Skip to content
Longterm Wiki
Back

Epoch AI: Compute Trends in AI Training

web

Credibility Rating

4/5
High(4)

High quality. Established institution or organization with editorial oversight and accountability.

Rating inherited from publication venue: Epoch AI

This Epoch AI dataset is frequently cited in AI governance discussions, including debates about compute-based regulatory thresholds (e.g., the EU AI Act's GPAI rules and US executive orders referencing 10^26 FLOPs as a frontier model threshold).

Metadata

Importance: 72/100dataset

Summary

Epoch AI's database tracking historical and current trends in computational resources used to train AI models. It provides empirical data on how training compute has scaled over time, enabling analysis of AI progress and forecasting future capabilities.

Key Points

  • Tracks training compute (in FLOPs) for major AI models over time, showing exponential growth trends
  • Data covers decades of AI development, illustrating the dramatic increase in compute since deep learning's rise
  • Useful for understanding hardware requirements, cost trends, and the relationship between compute and AI capabilities
  • Supports policy and governance discussions around compute thresholds for frontier AI models
  • Empirical foundation for arguments about compute as a key lever for AI oversight and regulation

Cited by 1 page

PageTypeQuality
EU AI ActPolicy55.0

Cached Content Preview

HTTP 200Fetched Mar 20, 20261 KB
# Page not found

The page you’re looking for doesn’t exist or has been moved.

[Go home](https://epoch.ai/)

Feedback

We value your privacy

Our website uses cookies to enhance your browsing experience and analyze site traffic. By clicking “Accept All,” you consent to our use of cookies as described in our [Privacy Policy](https://epoch.ai/privacy) and [Cookie Policy.](https://epoch.ai/cookies) If you wish to withdraw your consent, you can contact us at [ops@epoch.ai](mailto:ops@epoch.ai).

Accept AllReject All
Resource ID: b5265b94ee633a33 | Stable ID: MGQ2YWI2ND