Model capabilities
15.5 ECI/year
Model capabilities have grown faster since early 2024.
Model Performance
90% CI +13 to +18 ECI
Model Performance
Model capabilities have grown faster since early 2024
The Epoch Capabilities Index tracks model capabilities across a range of evaluations, enabling analysis of long-term capabilities trends. ECI shows an increase in the pace of progress around April 2024, with state-of-the-art models improving by 15.5 points per year—about as much as the jump from GPT-4 to o1.
AI capabilities progress has sped up
|CC-BYepoch.aiCompute stock growth
3.4 ×/year
6.8 months
0.53 OOM/year
The total computing power of the stock of AI chips is growing at a rate of 3.4×/year.
The total computing power of the stock of AI chips is doubling every 6.8 months.
The total computing power of the stock of AI chips is growing by 0.53 OOMs per year.
AI Companies
90% CI 3.2× to 3.7× 6.4 to 7.2 months 0.51 to 0.57 OOM
AI Companies
The total computing power of the stock of AI chips is growing at a rate of 3.4×/year
The computing power of the total stock of AI chips has grown at 3.4x per year, doubling every 7 months since 2022, based on revenue data, other financial disclosures, and analyst reports.
Global AI computing capacity is doubling every 7 months
|CC-BYepoch.aiTraining compute
5 ×/year
5.2 months
0.7 OOM/year
Training compute for frontier language models has been growing at 5× per year since 2020.
Training compute for frontier language models has been doubling every 5.2 months since 2020.
Training compute for frontier language models has been growing at 0.7 OOMs per year since 2020.
Training Runs
90% CI 4× to 6× 4.6 to 6.0 months 0.6 to 0.8 OOM
Training Runs
Training compute for frontier language models has been growing at 5× per year since 2020
The amount of compute used to train frontier language models has grown exponentially. Since 2020, the trend among top-5 models has grown by a factor of ~10,000.
Training compute has grown 5x per year since 2020
|CC-BYepoch.aiBuild time
2.1 years
Gigawatt-scale data centers can be built in about 2 years.
Data Centers
90% CI 1 to 3.6 years
Data Centers
Gigawatt-scale data centers can be built in about 2 years
Based on 7 planned or in-progress data center projects expected to exceed 1 GW of total facility power, the median construction-to-gigawatt timeline is just over 2 years. Power estimates are informed by satellite imagery, permits, and company disclosures. These plans are subject to change.
|CC-BYepoch.aiLargest AI data center
1,100,000 H100e
The largest known AI data center has a computing capacity equivalent to 1.1 million NVIDIA H100 chips.
Data Centers
Data Centers
The largest known AI data center has a computing capacity equivalent to 1.1 million NVIDIA H100 chips
SpaceXAI’s Colossus 2 in Memphis, Tennessee has an estimated computing capacity of 1.1 million H100-equivalents, making it the largest known AI data center. Meta Hyperion is expected to have over 3x more computing capacity, at 3.7M H100-equivalents by January 2028.
Frontier AI data centers
|CC-BYepoch.aiFLOP/s per dollar
1.37 ×/year
2.2 years
0.14 OOM/year
AI chip performance per dollar has improved by 37% per year.
AI chip performance per dollar has doubled every 2.2 years.
AI chip performance per dollar has improved by 0.14 OOMs per year.
Hardware
90% CI 1.30× to 1.45× 1.9 to 2.6 years 0.11 to 0.16 OOM
Hardware
AI chip performance per dollar has improved by 37% per year
The compute performance you can buy for a dollar has improved by roughly 40% per year across over 20 AI accelerators released between 2012 and 2025. Much of this is driven by manufacturers introducing more powerful and more expensive chips — the GB300 costs nearly 9× the P100’s release price, but delivers about 24× the performance per dollar.
AI chip performance per dollar has improved by 1.4x per year
|CC-BYepoch.ai
