On Monday, Decagon CEO Jesse Zhang released a provocative new theory titled “Everyone Was Wrong About Open Source AI in the Enterprise.” This post grapples with one of the most interesting paradoxes of today’s AI economy. He says his company is also transitioning to a more mature AI deployment toward a lighter model. But overall spending on expensive, newer models has barely budged.
This is a new way to think about the relationship between frontier and open source models. According to Zhang, they are not competitors, and the success of the open source model does not come at the expense of Frontier Labs. Instead, they are two stages of the same life cycle, with an expensive frontier model used to demonstrate use cases that can be transferred to cheaper open source alternatives as they mature.
As more mature use cases transition to lighter models, new use cases will continue to emerge and overall spending on frontier models will show little reduction.
Zhang doesn’t provide much data to support his point, but the data isn’t difficult to find. According to Vercel’s AI Gateway dashboard, in the past week alone, DeepSeek has surged to the lead in token volume and is currently handling more than a third of the tokens passing through the company’s infrastructure. Z.ai, the lab that developed the popular GLM-5.2 model, jumped to fourth place during the same period.
However, if you scroll down to total token spending, you’ll see that Anthropic still accounts for over half of the platform’s overall AI spending. Given that most of the recent changes have come from Anthropic’s own price increases, the company has seen its share decline slightly over the past month, but not by much.
OpenRouter tells a similar story, capturing a much larger (but slightly less enterprise) segment of the market. DeepSeek V4 Flash is the main winner in terms of overall usage, processing 5.3 trillion tokens per week. The most popular Frontier model, the Opus 4.8, handles just over 2 trillion. OpenRouter doesn’t rank models by total spend, but the average token cost for Opus 4.8 is about 23 times higher than V4 Flash ($1.37 per million tokens compared to just 6 cents). This means that Opus still accounts for the lion’s share of spending.
Those numbers don’t even capture Nvidia’s Nemotron, its latest offering that’s poised to leap to the forefront thanks to Nvidia’s powerful connectivity and the extreme adaptability of the model itself.
While these numbers don’t fully substantiate Zhang’s claims about the AI life cycle, they do show that leading labs like Anthropic aren’t suffering too much from the rise of open source. At least not yet. One explanation is that the market for tasks that AI can handle is growing so rapidly that the best models can maintain their position simply by dominating early-stage deployments. As Zhang puts it, “Pioneering labs will continue to own discovery. Open source will increasingly own production.” Another explanation could be that even if customers switch to open source, many use cases are too difficult to fully replace with cheaper alternatives.
Either way, the two-stage economy of these models could become a relatively stable feature of the AI economy.
Last September, I wrote about the possibility of the Foundation Research Institute selling coffee beans to Starbucks. In other words, it acts as a commodity input while the application layer reaps the benefits. Some of those predictions have come true. For example, vertical AI plays have shifted to lighter models, and the economics of “GPT wrapper” startups have largely remained stable.
However, we are also seeing that frontier providers offering tokens on exchange are able to maintain premium token prices, which is the most desirable part of the market. And that doesn’t seem likely to change anytime soon.
We may receive a small commission if you purchase something through links included in this article. This does not affect our editorial independence.
