numbers are honest. They describe different populations, and anyone quoting only one of them is choosing a story.
Who writes the rules?
The European Union, though less firmly than a year ago.
- August 2024. The EU AI Act enters into force, with obligations phased in over several years.
- November 2025. The European Commission proposes a "Digital Omnibus" to simplify the Act.
- June 2026. The European Parliament and the Council approve it.
- 27 July 2026. The amendments enter into force. Obligations for stand-alone high-risk AI systems, covering uses such as hiring, credit scoring, education and access to essential services, move from 2 August 2026 to 2 December 2027. High-risk AI built into regulated products moves to August 2028.
- 2 August 2026. Most transparency obligations, such as telling people when they are dealing with an AI system, apply on the original schedule.
The stated reason for the delay was that technical standards and national authorities were not ready. The effect is that the part of the law dealing with the highest-stakes uses will arrive 16 months later than planned.
Even so, the EU holds an asset neither rival has. When the Pew Research Center asked adults in 25 countries whom they trust to regulate AI effectively, a median of 53% said they trust the EU. The figure for the United States was 37%, and for China 27%. Even among Americans, 43% said they trust the EU on this and 40% said they do not.
Source: Pew Research Center, surveys conducted January to April 2025.
At home, Americans are sceptical of their own government on this. In the same Pew study, 44% of US adults said they trust their country to regulate AI effectively and 47% said they do not, against a median of 55% across all 25 countries. A separate survey reported in the AI Index puts American trust in the government to regulate AI responsibly at 31%, the lowest of any country it covered, where the average was 54%.
Europe is also building public computing capacity as a matter of policy. The Index counts state-backed AI supercomputing clusters in Europe and Central Asia rising from 3 in 2018 to 44 in 2025.
So who is winning?
It depends on which race you mean, and that is the honest answer rather than an evasion.
- If the race is to the most capable model, the US is ahead by a nose and China is level with it.
- If the race is for the infrastructure to run AI at scale, the US leads today on chips and data centres, and China leads on the electricity that both will need.
- If the race is to put AI to work in the physical economy, China's factories give it the lead.
- If the race is to decide the terms on which everyone else uses AI, the EU is the only one of the three with both a comprehensive law and the trust of people beyond its borders, and it has just given itself more time.
The EU's position is the weakest on scale and the most interesting on leverage. It cannot outspend either rival. What it has is a law already on the books and more trust abroad than either of them.
What do these numbers not show?
- Private investment is not total investment. The measure that makes the US lead look like 23 to 1 leaves out most Chinese state funding. The true ratio is smaller, and nobody outside China can say by how much.
- "Europe" is not the EU. Several of the figures above, including investment and data centre electricity, are for Europe as a region, which is wider than the European Union.
- A "notable model" is a definition, not a fact of nature. Different counting rules give different totals, and a count says nothing about how a model is used.
- Surveys disagree with each other. Pew finds 44% of Americans trust their government to regulate AI; the survey cited in the AI Index finds 31%. The wording differs, and so does the answer.
- Benchmark leads are narrow and unstable. A 2.7% gap on a leaderboard can reverse with a single release, and the Index itself reports growing concerns about how reliable such benchmarks are.
- Chip production figures for China are estimates. Published numbers for Huawei's output vary by a factor of two or more.
- Energy projections are scenarios. The IEA's own range for data centre demand in 2035 runs from 700 to 1,700 TWh.
- None of this measures who benefits. A country can lead on every metric here while most of its people see little of the gain. That is a separate question, and we looked at part of it in what the job forecasts actually say about AI.
The race framing itself deserves some suspicion. It is the language of the people who are asking for the money, the chips and the power. The numbers support a plainer description: three large economies are making three different bets on the same technology, and the rest of the world will live with whichever combination of them holds.
Sources
- Stanford Institute for Human-Centered AI, The 2026 AI Index Report, April 2026: investment, public funding, models, performance gap, computing capacity, talent, data centres, robots, adoption and public trust. Figures are taken from the report's overview and its chapter pages on research and development, technical performance, economy, policy and public opinion.
- International Energy Agency, Energy and AI, 2025: data centre electricity use and projections.
- Ember, Global Electricity Review 2026: China's generation and solar capacity.
- Chris McGuire, Council on Foreign Relations, China's AI Chip Deficit, December 2025: Nvidia and Huawei output.
- Center for a New American Security, America Must Expand AI Export Controls, 2026: remote access to compute and domestic chip supply.
- Center for Strategic and International Studies, DeepSeek, Huawei, Export Controls, and the Future of the U.S.-China AI Race, 2025.
- Goldman Sachs Research, Global AI Investment Is Forecast to Exceed $1 Trillion in 2026, August 2026.
- Eurostat, 20% of EU enterprises use AI technologies, December 2025.
- European Commission and EuroHPC Joint Undertaking: InvestAI, AI gigafactories and the July 2026 call for tenders.
- State Council of the People's Republic of China, "AI Plus" guideline, August 2025.
- Pew Research Center, How People Around the World View AI, October 2025.
- DLA Piper, The Digital AI Omnibus, updated August 2026: the AI Act timetable.