Dedicated frontier labs hold under 25% with overall capacity below 50%.
Lisan al Gaib#1215
OpenAI kicked off the AI compute buildout in 2023. But today it uses ~10% of the world's compute, and the top labs together are probably under half.
In this week's newsletter, @justjoshinyou13 discusses how much that share may change, and when it could hit a ceiling.
🧵
Users thanked Epoch AI Research for its report clarifying that OpenAI alone uses 10% of global compute while frontier labs control under half.
If a few frontier players come to dominate the entire AI industry, then either compute growth at the top labs has to slow down, or the overall AI buildout has to accelerate. With AI capex approaching $1T/year, the latter would have dramatic economic implications.
Based on disclosures about data center power capacity and compute spend, the compute used by OpenAI, Anthropic, and xAI is likely <30% of the world total.
Google and Meta are giant hyperscalers, but much of their compute goes to cloud and recommenders, not their frontier labs.
@Jsevillamol does this imply slowdown sooner than Whitfill (2025) proposed?
Frontier AI labs collectively have access to less than 50% of all AI compute in the world (!)
Frontier labs will probably grow their share of AI compute in 2026. For example, Anthropic is moving aggressively to secure compute on the back of its historic revenue growth.
But top labs could absorb most of the headroom in a few years, after which growth is capped by chip production.
This week’s Gradient Update was written by @justjoshinyou13.
All Gradient Updates are informal and opinionated analyses that represent the views of individual authors, not Epoch AI as a whole.
Read the full essay here: https://epochai.substack.com/p/frontier-labs-dont-use-most-ai-compute
@EpochAIResearch @justjoshinyou13 weren't there 20M+ H100e at the end of 2025? Was the data revised?
@Pier4r @EpochAIResearch This graph applies a discount to those numbers to account for deployment vs sales, since the lab figures are deployed compute
@herbiebradley @Jsevillamol @whitfill_parker
@Jsevillamol Can you break out what % of RoW is Microsoft/Amazon? And further Oracle and neoclouds? I’m pretty curious if a headline is “<10 companies control >90%.”
@justjoshinyou13 @EpochAIResearch aha! Thank you
@EpochAIResearch @justjoshinyou13 AI scaling is turning compute into one of the most strategic resources on the planet.
Dedicated frontier labs hold under 25% with overall capacity below 50%.
Lisan al Gaib#1215OpenAI kicked off the AI compute buildout in 2023. But today it uses ~10% of the world's compute, and the top labs together are probably under half.
In this week's newsletter, @justjoshinyou13 discusses how much that share may change, and when it could hit a ceiling.
🧵