US vs. China:
The AI Race Is Closer Than You Think
The Stanford AI Index Report 2026 reveals a tectonic shift: China is catching up — and in some dimensions, it has already caught up. Here is what the data actually says.
For years, the conventional wisdom held that the United States owned artificial intelligence. American labs built the frontier models. American investors wrote the checks. American universities trained the talent. That narrative is no longer the full picture. The Stanford AI Index Report 2026 — the most comprehensive annual measurement of global AI progress — presents a more complicated, more interesting reality. The gap between the US and China has closed dramatically, and in several dimensions, China has already taken the lead.
This is not hype. It is measurement. Let’s go through the data dimension by dimension.
01 — Model Development The US Leads in Volume. China Is Catching Up Fast.
The most visible dimension of the AI race is model production — which country releases the most notable AI systems. In 2025, the United States led with 50 notable AI model releases. China came in second with 30. No other country came close: South Korea had 5, Canada 1, France 1, the UK 1.
But total model count is only one metric. When you look at the broader model ecosystem — including smaller and less prominent releases — China has accelerated explosively. Between 2022 and 2025, China’s total model releases more than quintupled from 151 to 849. That’s not incremental progress. That is systematic capacity building at industrial scale.
Within the US, industry dominates production: 91.6% of all notable models in 2025 came from industry rather than academia. The top contributors globally were OpenAI (19 notable models), Google (12), and Alibaba (11) — meaning a Chinese company already sits in the top three globally by output.
02 — Model Performance The Gap Has Effectively Closed
This is the headline finding from the 2026 report, and it deserves to be stated plainly: the US–China AI performance gap has effectively closed.
The Arena Leaderboard — a community-driven benchmark where users blind-compare model outputs and vote — serves as one of the most reliable real-world proxies for model quality. As of March 2026, the top US model (Claude Opus 4.6) leads the top Chinese model (Dola-Seed-2.0 Preview) by just 39 Arena points, or 2.7%.
Source: Arena (formerly LMArena), March 2026. Style Control On, Public leaderboard.
The trajectory matters as much as the current snapshot. In early 2023, OpenAI held a commanding lead — its top model scored 1,322 vs Google’s 1,117. That gap closed steadily through 2024. In February 2025, DeepSeek-R1 briefly matched and surpassed the top US system entirely. As of March 2026, the top four models globally — Anthropic, xAI, Google, and OpenAI — are separated by fewer than 25 Arena points. DeepSeek (1,424) and Alibaba (1,449) trail modestly.
“This convergence is particularly notable because it has emerged from two distinct development environments and institutional contexts.”
DeepSeek-R1 was arguably the most consequential single model release of 2025. It introduced GRPO, a reinforcement-learning approach that trains reasoning without relying on labeled data or a separate critic model. The model’s strong performance relative to higher-cost US systems temporarily erased over one trillion dollars in US technology stock market value — a real-world measure of how seriously the financial markets now take Chinese AI capabilities.
03 — Research Publications China Leads in Volume and Citations
In academic AI research, China has overtaken every other nation by volume. In 2024, China accounted for 17.8% of all AI publications, compared to 11.1% from Europe and 7.6% from India. China also leads in citation share: 20.6% of all AI citations globally, just ahead of Europe at 19.5%, with the US at 12.6%.
The US has seen its publication share decline by 3 percentage points, though its citation share remained relatively stable. In the top 100 most-cited AI papers of 2024, China contributed 41 papers — up from 33 in 2021, while the US contributed 64 — still dominant, but the trend line is unmistakable.
China grew from 33 to 41 papers (2021–2024). US declined from 46 to 64 — no, wait, US was 46 in 2021 and rose to 64. Both growing, but China closing the gap.
04 — Patents China Dominates Volume. US Dominates Influence.
AI patents tell a story of two very different innovation strategies. China accounts for 74.2% of all granted AI patents globally — more than six times the US share of 12.1%. Between 2010 and 2024, China was granted 97,990 AI patents, compared to 15,920 for the US.
But volume is not the same as influence. When measuring forward citations — how often a patent is cited by later inventions, a proxy for downstream impact — the United States accounts for over 51.9% of all AI patent forward citations, despite producing only 12.1% of patents. China accounts for 29.8% of forward citations despite holding 74.2% of patents.
In other words: China files more patents, but US patents tend to be more foundational — they are cited more often and more broadly. This is a classic distinction between output and impact. One additional asymmetry worth noting: Chinese patents are cited frequently in US filings, while US patents appear far less often in Chinese ones.
05 — Investment The US Is in a Different League
On private AI investment, the United States operates at a scale that no other country can match. In 2025, the US attracted $285.9 billion in private AI investment — 23.1 times more than China’s $12.4 billion, and 48.5 times more than the UK’s $5.9 billion. The US also led in newly funded AI companies: 1,953 in 2025, compared to 161 in China.
More than half of all US private AI investment was generative AI-related ($163.6 billion), while China and Europe’s combined generative AI investment was just $4.7 billion. Since 2024, US private AI investment grew 160.2% — vs 32.2% for China and 7.2% for Europe.
There is an important caveat: these figures capture private investment only, not state-directed capital. The Chinese government deploys resources through government guidance funds — state-initiated investment vehicles that mix financial returns with strategic objectives. Between 2000 and 2023, an estimated $912 billion of these funds were deployed across industries, with approximately $184 billion directed toward AI companies. In 2025 alone, China announced a $138 billion state VC fund targeting AI and cutting-edge technologies. The private investment gap understates how much capital China is actually directing toward AI.
“Comparisons based solely on private investment likely understate how much capital China is directing toward AI.”
06 — Data Centers & Infrastructure US Hosts the Hardware. Taiwan Holds the Key.
The United States hosts 5,427 AI data centers — more than ten times any other country. US data centers also consume more AI-related energy than any other nation. The Stargate Project, a joint venture between OpenAI, SoftBank, and Oracle, announced plans to invest between $100 billion and $500 billion to build advanced AI data centers across the US by 2029.
But here is the structural vulnerability that the 2026 AI Index highlights explicitly: a single company, TSMC (Taiwan Semiconductor Manufacturing Company), fabricates almost every leading AI chip in the world. The entire global AI hardware supply chain runs through one foundry on one island. A TSMC–US facility began operations in 2025, but the dependency remains acute. This is not a US-vs-China issue — it is a global systemic risk that both superpowers share.
07 — Adoption The US Builds AI. It Doesn’t Use It Most.
Perhaps the most counterintuitive data point in the entire report: the United States, despite leading in investment and model development, ranked 24th in population-level AI adoption in the second half of 2025, with a usage rate of 28.3%. The UAE (64.0%) and Singapore (60.9%) lead globally. France (44.0%), Ireland (44.6%), and Spain (41.8%) all surpass the US by wide margins.
China’s adoption data is not available in this ranking (the survey covers top-30 economies using the Microsoft AI Economy Institute methodology), but the contrast between the US’s technological leadership and its relatively cautious adoption rate is striking. The report attributes this partly to a more skeptical public mood toward AI in North America.
08 — Full Scorecard Where Each Country Leads
| Dimension | 🇺🇸 United States | 🇨🇳 China | Edge |
|---|---|---|---|
| Notable model releases (2025) | 50 models | 30 models | US |
| Top model performance (Arena) | 1,503 (Claude Opus 4.6) | 1,464 (Dola-Seed-2.0) | US (+2.7%) |
| Private AI investment (2025) | $285.9B | $12.4B (+state funds) | US (private) |
| AI publication volume (2024) | ~7–8% share (declining) | 17.8% share (leading) | China |
| AI citation share (2024) | 12.6% | 20.6% | China |
| AI patent volume (2010–2024) | 12.1% of global total | 74.2% of global total | China |
| AI patent forward citations | 51.9% share | 29.8% share | US (influence) |
| AI data centers | 5,427 | <500 (est.) | US |
| Industrial robot installations | Declining YoY | 54% of global total | China |
| AI adoption rate (H2 2025) | 28.3% (rank: 24th) | N/A in this survey | Not US |
| Autonomous vehicle rides (2025) | Waymo: ~450K/week | Apollo Go: 11M rides YTD | China (volume) |
09 — What This Means Strategic Implications for Tech Leaders
The 2026 AI Index report forces a rethink of some comfortable assumptions. The US remains the leader in building frontier models and in private capital deployment — but the margin is shrinking on performance, and China is already ahead on research volume and patent output.
A few things stand out for technology leaders and developers:
The performance gap may not be the right metric. As the report notes, frontier models are becoming harder to distinguish on benchmarks. The top four systems globally sit within 25 Arena points of each other. Competitive advantage is increasingly shifting toward cost efficiency, latency, reliability, and domain-specific optimization. On those dimensions, the race is wide open.
China’s research infrastructure is not playing catch-up — it has caught up. With 17.8% of global AI publications, 74% of patents, and explosive model release growth, China has built a serious domestic AI research foundation. DeepSeek demonstrated that highly capable models can be built with different architectural approaches and at lower cost.
The investment gap is real, but partly misleading. US private investment dwarfs China’s. But state-directed Chinese funding (estimated at $184 billion in AI alone through 2023, plus the new $138 billion state VC fund announced in 2025) is not captured in the private figures. Total capital mobilized toward AI in China is substantially larger than the headline numbers suggest.
The hardware dependency is everyone’s problem. Both countries rely on TSMC. Both are exposed to the same geopolitical chokepoint. The US TSMC fab expansion that began in 2025 is a step toward reducing this risk, but it is a step, not a solution.
AI adoption is not following investment. The US leads in building AI, but its own population uses it less than France, Spain, or Ireland. The question of who benefits from AI is becoming separable from the question of who builds it.
“The U.S. still produces more top-tier AI models and higher-impact patents, while China leads in publication volume, citations, patent output, and industrial robot installations.”
The race between the US and China in AI is not heading toward a decisive winner — it is heading toward sustained, parallel competition across different layers of the stack. The US builds the models. China builds the research base. Both are investing heavily in infrastructure. And the gap at the frontier, by any serious measurement, is now smaller than a rounding error.
For anyone building AI products, deploying AI in their organization, or making investment decisions in the technology sector: the assumption that “US = frontier, China = follower” is now factually incorrect. The appropriate mental model is two systems running at near-parity, with different strengths, different funding structures, and different constraints — and the outcome uncertain enough that it warrants watching closely.
How does the US compare to China in AI model development?
In 2025, the US led with 50 notable AI model releases, while China had 30. However, China’s total model releases surged from 151 to 849 between 2022 and 2025, indicating rapid growth.
What is the current state of the US-China AI performance gap?
The AI performance gap has effectively closed, with the top US model leading the top Chinese model by just 39 Arena points as of March 2026.
Who leads in AI research publications and citations?
China leads in both volume and citations of AI research, accounting for 17.8% of all AI publications and 20.6% of citations globally as of 2024.
Discover more from The Tech Society
Subscribe to get the latest posts sent to your email.