Stanford just lately launched the 2026 AI Index, the sphere’s annual bodily. One discovering stopped me chilly: the nation that leads AI growth just isn’t the nation that leads AI adoption. I’ll come again to that. First, what is that this AI Index?
The AI Index is probably the most rigorous data-driven portrait of the place AI stands: a yearly checkup throughout technical efficiency, funding, the labor market, the atmosphere, public attitudes, regulation, the US-China race, and extra. 4 hundred pages, twelve headline takeaways, and a measurement equipment no different establishment has matched.
Many of the 2026 takeaways affirm what we already know or strongly suspect. AI efficiency is climbing, funding is exploding, the China hole has closed, younger software program engineers are shedding their jobs. Acquainted territory.
One quantity is shocking, although. The Index measures adoption (the share of a rustic’s inhabitants utilizing generative AI instruments) throughout two dozen economies within the second half of 2025. The leaders are usually not the international locations you’d guess.
The United Arab Emirates tops the listing at 64%. Singapore is second at 61%. Norway, Eire, and France spherical out the highest 5. Adoption correlates strongly with GDP per capita: richer international locations have higher infrastructure and extra data staff whose jobs profit from these instruments. That makes intuitive sense.
The US ranks twenty fourth, at 28.3%. That shocked me.
The nation rating determine, beneath, supplies the broader listing.
Here’s what makes the hole unusual.
In 2025, US personal funding in AI reached $285.9 billion, 23 instances China’s and greater than the remainder of the world mixed. Lots of the main fashions are educated in American labs. Even with expertise inflows down 89% since 2017, US researchers nonetheless outnumber every other nation’s by a large margin. By each supply-side measure, we’re the nation that builds AI.
By the Index’s adoption rating, we sit 23 locations behind the UAE, simply forward of the Czech Republic.
The hole just isn’t about ease of entry. Individuals can use the identical instruments, on the identical day, for a similar value (often zero) as anybody else.
The AI adoption vs GDP per capita scatterplot, beneath, plots every nation towards its earnings. The US sits about 13 factors beneath the development line, the most important hole of any rich nation.
The remainder of the report makes extra vital factors with some staggering statistics—for example, a leap in AI efficiency on cybersecurity benchmarks from 15% to 93%.
The unhealthy information
Seven findings give trigger for concern.
Vitality. Coaching and inference now eat gigawatts of electrical energy. Grok 4’s coaching run emitted 72,816 tons of CO2, the equal of 17,000 automobiles pushed for a yr, and international AI information middle capability reached 29.6 gigawatts, roughly New York State at peak demand.
Expertise flight. The variety of AI students relocating to the US has dropped 89% since 2017, with an 80% decline prior to now yr alone.
US-China parity. The efficiency hole between the highest American and high Chinese language fashions has closed to 2.7 share factors. China leads on publication quantity, patent output, and industrial robotic installations, aligned with what Cady and I predicted again in 2019.
Transparency collapse. The Basis Mannequin Transparency Index grades the key builders on how a lot they reveal about their fashions, from coaching information to downstream use. After two years of positive factors, the common fell from 58 to 40 in a single yr. Essentially the most succesful fashions disclose the least.
Entry-level squeeze. Employment amongst US software program builders aged 22 to 25 has fallen almost 20% from its 2022 peak. One in three organizations expects additional workforce reductions over the approaching yr.
College students lead however colleges lag. 4 in 5 US highschool and faculty college students now use AI for school-related duties. Half of center and excessive colleges haven’t any AI coverage, and solely 6% of lecturers say their faculty’s insurance policies are clear.
Public ambivalence. World optimism about AI rose to 59%, however solely 33% of Individuals anticipate AI to make their jobs higher, and solely 31% belief their authorities to manage it, the bottom determine of any nation surveyed.
The excellent news
4 findings are optimistic.
Technical efficiency. Frontier fashions now meet or exceed human efficiency on PhD-level science and competitors arithmetic, and the success charge of brokers on cybersecurity benchmarks jumped from 15% in 2024 to 93% in 2025.
Funding. World company AI funding hit $581.7 billion in 2025, up 130% yr over yr, with generative AI capturing almost half of all personal funding.
Science. AI-related publications within the pure, bodily, and life sciences rose 26% to twenty-eight% yr over yr. AI ran its first end-to-end climate forecasting pipeline, and astronomy constructed its first basis mannequin.
Medication. Medical-note instruments minimize doctor note-writing time by as much as 83% throughout a number of hospital programs.
The Index is at its most helpful when it surfaces a shock. This yr, that discovering is the diffusion hole: the nation that builds AI just isn’t the nation that makes use of it.
Why the hole? The report doesn’t say. However the adjoining findings trace at a solution. American staff anticipate AI to make their jobs worse. American voters mistrust the federal government that will regulate it. American corporations are deploying it extra slowly than corporations in China or Europe. Satirically, the nation that pioneered generative AI is underutilizing it. Quantity 24. Sheesh.
