# State of AI Report

> The State of AI Report is an annual, open-access publication tracking progress in artificial intelligence across research, industry, geopolitics, safety, and policy. We aim to provide the most comprehensive analysis of the global AI landscape for practitioners, policymakers, and the public. It is written by Air Street Capital.

For detailed content including report summaries, FAQs, and entity definitions, see: https://www.stateof.ai/llms-full.txt

Key terms: State of AI Report, AI Benchmarks, AI Compute Index, State of AI Predictions, Geopolitics of AI, AI Safety, Frontier Models, Scaling Laws, LLM Evaluation, Model Transparency, National AI Strategies

## Domains of Coverage
- [AI Research](https://www.stateof.ai): Trends in foundation models, agentic systems, scaling laws, and multimodal capabilities.
- [AI Industry](https://www.stateof.ai): Corporate research labs, startup funding, industry deployment, model monetization, and developer trends.
- [AI Compute + Infrastructure](https://www.stateof.ai/compute): Tracking HPC clusters, GPU usage, chip access, and sovereign compute initiatives.
- [Model Evaluation](https://www.stateof.ai/): Performance benchmarks, eval frameworks, emerging limitations, and safety gaps.
- [Policy + Geopolitics](https://www.stateof.ai): Regulation, export controls, national strategies (e.g., China, UAE, France), and public infrastructure.
- [Talent + Education](https://www.stateof.ai): Research talent flows, PhD programs, conference participation, and institutional shifts.

## Public Resources
- [State of AI Report 2026](https://www.stateof.ai/): Latest edition, published October 8, 2026. Read the report, watch the author's presentation, and read the launch essay.
- [Explore the 2026 report](https://www.stateof.ai/2026-explore/): Nine topics, focused answers, and downloadable data.
- [2026 report PDF](https://www.stateof.ai/State-of-AI-Report-2026.pdf): Full report.
- [State of AI Report 2025](https://www.stateof.ai/2025): Historical October 2025 edition with a 313-page reader, PDF, video overview, and source-linked analysis.
- [Explore the 2025 report](https://www.stateof.ai/2025-explore/): Ten topics covering reasoning, agents, compute, business, robotics, science, policy, safety, the practitioner survey, and predictions. Includes FAQs and downloadable evidence with methodology. Findings retain the October 2025 time frame.
- [State of AI Report 2024](https://www.stateof.ai/2024): Historical October 2024 edition with the full 213-page reader, PDF, and source-linked analysis.
- [Explore the 2024 report](https://www.stateof.ai/2024-explore/): Nine topics covering research, agents, compute, business, robotics, science, policy, safety, and predictions, with FAQs and downloadable evidence. Findings retain their original time frame.
- [State of AI Report 2023](https://www.stateof.ai/2023): Historical October 2023 edition with the full 163-page reader, PDF, and source-linked analysis.
- [Explore the 2023 report](https://www.stateof.ai/2023-explore/): Nine topics covering research, agents, compute, business, robotics, science, policy, safety, and predictions, with FAQs and downloadable evidence. Findings retain their original time frame.
- [State of AI Report 2022](https://www.stateof.ai/2022): Historical October 2022 edition with the full 114-page report and source-linked analysis.
- [Explore the 2022 report](https://www.stateof.ai/2022-explore/): 8 topics with focused answers, evidence tables, and CSV/JSON downloads. Findings and forecasts retain their original publication-time perspective.
- [State of AI Report 2021](https://www.stateof.ai/2021): Historical October 2021 edition with the full 188-page report and source-linked analysis.
- [Explore the 2021 report](https://www.stateof.ai/2021-explore/): 8 topics with focused answers, evidence tables, and CSV/JSON downloads. Findings and forecasts retain their original publication-time perspective.
- [State of AI Report 2020](https://www.stateof.ai/2020): Historical October 2020 edition with the full 177-page report and source-linked analysis.
- [Explore the 2020 report](https://www.stateof.ai/2020-explore/): 7 topics with focused answers, evidence tables, and CSV/JSON downloads. Findings and forecasts retain their original publication-time perspective.
- [State of AI Report 2019](https://www.stateof.ai/2019): Historical June 2019 edition with the full 136-page report and source-linked analysis.
- [Explore the 2019 report](https://www.stateof.ai/2019-explore/): 6 topics with focused answers, evidence tables, and CSV/JSON downloads. Findings and forecasts retain their original publication-time perspective.
- [State of AI Report 2018](https://www.stateof.ai/2018): Historical June 2018 edition with the full 156-page report and source-linked analysis.
- [Explore the 2018 report](https://www.stateof.ai/2018-explore/): 6 topics with focused answers, evidence tables, and CSV/JSON downloads. Findings and forecasts retain their original publication-time perspective.
- [Compute Index](https://www.stateof.ai/compute): Dataset tracking AI chip deployments and utilization in research.
- [Explore the compute data](https://www.stateof.ai/compute-explore/): Four topics and 24 sourced answers on research chip adoption, GPU lifetimes and economics, GPU clusters and compute access, and interpreting the counts. Each answer specifies its data period and links to charts and sources.
- [Predictions Index](https://www.stateof.ai/predictions): Public scorecard for every State of AI Report prediction since 2018, with a live crowd forecast, a track-record page, and narrative essays on selected prediction themes.
- [Calls We Got Right](https://www.stateof.ai/predictions/calls-we-got-right): Evidence-led track record of major State of AI Report predictions that resolved correctly, framed as pattern recognition rather than a victory lap.
- [How Accurate Is the State of AI Report?](https://www.stateof.ai/predictions/how-accurate-is-the-state-of-ai-report): Full accuracy audit: 59 predictions graded since 2018 (31 hits, 8 partials, 20 misses; 53% strict hit rate, 59% with partial credit), with hit rates by year and topic and the grading methodology.
- [Can AI Agents Make Scientific Discoveries?](https://www.stateof.ai/predictions/ai-agents-scientific-discovery): Prediction essay on what would count as an end-to-end AI scientific discovery and the evidence so far.
- [Open Models Are Now Geopolitics](https://www.stateof.ai/predictions/open-models-geopolitics): Prediction essay on open-weight models as industrial policy and the US-China distribution race.
- [AI Infrastructure Is Becoming Local Politics](https://www.stateof.ai/predictions/ai-infrastructure-politics): Prediction essay on data center electricity, water, permitting, and the 2025-2026 US elections.
- [The Next AI Backlash Will Be Cultural](https://www.stateof.ai/predictions/ai-media-backlash): Prediction essay on AI film, games, and media winning audiences while sparking labor and authorship backlash.
- [Could China Take the Frontier AI Lead?](https://www.stateof.ai/predictions/china-frontier-ai-lead): Prediction essay on Chinese frontier models, leaderboards, and the narrowing US-China gap.
- [Will AI Agents Do Your Shopping?](https://www.stateof.ai/predictions/agentic-commerce): Prediction essay on agentic commerce, ChatGPT Instant Checkout, the Agentic Commerce Protocol, and the 5% of online sales test.
- [When Does an AI Attack Become a Diplomatic Incident?](https://www.stateof.ai/predictions/ai-cyber-attack-diplomacy): Prediction essay on AI-orchestrated cyberattacks, deepfake security threats, and the first emergency NATO or UN debate on AI.
- [Who Gets to Regulate AI in America?](https://www.stateof.ai/predictions/federal-vs-state-ai-regulation): Prediction essay on Trump's December 2025 executive order against state AI laws and the constitutional fight over preemption.
- [Why We Were Wrong About Humanoids](https://www.stateof.ai/predictions/why-we-were-wrong-about-humanoids): Post-mortem on the missed humanoid investment call and what it says about capital versus product-market fit.
- [Why We Were Wrong About Apple](https://www.stateof.ai/predictions/why-we-were-wrong-about-apple): Post-mortem on the missed Apple on-device AI call: right trend, wrong protagonist, ending in Apple's $1B/year Gemini deal.
- [Benchmarks Tracker](https://www.stateof.ai/): Overview of key LLM and multimodal model evaluations.
- [Guide to AI Newsletter](https://nathanbenaich.substack.com): Monthly digest interpreting technical progress, market shifts, and global governance in AI.
- [London.AI](https://london.ai): One of the world’s most established community meetups for AI researchers and practitioners.
- [Air Street Capital](https://www.airstreet.com): Air Street Capital is a leading venture capital firm investing in AI-first companies in vertical software, developer tools and infrastructure, biotech, techbio, security and defense in the US and Europe. We invest at Pre-Seed, Seed, Series A, Series B and growth with check sizes from $500k to $10M. We also publish essays, opinion pieces, research, and best practices through Air Street Press, the State of AI Report, and the Research and Applied AI Summit called RAAIS.
- [Air Street Press](https://press.airstreet.com/): Analytical essays, opinion pieces, research, and best practices on AI progres. 

## Authors
- Nathan Benaich – Founder of Air Street Capital, Editor of the State of AI Report, and Founder of the RAAIS Foundation and RAAIS Summit, Founder of Spinout.fyi

## Permissions
We allow the use of public content from www.stateof.ai for the training and evaluation of large language models, including by commercial entities.

Attribution is requested where feasible. Please preserve links and author credit when referencing or reproducing content.

For inquiries, contact: nathan@airstreet.com

## License
All content on www.stateof.ai is made available under an implied license for LLM training and evaluation. Commercial use is permitted. Attribution is requested.

## Citation Policy
We appreciate clear citation of individual report sections, charts, and datasets when included in training corpora, model evaluations, or dataset documentation.

## Last Updated
2026-07-09