Stanford's Institute for Human-Centered AI (HAI) released the 2026 AI Index Report today, and its headline finding cuts against a year of plateau chatter: AI capability is still accelerating. But the accompanying data also paints a portrait of concentration — of compute, capital, and consumer value — that raises uncomfortable questions about who actually benefits from the boom.
Capability is still climbing
The report documents breakthrough results across science and complex reasoning benchmarks, with frontier models making gains that outpace what many analysts expected a year ago. A central storyline is the narrowing US–China gap: American and Chinese models have swapped the top spot on major leaderboards repeatedly since early 2025, and as of March 2026 Anthropic's leading model edges out its closest Chinese competitor by just 2.7%.
The flip side is transparency. The Foundation Model Transparency Index, which grades how openly labs disclose training data, compute, capabilities, and risk information, fell to an average score of 40 — down from 58 the prior year. The most capable models, the report notes, are increasingly the least transparent.
Money and megawatts
Global corporate AI investment hit $581.7 billion in 2025, up roughly 130% year over year. The United States accounted for $285.9 billion of that — about 23 times China's $12.4 billion, though Chinese progress suggests capital efficiency matters more than raw spend.
The infrastructure bill is getting hard to hide. AI data-center power capacity has reached 29.6 gigawatts, a draw the report compares to peak demand for all of New York state. Water use is climbing too: annual inference for a model like GPT-4o may consume more drinking water than 12 million people use in a year, according to the report's estimates.
Adoption outpaces governance
Generative AI reached 53% population penetration within three years, faster than either the personal computer or the internet. The estimated value of GenAI tools to US consumers hit $172 billion annually by early 2026, with median per-user value tripling since 2025.
Education is where the governance lag is starkest. Four in five US high school and college students now use AI for schoolwork, but only half of middle and high schools have AI policies in place — and just 6% of teachers say those policies are clear.
Sentiment is more complex than the headlines suggest
Global optimism about AI's benefits rose to 59% (from 52%), but nervousness ticked up to 52% as well. That isn't contradiction so much as ambivalence — the public is warming to the upside and the downside simultaneously.
Why it matters
The Index's cumulative message is that the narrative of an "AI plateau" doesn't survive contact with the data. Capability is compounding, adoption is compounding, and costs — in dollars, watts, and water — are compounding with them. The open question, the report implies, isn't whether AI keeps advancing. It's whether the gains spread beyond the handful of labs and countries currently capturing them.
