The Stanford AI Index 2026, in 18 slides

AI is scaling faster than the systems around it can adapt

Every spring, Stanford's Institute for Human-Centered AI (HAI) publishes the AI Index Report, an independent, 400-plus-page audit of where artificial intelligence actually is rather than where Silicon Valley marketing says it is. The 9th edition landed this month, and the headline from the report's co-chairs is hard to shake:

“AI is scaling faster than the systems around it can adapt.”

I've distilled the report into a mobile-friendly carousel — 18 portrait slides — and it is available below to swipe through.

What's in the carousel

The 2026 Index tracks eight domains. My summary follows the same structure but pulls out the numbers I think matter most for leaders.

Capability and adoption. Organisational adoption of generative AI has reached 88%. Roughly 53% of the global population has used an AI tool within three years — faster uptake than the PC or the internet. More than 90% of notable frontier models in 2025 came from industry, not academia.

Global competition. The US–China performance gap has effectively closed. The US still leads on investment ($285.9B in 2025, 23 times China's disclosed figure) and infrastructure (5,427 AI data centres, ten times any other country). China leads on publication volume, patent output, and industrial robot installations. DeepSeek-R1 briefly matched the top US model in February 2025.

The jagged frontier. Gemini Deep Think won gold at the International Mathematical Olympiad. The same class of model reads an analogue clock correctly just 50.1% of the time. Robots succeed at household tasks only 12% of the time outside the lab.

Economy and labor. Productivity gains of 14–26% in customer support and software development. At the same time, employment of US developers aged 22–25 fell nearly 20% from 2024, even as headcount for older developers grew.

Responsible AI. 362 documented AI incidents in 2025, up from 233 the year before. Reporting of responsible-AI benchmarks is far less consistent than reporting of capability benchmarks. Grok 4's training produced an estimated 72,816 tonnes of CO₂-equivalent emissions.

Science, medicine, and education. Frontier models now outperform human chemists on ChemBench on average. Physicians using AI scribes report up to 83% less time writing clinical notes. At the same time, only 5% of clinical AI studies used real patient data, and only 6% of teachers say their AI policies are clear.

Policy and public opinion. 73% of AI experts expect AI to have a positive impact on jobs. Only 23% of the general public agrees. That 50-point gap is arguably the single most important governance problem in the report.

Why this matters

Two questions I would put to any leadership team after reading this.

First, how is your organisation moving beyond 88% adoption to actual productivity transformation?

Adoption is no longer the moat. Almost everyone has the tools. The differentiator now is whether you can re-engineer the work around them.

Second, are you monitoring the jagged frontier? The gap between what AI is brilliant at and what it is unreliable at is wider than most risk frameworks assume. Knowing where that edge sits — for your tasks, your data, and your regulatory environment — is a strategic capability in its own right.

Source

All data in the carousel and this post is drawn from the Stanford HAI AI Index Report 2026 (9th Annual Edition), published April 2026.