AI-Wiki

Artificial Intelligence Index Report 2026

10 min read

Artificial Intelligence Index Report 2026 (9th edition)

TL;DR

The 9th annual AI Index from Stanford HAI. Cite as: Sajadieh et al. 2026.

Headline framing (from Co-chairs Yolanda Gil and Raymond Perrault):

“The data does not point in a single direction. It reveals a field that is scaling faster than the systems around it can adapt.”

The report tracks “the gap between what AI can do and how prepared we are to manage it” — running through every chapter. New in 2026: standalone chapters on AI in Science (Ch 5) and AI in Medicine (Ch 6, in collaboration with Schmidt Sciences); an analytical framework on AI sovereignty; new estimates of generative AI’s economic value alongside emerging labor-market evidence.

This page summarizes the front matter and the 15 Top Takeaways; per-chapter deep-reads of the ~430-page report are deferred.

What’s new vs. the 2025 edition

  • Chapter structure: Medicine split out from Science (was combined as one chapter in 2025); ninth chapter added.
  • Editorial leadership:
    • Co-chairs: Yolanda Gil (USC, ISI) — new chair, was chair-elect in 2025; Raymond Perrault (SRI International) — was chair in 2025, now co-chair.
    • Editor-in-Chief: Sha Sajadieh (Stanford) — replaces Nestor Maslej who led the 2025 edition (Maslej remains a contributor).
    • Steering Committee additions: Vipin Kumar (U Minnesota), Dan Weld (U Washington), Elham Tabassi (Brookings).
    • Departures from steering committee vs 2025: John Etchemendy (Stanford), Katrina Ligett (Hebrew U).
  • New supporting partner: Infosys; remaining: Google, NSF, OpenAI, Open Philanthropy, Quid.
  • New analytics/research partner: Schmidt Sciences (collaborated on Medicine chapter).
  • Same data partners as 2025, plus Zeki (was already there, may have been newer).

The 15 Top Takeaways

1. AI capability is not plateauing

  • Industry produced >90% of notable frontier models in 2025.
  • SWE-bench Verified: 60% → near 100% in one year.
  • Several models meet/exceed human baselines on PhD-level science questions, multimodal reasoning, competition mathematics.
  • Organizational adoption: 88% (up from 78% in 2024, per the 2025 edition).
  • 4 in 5 university students now use generative AI.

2. The U.S.-China AI model performance gap has effectively closed

  • DeepSeek-R1 briefly matched the top U.S. model in February 2025.
  • As of March 2026, Anthropic’s top model leads by just 2.7%.
  • U.S. still produces more top-tier models and higher-impact patents.
  • China leads in publication volume, citations, patent output, industrial robot installations.
  • South Korea leads in AI patents per capita.

3. United States hosts the most AI data centers, fabricated by one Taiwanese foundry

  • 5,427 U.S. data centers — more than 10× any other country.
  • U.S. AI data center energy consumption higher than any other country.
  • TSMC fabricates almost every leading AI chip → global supply chain dependent on one foundry in Taiwan.
  • TSMC-U.S. expansion began operations in 2025.

4. The “jagged frontier” is now an AI Index narrative

  • Gemini Deep Think earned a gold medal at IMO, yet the top model reads analog clocks correctly only 50.1% of the time.
  • AI agents on OSWorld: 12% → ~66% task success in one year (real computer tasks across operating systems).
  • Still fail roughly 1 in 3 attempts on structured benchmarks.
  • This is the same concept introduced empirically in Dell’Acqua et al. 2026 — see jagged-frontier.

5. Robots still fail at most household tasks

  • Robots succeed in only 12% of household tasks.
  • RLBench robotic manipulation (software-based simulations): 89.4% success.
  • Wide gap between predictable lab settings and unpredictable household environments.

6. Responsible AI not keeping pace with capability

  • Documented AI incidents: 362 in 2025, up from 233 in 2024.
  • Almost all leading frontier developers report capability benchmarks; reporting on responsible AI benchmarks remains spotty.
  • Recent research finds improving one RAI dimension (e.g., safety) can degrade another (e.g., accuracy) — a tension previously implicit but now empirical.

7. United States leads in AI investment, but talent attraction is declining

  • U.S. private AI investment: $285.9B in 2025, ~23× China’s $12.4B.
    • For context, the 2025 edition reported $109.1B for U.S. private investment in 2024 — a ~2.6× year-on-year increase.
  • 1,953 newly funded AI companies in U.S. in 2025 — more than 10× the next country.
  • AI researchers/developers moving to the U.S. dropped 89% since 2017, with an 80% decline in the last year alone.
  • China’s $12.4B figure likely understates total spending given government guidance funds.

8. AI adoption spreading at historic speed

  • Generative AI: 53% population adoption within three years — faster than the PC or the internet.
  • Adoption correlates strongly with GDP per capita.
  • Outliers: Singapore (61%), UAE (54%) higher than expected; U.S. ranks 24th at 28.3%.
  • Estimated value of GenAI tools to U.S. consumers: $172B annually by early 2026.
  • Median value per user tripled between 2025 and 2026.

9. Productivity gains and entry-level employment decline coincide

  • Productivity gains of 14–26% in customer support and software development.
  • Weaker or negative effects in tasks requiring more judgment.
  • AI agent deployment remains in single digits across nearly all business functions.
  • U.S. software developers ages 22–25: employment fell nearly 20% from 2024, even as headcount for older developers continues to grow.
  • Mirrors the Brynjolfsson Canaries paper’s finding via different methodology and a more recent timeframe; see ai-employment-effects.

10. AI’s environmental footprint is expanding

  • Grok 4 training emissions: 72,816 tons CO₂eq.
  • AI data center power capacity: 29.6 GW — comparable to New York state at peak demand.
  • Annual GPT-4o inference water use alone may exceed the drinking water needs of 1.2 million people.

11. AI for science can outperform human scientists, but bigger isn’t always better

  • Frontier models outperform human chemists on average on ChemBench.
  • Score <20% on replication in astrophysics, 33% on Earth observation questions.
  • MSAPairformer (111M params) beat previous protein methods on ProteinGym.
  • GPN-Star (200M params, genomics) outperformed a model nearly 200× larger.
  • Most AI foundation models for science come from cross-sector collaborations, contrasting with the industry-dominated landscape of general-purpose AI.

12. AI is transforming clinical care, but rigorous evidence remains limited

  • Ambient AI scribes saw substantial 2025 adoption: physicians reported up to 83% less time spent writing notes; significant burnout reductions.
  • Of 500+ clinical AI studies reviewed: nearly half rely on exam-style questions rather than real patient data; only 5% use real clinical data.
  • The evidence base for clinical AI remains thin despite mass deployment.

13. Formal education lagging; people learning AI skills at every life stage

  • >80% of U.S. high school and college students use AI for school-related tasks.
  • Only half of middle and high schools have AI policies; just 6% of teachers say those policies are clear.
  • AI engineering skills accelerating fastest in UAE, Chile, South Africa.
  • New AI PhDs in U.S. and Canada +22% from 2022 to 2024 — and the new PhDs took jobs in academia, not industry (a notable reversal).

14. AI sovereignty as a defining feature of national policy

  • Expansion of national AI strategies, particularly in developing economies.
  • State-backed AI supercomputing investments rising in parallel.
  • Open-source contributions from rest-of-world now outpace Europe and approach the U.S. on GitHub.
  • Fueling more linguistically diverse models and benchmarks.

15. Expert-public trust gap is large; institutional trust is fragmented

  • On AI’s impact on jobs: 73% of experts expect a positive impact vs 23% of public — a 50-point gap.
  • Globally, the EU is trusted more than the U.S. or China to regulate AI effectively.
  • The U.S. reported the lowest level of trust in its own government to regulate AI (31%).

Key claims by chapter (highlights only)

Chapter 1 — Research and Development

  • Industry produced 91.18% of notable AI models in 2025; Industry-academia collaboration 4.90%; Other 1.96%; pure Academia 1.96%.
  • 2025 by country: United States 59 notable models, China 35, South Korea 8.
  • 2025 by organization: OpenAI 20, Google 14, Alibaba 11, Anthropic 7, xAI 5, DeepSeek 4, LG AI Research 4, Meta 4, Tsinghua University 4 (the only academic institution in the top 9).
  • Disclosure dropping: training code, dataset sizes, parameter counts increasingly withheld for the most capable systems.
  • Parameters stable in the trillion-parameter range for three years; training compute (estimable independently) continues to rise.
  • Synthetic data still not replacing real data in pre-training, but OLMo 3.1 Think 32B with ~90× fewer parameters than Grok 4 achieves comparable benchmark results via pruning, deduplication, curation alone.
  • Global AI compute capacity: 17.1M H100-equivalents, growing 3.3× per year since 2022.
  • Nvidia: >60% of total compute. Google and Amazon supplying much of the remainder; Huawei holding small but growing share.
  • Open source: 5.6M projects on GitHub; Hugging Face uploads tripled since 2023; U.S.-based projects still attract most engagement (30M cumulative GitHub stars on projects past the 10-star threshold).
  • Talent: AI researchers/developers moving to U.S. -89% since 2017, -80% last year alone.
  • Per-capita researcher density: Switzerland and Singapore lead.
  • Female AI researcher representation: Saudi Arabia 32.3%, Canada 29.6%, Australia 30.1% — no country approaches parity.

Chapters 2–9: not yet ingested

Chapters 2 (Technical Performance), 3 (Responsible AI), 4 (Economy), 5 (Science), 6 (Medicine — new), 7 (Education), 8 (Policy and Governance), 9 (Public Opinion) span ~360 of the report’s pages and are deferred to subsequent ingests.

Methodology notes

  • AI tooling acknowledgement: “The AI Index is written by a team of human researchers. The authors used ChatGPT and Claude to help refine and copy edit drafts. All images in this publication were generated with AI by Johanna Friedman (2026), Gemini 3.1 (W-Nanobanana2), Gemini 3 (W-Nanobanana Pro).” — visibly more AI-assistance disclosure than the 2025 edition.
  • Notable models dataset: still curated by Epoch AI; year-over-year counts may differ from prior reports as new/historic models are continually added.
  • License: CC BY-ND 4.0 (no derivatives) — same as 2025.

Quotes worth saving

“A year ago, this report documented AI’s arrival as a mainstream force. This year’s data shows what happens after arrival.” — Co-chairs

“The data does not point in a single direction. It reveals a field that is scaling faster than the systems around it can adapt.” — Co-chairs

“AI capability is not plateauing. It is accelerating and reaching more people than ever.” — Top Takeaway #1

“AI models can win a gold medal at the International Mathematical Olympiad but cannot reliably tell time — an example of what researchers call the jagged frontier of AI.” — Top Takeaway #4

“Productivity gains from AI are appearing in many of the same fields where entry-level employment is starting to decline.” — Top Takeaway #9

Limitations / caveats (per report)

  • Notable-model counts depend on Epoch AI’s curation; not a census of all models.
  • Disclosure decline among frontier labs makes some technical-performance metrics estimates rather than ground truth.
  • Most-cited AI papers (top 100) lag 1–2 years; 2024 paper data is as recent as available.
  • Public-opinion data gathered in survey years that may pre-date the latest model releases.

Graph View

Backlinks