Why the rise of open source AI isn’t hurting Anthropic yet

On Monday, Decagon CEO Jesse Zhang unveiled a provocative new theory under the headline “Everyone is wrong about open source AI in the enterprise.” The post explores one of today’s most compelling contradictions in the AI economy: while more mature AI deployments are shifting toward lighter models—even within his own company—overall spending on expensive state-of-the-art models remains largely unchanged.

This perspective reframes the relationship between frontier and open source models. According to Zhang, they are not competitors, and the success of open source models does not come at the expense of frontier labs. Instead, they represent two phases of the same lifecycle: expensive frontier models are used to validate use cases, which are then handed off to cheaper open source alternatives as they mature.

As established use cases migrate to lighter models, new use cases continuously emerge, keeping overall spending on frontier models steady.

Although Zhang provides limited data to support his argument, the evidence is readily available. Vercel’s AI gateway dashboard reveals that over the past week, DeepSeek surged to the top for token volumes, processing just over a third of all tokens through its infrastructure. Meanwhile, Z.ai—the lab behind the popular GLM-5.2 model—climbed to a respectable fourth place during the same period.

However, when examining overall token spend, Anthropic still accounts for more than half of the total AI expenditure on the platform. Although this share has dipped slightly over the past month due to Anthropic’s own price increases, the decline is not significant.

Image Credits:Vercel dashboard / data export

OpenRouter presents a similar narrative, capturing a larger (though slightly less enterprise-focused) segment of the market. DeepSeek V4 Flash leads in overall usage, processing 5.3 trillion tokens weekly. The most popular frontier model, Opus 4.8, handles just over 2 trillion tokens. While OpenRouter does not rank models by total spend, it shows that the average token cost for Opus 4.8 is roughly 23 times higher than V4 Flash ($1.37 per million tokens versus just 6 cents), suggesting that Opus still captures the majority of spending.

These figures do not even include Nvidia’s newest entry, Nemotron, which is poised to leap to the forefront thanks to Nvidia’s strong industry connections and the model’s exceptional adaptability.

While these numbers do not fully prove Zhang’s point about AI lifecycles, they do indicate that frontier labs like Anthropic are not yet suffering significantly from the rise of open source—at least not for now. One explanation is that the market for AI-addressable tasks is expanding so rapidly that top models can maintain their dominance by controlling early-stage deployments. As Zhang states, “The frontier labs will keep owning discovery. Open source will increasingly own production.” Another possibility is that, even as clients adopt open-source models, many use cases are too complex to be entirely replaced by cheaper alternatives.

Regardless, this two-tiered model economy may become a stable feature of the AI landscape.

As recently as last September, I wrote about the possibility that foundation labs would end up selling coffee beans to Starbucks—essentially serving as commodity inputs while the application layer reaped the benefits. Some aspects of this prediction have materialized: vertical AI companies have shifted to lighter models, and the economics of “GPT wrapper” startups have remained largely stable.

Yet, we are also seeing that, token for token, frontier providers have retained the most valuable segment of the marketplace: premium token pricing. This trend does not appear likely to change anytime soon.