Who's winning the AI race?

As of -- UTC

There are thousands of evals to judge the best AI. But the best one is: What are people paying for? Using OpenRouter usage data, we can calculate the API revenue of the top models:

Weekly OpenRouter Revenue by Model

Show

Estimated revenue per model using OpenRouter pricing, based on weekly token usage. Revenue assumes 1.5% completion tokens and 98.5% prompt tokens. The historical dataset includes an “Others” bucket that aggregates models without per-model breakdowns. Those tokens are excluded from lab revenue estimates.

And from this, we can get a rough estimate of the relative API usage dominance of the different labs:

Weekly OpenRouter Revenue by Lab

Estimated revenue per lab using OpenRouter pricing, based on weekly token usage. Revenue assumes 1.5% completion tokens and 98.5% prompt tokens. The historical dataset includes an “Others” bucket that aggregates models without per-model breakdowns. Those tokens are excluded from lab revenue estimates.

Normalizing to 100% makes it easier to see how each lab's share of the pie is changing over time:

OpenRouter Revenue Market Share by Lab

Each lab's estimated revenue as a percentage of total weekly OpenRouter revenue. Same methodology as the absolute revenue chart above.

We also need to look into the future to see what kind of compute each lab will have access to. Using satellite imagery and public records, Epoch AI tracks the frontier data centers being built by each lab. Here's how much compute each lab will have access to over the next few years:

Frontier Datacenter Compute Available by Lab

Data from Epoch AI, licensed under CC BY 4.0. Covers major frontier data centers tracked via satellite imagery and public records. Labs with primarily cloud/distributed compute (e.g., DeepSeek, Mistral) may not appear. Facility-to-lab attribution confidence varies — hover for details. Chart includes projected future capacity based on construction progress.

Frontier Datacenter Compute Market Share by Lab

Each lab's share of total tracked training compute capacity over time. Same data source as the absolute compute chart above.

Weekly New App Ratings

Rating counts from Google Play and Apple App Store. iOS ratings weighted 3× to approximate download parity (iOS users rate apps less frequently than Android users). Weekly change in cumulative ratings is used as a proxy for download velocity.