The Epoch AI Brief - September 2025
New AI Companies Data Hub, AI in 2030 report, Grok 4’s training footprint, implications of long-context inference, two new podcast episodes, and more.
Hello! In this edition of the Epoch AI brief:
We launched our AI Companies Data Hub tracking financial and operational metrics—revenue, funding, workforce, usage, and compute spending—for frontier AI companies.
We published a report forecasting AI’s trajectory through 2030, analyzing whether scaling will continue and what scientific R&D capabilities this enables.
We’ve published four new Data Insights covering GPQA Diamond scores accuracy, comparison of Grok 4’s training footprint, AI performance in high school math contests, and GPT model generation leaps.
We’ve published four new Gradient Updates on GPT-5 training compute, long-context inference implications, AI compute-based policy challenges, and possibility of a compute scaling slowdown due to lead times.
Two new Epoch After Hours podcast episodes featuring debates on AI progress forecasting through 2040 and economic theory’s role in understanding AGI and the potential “economic singularity”.
We’re hiring a Lead Editor to help expand our audience and make our research more accessible to decision-makers.
Publications & Announcements
New AI Companies Hub
The AI industry has changed rapidly in recent years, with frontier companies like OpenAI and Anthropic seeing fast exponential growth in their revenues and valuations. This growth has important implications for AI’s trajectory: AI companies are continually improving their technology by scaling up compute and labor inputs, and their revenue and usage track how AI is already impacting the world.
To help researchers, policymakers, and the public understand these trends, we have created our AI Companies Data Hub. The hub tracks financial and operational metrics—revenue, funding, staff, usage, and compute spend—for the key companies developing frontier AI models, along with interactive visualizations. This supplements our data hubs on AI models and GPU clusters and provides a more holistic view of the resource inputs and economic impact of the AI industry.
Here are some highlights from this new dataset:
Revenue — The combined revenue rates of OpenAI and Anthropic have grown around 10x since early 2024.
Funding — Collectively, the frontier AI companies in our data have raised ~$100B in equity and debt funding.
Usage — ChatGPT alone surpassed 700 million weekly active users by August 2025, processing over 3 billion daily messages.
Staff — OpenAI and Anthropic have both expanded from small startups to thousands of full-time staff, though they are still well behind Google’s flagship AI effort, Google DeepMind.
Compute spend — Compute for research, training, and inference is expensive: OpenAI’s cloud compute bill for 2025 will exceed $15 billion!
Start exploring the data now — and stay tuned as we expand our coverage to new companies and datasets!
What will AI look like 2030?
Our new report zooms in on two things: (1) whether scaling continues (compute, data, power, capital), and (2) the capabilities this enables—especially for scientific R&D.
On scaling — It is likely to continue till 2030. Building on our past work, we forecast that: Training clusters would cost hundreds of billions of dollars; Compute scaling is likely “not hitting a wall”; Synthetic & multimodal data may be needed to ease bottlenecks; and Power demands will increase but be manageable in principle.
On scientific R&D — Given that we expect scaling to continue, what does this mean for the resulting AI capabilities? We focus on scientific R&D—a stated priority for leading labs—and assess both benchmarks and real-world use, allowing us to forecast the kinds of tasks AI will be able to automate. Despite benchmarks’ weaknesses, progress has tracked real-world improvements, and usage already supports productivity gains. People already spend billions to use AI for coding, writing, and research. AI may be a transformative tool well before it can work autonomously. We explore future capabilities across four domains: software engineering, mathematics, biology and weather prediction.
Check out the full report for more!
Data Insights
Our Data Insights offer digestible snapshots of complex trends in AI. Since our August newsletter, we’ve published four new insights. Here’s what we’ve found:
AI developers accurately report GPQA Diamond scores for recent models
LLMs have not yet solved the hardest problems on high school math contests
GPT-5 and GPT-4 were both major leaps in benchmarks from the previous generation
Gradient Updates
Since our August newsletter, we’ve published four new issues of Gradient Updates, our weekly newsletter containing shorter-form research and commentary on important issues on AI:
Why GPT-5 used less training compute than GPT-4.5 (but GPT-6 probably won’t)
Three challenges facing compute-based AI policies (in collaboration with GDM’s AI Policy Perspectives Team)
Benchmarking updates
Sonnet 4.5 sets a new SOTA of 65% (±2%) on SWE-bench with our scaffold (based on SWE-agent). The new model beats Sonnet 4 by 4 percentage points.
We also benchmarked DeepSeek-V3.1 on SWE-Bench verified. It achieved below SotA at 52 ± 2.24%.
Podcast updates
We have two new episodes of the Epoch After Hours podcast:
Epoch AI’s own Anson Ho interviewed Dr. Philip Trammell about economic theory and AGI. Among other things, they discussed:
The value of economic theory in thinking about AGI
Phil’s attempts at detecting whether the “economic singularity” is coming
What’s wrong with existing work on explosive growth and the economics of AI
Jaime Sevilla and Yafah Edelman debate where current AI trends carry us —and where they break. They disagree on mechanisms and outcomes, but agree on this:
Fast diffusion by the end of the decade
Broad cognitive automation by ~2035
Extreme uncertainty after those timelines
Organizational Updates
Careers
We’re hiring a Lead Editor to help expand our audience and ensure our research informs decisions about AI’s future! We’re looking for someone deeply embedded in AI discussions, with a track record of making complex technical concepts engaging. The role is fully remote, and applications are rolling.
Subscribe only to the updates you care about
We’ve received consistent feedback from readers that they would like a way to subscribe and receive updates on all of our content. That’s why we’ve decided to bring all of our content into Substack.
We’ve introduced additional newsletters across a wider range of categories, mirroring the structure of our website. You’ll be able to subscribe to updates on Gradient Updates, Data Insights, Papers & Reports, Epoch After Hours, The Epoch Brief or general Announcements.
By default, Substack subscribes you to all of our content, but you can opt out of individual newsletters by clicking the Manage Subscription option on our Substack page.
great works