Anthropic Keeps Majority AI Spend as DeepSeek Leads Tokens

A curious split showed up in this week's AI usage data that is worth staring at, because it undermines the easy narrative that open source is eating the frontier labs' lunch. TechCrunch reported that DeepSeek has surged into the lead on Vercel's AI gateway, now processing just over a third of the tokens passing through the platform, while Anthropic still accounts for more than half of the overall AI spend on that same infrastructure.

The pricing gap explains most of it. On OpenRouter, DeepSeek V4 Flash is the main winner on overall usage, processing 5.3 trillion tokens weekly, and Opus 4.8 handles just over 2 trillion, but Opus is priced at $1.37 per million tokens against V4 Flash's 6 cents, roughly 23x higher. Volume and revenue point in opposite directions.

Decagon CEO Jesse Zhang has a framing for that split that lands better than most enterprise AI takes I have read this year. In a post titled 'Everyone is wrong about open source AI in the enterprise,' he argues frontier and open source are not competitors but two phases of the same life cycle, with expensive frontier models being used to prove out use cases that then get passed along to cheaper open source alternatives as they mature. The soundbite is his: 'The frontier labs will keep owning discovery. Open source will increasingly own production.'

The honest caveats are worth naming. Vercel's AI gateway and OpenRouter are two specific slices of usage, not the whole market. The reporter notes Zhang doesn't give much data to support the point, and cautions that these rankings don't even capture Nvidia's Nemotron, which is poised to leap to the front of the pack by virtue of Nvidia's strong connections and the model's own extreme adaptability. Today's snapshot is temporary.

If Zhang is right, the strategic read is that Anthropic's revenue is not being drained by cheaper open source so much as it is being fed by it. Every new agent, coding assistant or workflow that eventually migrates to DeepSeek or GLM had to start on something like Opus first. The moat is discovery work, not production tokens, at least until frontier progress slows or a peer catches Anthropic on that end.