Compute used to train frontier language models has grown roughly 5× per year since 2020
Epoch AI reports that the compute used for training frontier language models has increased by 5 times per year since 2020. This trend is based on the amount of compute used to train the top-5 models. Epoch AI notes that this scale-up comes from deploying more chips, longer training times, and more powerful processors.
Verified 2026-07-31
Track with InsiderWhy it matters
The computing used to build top models keeps growing fast, which drives demand for chips, power and data centres.
More on Compute
- Epoch AI tracks how much AI compute a dollar buys, measured in constant 2025 dollars across the chips companies actually purchased each quarter. Since 2023, AI chip price-performance has grown 49% a year, doubling every 1.7 years
- Amazon Web Services, the cloud unit that rents out computing power for AI, grew segment sales 39% year-over-year to $46.0 billion in the third quarter of 2026
- Applied Digital had signed long-term hyperscaler leases for 1,410 megawatts of AI data-centre capacity across five campuses as of May 31, 2026, worth approximately $36 billion in contracted revenue over the initial 15-year lease terms 1.41 gigawatts of AI data-centre capacity leased for $36 billion
The AI boom,with receipts.
6 sourced facts every weekday, plus the Weekly Brief, AI company deep dives and so much more. Free.
Free · No card · Unsubscribe anytime · See today's issue →
Join now and you start with 100 Receipts on The Receipts Ledger — your first step toward a free year of Insider or Pro. How it works →