Research institute · Research orgs · San Francisco, CA · Est. 2021
Epoch AI is a data-first research nonprofit that investigates the future trajectory and societal implications of artificial intelligence. Its stated mission is to help people understand what is happening in AI from a neutral perspective grounded in evidence, by studying AI’s key drivers and bottlenecks, tracking capabilities over time via evaluations and benchmarks, and forecasting downstream impacts on the economy and society.
The organization operationalizes this mission through open data programs and benchmarking infrastructure: it curates and analyzes datasets on AI models and infrastructure (including compute, hardware, and data centers), publishes frequent short analyses intended to be accessible and citable, and maintains benchmarking tools and datasets that can be used by researchers, developers, and policymakers. In 2025, Epoch AI reported launching and expanding several major initiatives, including a revamped benchmarking hub and the Epoch Capabilities Index (ECI), completing FrontierMath Tier 4 commissioned by OpenAI, releasing its GATE macroeconomic framework for AI automation effects, and expanding its data-center tracking (including a GPU Clusters Data Explorer and Frontier Data Centers Data Explorer). In addition to open research output, Epoch AI states it offers commissioned research, model evaluations, and consultations for companies, nonprofits, and government bodies, and it maintains a public transparency page listing donations of $70,000+ and certain collaborations/consultations.
No curated programs yet. Check the careers page ↗.
Whether AI can do our jobs, how far behind open-weight models are in cybersecurity, and Gemini closing in on the frontier
Can AI automate AI R&D yet?Not yet, but it will claim otherwise
The Epoch Brief - October 8, 2026Plummeting AI costs, billions of agents, how Chinese labs make money, AI’s impact on behaviors, and our benchmark auditing initiative
Hundreds of millions of AI agents are coming. Is there work for them?The AI compute buildout could support more AI agents than the US has workers
AI is getting cheaper faster than any other transformative technologyThe price is falling 13x year over year
How far behind Nvidia is Huawei?Huawei will substantially improve its AI compute solutions by 2030, but export controls constrain its most important scaling levers, making it unlikely to catch up with Nvidia.
The Epoch Brief - September 12, 2026Huawei's long odds against Nvidia, the GDP growth we aren't measuring, GPT-6 Astra sets records, a new math benchmark, and more
An update on AI’s most important numberHow quickly are Anthropic and OpenAI growing their revenue?
The Nvidia-sized hole in US GDP statisticsGDP growth is understated by about 0.3 percentage points because of missing value from fabless chipmakers, primarily Nvidia.
9 Big questions benchmarks can help answerA good benchmark asks more than just “can AI do this specific task?”
Will financing bottleneck AI compute? An Anthropic case studyWhy Anthropic's buildout suggests that financing is unlikely to be the immediate blocker to frontier AI compute growth.
The Epoch Brief - July 31, 2026Expanding FrontierMath: Open Problems, how "parallelizability" determines a technological singularity, the realities of AI energy use, and signs of AI uplift
Epoch AI reports a rise in the share of arXiv math preprints that acknowledge AI use from 4% in April 2026 to 25% in August 2026, including discussion of classification choices for “substantial research” versus research assistance.
Trade data is consistent with more than $3 billion of chips smuggled into China via MalaysiaEpoch AI analyzes server import/export trade data patterns between Malaysia and China and reports a pattern consistent with more than $3B worth of chips smuggled into China via Malaysia, and provides discussion of uncertainty and implications.
GPT-6 Astra leads on math benchmarks, but not on software engineeringEpoch AI reports GPT-6 Astra as topping the Epoch Capabilities Index (ECI) and provides a breakdown where it highlights differences between math and software-engineering subsets within the ECI framework.
Scaling Our Global AI Data Center ResearchEpoch AI reports that its AI Data Centers explorer now covers an estimated 44% of global AI compute across 86 sites, with a redesigned interface and expanding coverage.
Near-daily AI use among US adults has more than doubled in six monthsBased on Epoch AI/Ipsos polling, the organization reports that the share of US adults using AI at least 6 days a week rose from 8% in March 2026 to 19% in August 2026.
ExploitBench v0.1 – Benchmark ReviewEpoch AI publishes a rubric-based audit review for ExploitBench v0.1 and reports a verdict and key quality findings (as presented on the review page).