Stanford AI Index: lead narrows

Stanford’s 2026 AI Index says the United States still leads China on top AI model performance, but the gap had narrowed to 2.7% by March 2026. (hai.stanford.edu) The report was released by Stanford’s Institute for Human-Centered Artificial Intelligence on April 13, 2026, as the ninth edition of its annual audit of the field. It says U.S. and Chinese models have traded the lead multiple times since early 2025, and that DeepSeek-R1 briefly matched the top U.S. model in February 2025. (hai.stanford.edu) Stanford says the United States still produces more top-tier models and higher-impact patents, while China leads in publication volume, citation share, patent output, and industrial robot installations. The same report says U.S. private AI investment remains far larger, at 23 times China’s level. (hai.stanford.edu 1) (hai.stanford.edu 2) The report frames that race alongside a second trend: systems are getting stronger faster than the tools used to test, govern, and explain them. Stanford says frontier developers routinely publish capability scores, but reporting on safety, fairness, and other responsible-AI benchmarks remains sparse. (hai.stanford.edu 1) (hai.stanford.edu 2) That gap shows up in disclosure. Stanford says the average score on the Foundation Model Transparency Index fell to 40 in 2025 from 58 in 2024, with major blind spots around training data, compute resources, and post-deployment impact. (hai.stanford.edu) It also shows up in safety and reliability. The AI Incident Database logged 362 incidents in 2025, up from 233 in 2024, and Stanford says hallucination rates on a new benchmark ranged from 22% to 94% across 26 leading models. (hai.stanford.edu) The systems are also getting more expensive to build and run. Stanford says global corporate AI investment more than doubled in 2025, private investment grew 127.5%, and generative AI funding rose more than 200%. (hai.stanford.edu) Those gains come with a larger physical footprint. Stanford estimates Grok 4’s training emissions at 72,816 tons of carbon dioxide equivalent in 2025, says AI data-center power capacity reached 29.6 gigawatts, and estimates annual GPT-4o inference water use could exceed the drinking-water needs of 12 million people. (hai.stanford.edu) The benefits are spreading unevenly. Stanford says generative AI reached 53% population-level adoption within three years, but adoption tracks closely with national income, and the United States ranked 24th at 28.3% while Singapore reached 61% and the United Arab Emirates 54%. (hai.stanford.edu 1) (hai.stanford.edu 2) Workers and the public are responding differently than industry. Stanford says employment for U.S. software developers ages 22 to 25 fell nearly 20% from 2024, while 64% of Americans expect AI to lead to fewer jobs over the next 20 years and only 31% trust the U.S. government to regulate AI responsibly. (hai.stanford.edu) (hai.stanford.edu) Stanford’s bottom line is not that the U.S.-China race is over. It is that the performance lead is now thin, the spending is still climbing, and the institutions meant to measure and manage AI are still trying to catch up. (hai.stanford.edu)