41T tokens served in a day: startup claims first quadrillion tokens/month for open-weight models

ycombinator · x · 2026-11-18

thdxr (amplified by the Y Combinator account) reports his inference business served 41 trillion tokens in a single day, and claims they'll be the first to process a quadrillion tokens per month serving open-weight models. The figure offers a rare concrete datapoint on how much real production traffic open-weight model inference now handles, and signals that open-weight inference is scaling into a token-throughput business at massive volume.

Original post →

More from Infra

Infra channel →