Stanford AI Index 2026: Record Power Use, US‑China AI Race
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Stanford AI Index released its 2026 report, noting that AI model performance continues to improve and that adoption rates exceed those of personal computers and the internet.
- AI data centers worldwide now draw 29.6 gigawatts of power—enough to power New York at peak demand—while OpenAI’s GPT‑4o alone may use water equivalent to the drinking needs of 1.2 million people.
- United States hosts an estimated 5,427 AI data centers (over ten times any other country) and leads in model power and capital, whereas China leads in AI research publications, patents, and robotics, making the two nations nearly tied on model performance per the Arena ranking.
- Anthropic is the top‑ranked LLM as of March 2026, followed closely by xAI, Google, and OpenAI, while Chinese models such as DeepSeek and Alibaba lag only modestly.
- Benchmarks are struggling to keep pace: a popular math benchmark has a 42% error rate, and companies increasingly with training details, hindering independent safety testing.
- AI adoption has reached over 50 % of the global population, with 88 % of organizations and 80 % of university students using AI, yet software‑developer employment for ages 22‑25 has fallen nearly 20 % since 2022.
- Regulators worldwide are scrambling: the EU AI Act bans predictive policing and emotion‑recognition, while the U.S. federal government under President Trump seeks to limit state AI regulation, even as 150 state AI bills pass, including California’s SB 53 and New York’s RAISE Act.
Why it matters: The report shows AI firms and data‑center operators reap massive revenue while straining power grids and water supplies, whereas young software developers see job declines; governments lag in regulation, leaving public safety, market fairness, and consumer trust at risk for the broader economy.
