Why the rise of open source AI isn’t hurting Anthropic … yet
TL;DR
Decagon CEO Jesse Zhang argues that enterprises use expensive frontier models to discover new AI use cases, then move mature workloads to cheaper open source models. Vercel data partly supports the pattern: DeepSeek recently handled more than a third of token volume there, while Anthropic still captured more than half of AI spend on the platform, according to TechCrunch.
Nauti's Take
The uncomfortable truth for open source is that usage is not the same as revenue. Processing 5.3 trillion cheap tokens wins visibility, but not necessarily the richest part of the customer base.
Anthropic should worry when companies start new use cases directly on open models instead of reaching for Claude-class systems first. Until then, open source looks more like the cost reducer in production than the revenue killer at the frontier.
Briefingshow
For AI teams, token volume alone is a weak proxy for market power. Open source models can absorb production workloads while Anthropic keeps earning from the expensive discovery and premium layer. The argument is still lightly evidenced, but it matches a familiar enterprise pattern: buy quality first, optimize cost later.
Sources