Global AI computing capacity is doubling every 7 months
A · Strong evidence Epoch AI data insight with downloadable chip-level data, a stated regression method and confidence intervals; short, but every step is checkable.
Summary based on the full text·Original: about 5 minutes to read
In short
A January 2026 data insight by Josh You, Venkat Somala, Yafah Edelman and Luke Emberson at Epoch AI. Total computing capacity from AI chips across all major designers has grown about 3.3x a year since 2022, a doubling every seven months. NVIDIA accounts for over 60% of the total, with Google and Amazon most of the rest.
Key numbers
- 3.3x per yeargrowth in global AI computing capacity since 2022 (90% CI: 2.7x to 4.1x)
- 7 monthsimplied doubling time of global AI compute (90% CI: 6 to 8 months)
- over 60%NVIDIA chips' share of total AI compute
- less than 10%share of chip production from manufacturers other than Nvidia and Google as of 2024Q1
As reported by the source.
English·Web page·opens epoch.ai
What it reports
- Global AI compute, measured in H100 equivalents, grew about 3.3x a year since 2022 (90% CI 2.7x to 4.1x).
- That is a doubling time of 7 months (90% CI 6 to 8 months).
- NVIDIA chips are over 60% of total compute. Google TPUs and Amazon chips make up most of the remainder.
- The page publishes a chip-by-quarter table with unit estimates, H100-equivalent compute, power draw and cost estimates.
How they got it
Chip sales estimates from Epoch's AI Chip Sales data hub, built mainly from company financial documents, converted to H100 equivalents. A log-linear regression on quarterly H100-equivalent sales gives the growth rate, with analytical 90% confidence intervals from the slope's standard error. The data can be downloaded as CSV.
Limits
- Nvidia data start in 2022Q1, Google in 2022Q4 and others only in 2024Q1. Epoch does not correct for the gap but says those makers were under 10% of production.
- The data track chip sales, not deployments, and ignore depreciation and the lag between sale and use.
- Unit counts are estimates with wide 5th to 95th percentile ranges.
Why it matters for companies
Compute supply is growing far faster than most enterprise AI budgets or adoption rates, so the binding constraint for a typical company is integration, data and skills, not access to models. The concentration figure matters for vendor risk: most capacity runs on one chip maker's hardware and a few cloud providers, which affects pricing power and resilience planning.
Summary based on the full text. Read the original for the full argument and data.Read the original on epoch.ai
- Tags
- AI capability