2026 Stanford AI Index: Closing US‑China Gap and 88% AI Adoption Explained

Report Overview

The Stanford Institute for Human-Centered AI released its ninth annual AI Index Report 2026 , a 400‑plus‑page assessment tracking advances in AI technology, research output, investment, talent, policy, and public perception. It is regarded as the most comprehensive independent annual evaluation of global AI development.

Resource Growth and Model Concentration

AI‑related resources continued to expand in 2025, yet the number of high‑profile models released fell slightly and the industry is increasingly dominated by a few institutions. Over 90% of well‑known frontier models are now owned by industry players, and the most powerful systems are the least transparent, with decreasing disclosure of training code, dataset size, and parameter counts.

Since 2022, compute used to train these models has grown roughly 3.3× per year, largely relying on a single Taiwanese chip‑foundry, exposing a fragile hardware supply chain.

US‑China Performance Gap Narrowing

While the United States still leads in top‑tier model development, the performance gap with Chinese models has effectively been "flattened." Chinese research excels in paper volume, citation share, and patent grants, and smaller nations such as Switzerland and Singapore lead in per‑capita AI researcher counts.

From early 2025, US and Chinese models have alternated in leading benchmarks. In February 2025 DeepSeek‑R1 briefly matched the best US models, and by March 2026 the leading models’ Elo scores differed by less than 25 points, indicating strong convergence.

Frontier Model Convergence

In the past year, the performance gap among leading models has shrunk, with top models from Anthropic, xAI, Google, OpenAI, Alibaba, and DeepSeek all residing in the first Elo tier. The gap between the top four models fell from 97 points in 2024 to under 25 points by March 2026.

Notably, Gemini Deep Think won a gold medal at the International Mathematical Olympiad but correctly read analog clocks only 50.1% of the time, highlighting lingering limitations in real‑world reasoning.

Adoption Surge

Enterprise AI adoption reached 88% in 2025, up from 78% in 2024. More than half of respondents reported AI use in at least three business functions, and 79% regularly employed generative AI, surpassing the 71% rate in 2024. Adoption growth was especially strong in China and Europe, outpacing global averages by 13% and 11% respectively.

Labor‑Market Impact

AI’s influence on employment is uneven. Since 2024, software developers aged 22‑25 have seen a near‑20% drop in employment rates. One‑third of firms anticipate AI‑driven staff reductions within a year, with the highest expected cuts in services, supply‑chain, and software engineering.

Expert‑Public Perception Gap

73% of AI experts expect a positive impact on work, versus only 23% of the public. Similar divides appear regarding AI’s effects on the economy and healthcare. Trust in governmental AI regulation varies widely, with the U.S. public showing the lowest confidence (31%).

Environmental and Transparency Concerns

Training the latest large language models (e.g., xAI’s Grok‑4) can emit over 72,000 tons of CO₂, a sharp rise from previous estimates. Inference emissions also vary, with the least efficient models emitting more than ten times the carbon of the most efficient ones. Data‑center power demand has climbed to 29.6 GW, comparable to New York State’s peak load, and GPT‑4o’s annual inference water usage could exceed the drinking water needs of 1.2 million people.

Model transparency is deteriorating: major labs (OpenAI, Anthropic, Google) have stopped disclosing training data size, parameter counts, and training duration. Of the 95 most influential AI models released in 2025, 80 lack publicly available training code, making the strongest models the least open.

Key Takeaways

AI resources keep growing, but model development is concentrating among a few firms.

The US‑China performance gap has largely vanished, signaling a more balanced global AI landscape.

Top‑tier models are converging in performance, shifting competition toward cost, reliability, and domain‑specific strengths.

Enterprise AI adoption surged to 88%, outpacing previous years and varying by region.

AI’s labor‑market impact is uneven, with notable job declines for young developers.

Transparency and environmental sustainability are emerging challenges as models become more powerful.