Stanford recently released the 2026 AI Index, the field’s annual physical. One discovery stopped me cold: the world leading AI development it is not a country that leads AI adoption. I will come back to that. First, what is the AI Index?
The AI Index is the most robust data-driven picture of where AI stands: an annual assessment of technology performance, investment, the labor market, the environment, public attitudes, regulations, the US-China race, and more. Four hundred pages, taken from twelve articles, and measuring equipment no other institution has matched.
Much of the takeover of 2026 confirms what we already know or strongly suspect. AI performance is increasing, investment is exploding, the China gap is closing, new software engineers are losing their jobs. A known field.
One number is surprising, however. The Index measures the adoption (the share of the country’s population using productive AI tools) in all twenty economies in the second half of 2025. Leaders are not countries you can guess.
The United Arab Emirates tops the list with 64%. Singapore is second at 61%. Norway, Ireland, and France round out the top five. Adoption is highly correlated with GDP per capita: richer countries have better infrastructure and more knowledge workers whose jobs benefit from these tools. That makes intuitive sense.
The United States ranks 24th, at 28.3%. That scared me.
The country figure, below, provides a comprehensive list.
This is what makes the gap so confusing.
By 2025, US private investment in AI will reach $285.9 billion, 23 times that of China and more than the rest of the world combined. Many of the best models are trained in American labs. Even though talent is down 89% since 2017, US researchers still outnumber any other country by a wide margin. By all supply-side measures, we are a country that builds AI.
In terms of the adoption Index, we are sitting 23 places behind the UAE, just ahead of the Czech Republic.
The gap is not about accessibility. Americans can use the same tools, on the same day, for the same price (usually zero) as anyone else.
I AI adoption vs GDP per capita The scatterplot, below, plots each country against its income. The US sits nearly 13 points below the trend line, the largest gap among rich countries.
Another report makes more important points with some startling statistics—for example, the jump in AI performance in cybersecurity benchmarks from 15% to 93%.
Bad news
Seven findings give cause for concern.
Power. Training and explanation now consume gigawatts of electricity. The training run of Grok 4 emitted 72,816 tons of CO2, which is equivalent to 17,000 cars driven for a year, and the capacity of the global AI data center reached 29.6 gigawatts, almost in New York State with the highest demand.
A flight of talent. The number of AI scholars immigrating to the US has dropped 89% since 2017, and dropped 80% last year alone.
US-China parity. The performance gap between American top models and Chinese top models has closed to 2.7 percent. China is leading the way in publication volume, patent output, and industrial robotics installations, which is consistent with what Cady and I predicted back in 2019.
A wrap of transparency. The Foundation Model Transparency Index measures major developers on how much they reveal about their models, from training data to stream implementations. After two years of benefits, the average dropped from 58 to 40 in one year. The most talented models reveal the least.
Entry level compression. Employment of US software developers aged 22 to 25 is down nearly 20% from 2022. One in three organizations expect further redundancies in the coming year.
Students are earning but schools are failing. Four out of five US high school and college students are now using AI in school-related activities. Half of middle and high schools do not have an AI policy, and only 6% of teachers say their schools’ policies are clear.
Social disagreement. Global confidence in AI increased to 59%, but only 33% of Americans expect AI to improve their jobs, and only 31% trust their government to control it, the lowest of any country surveyed.
Good news
Four findings are positive.
Technical performance. Frontier models now meet or exceed human performance in PhD-level science and competitive statistics, and the success rate of agents in cybersecurity benchmarks jumped from 15% in 2024 to 93% in 2025.
Investment. Global corporate investment in AI will reach $581.7 billion by 2025, up 130% year-over-year, with productive AI capturing nearly half of all private equity.
Science. AI-related publications in the natural, physical, and health sciences increased by 26% to 28% year over year. AI ran its first end-to-end weather forecasting pipeline, and astronomy built its first base model.
Medicine. Clinical tools reduce physician note writing time by up to 83% across multi-hospital systems.
The Index is very useful when it finds a surprise. This year, what you get is a distribution gap: the country that builds AI is not the country that uses it.
Why is there a gap? The report does not say. But the immediate answers provide the answer. American workers expect AI to make their jobs worse. American voters do not trust the government to control it. American firms use it less than firms in China or Europe. Ironically, the country that invented manufacturing AI uses it the least. Number 24. Sheesh.
