Decagon CEO Jesse Zhang has put forward a theory that challenges the assumption that cheaper open source AI models are eroding the business of frontier labs such as Anthropic. In a blog post titled "Everyone is wrong about open source AI in the enterprise," he argues that expensive frontier models and cheaper open source ones are not rivals but two stages of a single life cycle: costly cutting-edge models prove out new use cases, which are then handed down to lighter open source models as they mature, while fresh use cases keep frontier spending buoyant. This matters because it reframes a widely held fear that leading AI firms risk becoming low-margin commodity suppliers.

Available data lends the idea some support. On Vercel's AI gateway, DeepSeek has surged to lead token volumes, yet Anthropic still commands more than half of overall spend. OpenRouter shows a similar split: DeepSeek V4 Flash processes 5.3 trillion tokens weekly against just over 2 trillion for Anthropic's Opus 4.8, but Opus costs roughly 23 times more per token, so it likely still captures the bulk of spending. The author notes that frontier providers have retained the most valuable slice of the market — the premium token price — and that this two-tiered economy may prove a stable, lasting feature, though Nvidia's new Nemotron model could shake up the rankings.