On Monday, Decagon CEO Jesse Zhang published a provocative latest theory, published under the title “Everyone is wrong about open source AI in the enterprise.” The post grapples with one of the most exciting contradictions of today’s AI economy: More mature AI deployments are switching to lighter models, he stated, even at his own company. But the overall spend on expensive latest models has barely budged.
It’s a new way to think about the relationship between frontier and open sources models. In Zhang’s telling, they aren’t competitors, and open source models’ success isn’t coming at the expense of frontier labs. Rather than, they’re two phases of the same life cycle, with expensive frontier models being used to prove out use cases that can be surpassed along to cheaper open source options as they mature.
As more mature use cases switch to lighter models, new use cases arising — and the overall spend on frontier models barely goes down.
Zhang doesn’t give much data to assist the point, however the data isn’t hard to find. Vercel’s AI gateway dashboard suggests that, in just the past week, DeepSeek has surged into the lead for token volumes, now processing just over a third of the tokens passing via the agency’s infrastructure. Z.Ai — the lab behind the popular GLM-5.2 model — jumped into a respectable fourth place over the same period.
But in you scroll down to overall token spend, you’ll see Anthropic still accounts for more than half of the general AI spend at the platform. Given that a lot of the current changes comes from Anthropic’s own rising prices, the share has dropped barely over the last month, but not extensively.
OpenRouter tells a similar story, capturing aa much larger (but slightly much less enterprise-y) segment of the market. DeepSeek V4 Flash is the primary winner on overall usage, processing 5.3 trillion tokens weekly. The most popular frontier model, Opus 4.8, manges simply over 2 trillion. OpenRouter doesn’t rank models by total spend, however it registers the average token price for Opus 4.8 as roughly 23x higher than V4 Flash ($1.37per million tokens, compared to just 6 cents), which would suggest Opus was still possibly capturing the lion’s share of spending.
Those figures don’t even capture the newest arrival, Nvidia’s Nemotron, which is poised to leap to the front of the pack via virtue of Nvidia’s sturdy connections and the model’s own excessive adaptability.
Those figures don’t fully show Zhang’s point about the AI life cycles, however they do display frontier labs like Anthropic aren’t suffering too much from the rise of open supply — at least not yet. One explanation is that the marketplace of AI-addressable tasks is developing so rapidly that the top models are able to hold their position just by dominating early-stage deployments. As Zhang puts it, “The frontier labs will hold proudly owning discovery. Open source will increasingly own manufacturing.” Another explanation is probably that, at the same time as customers move to open source, many use cases are so hard that they can’t be absolutely replaced with cheaper options.
Either way, this two-tiered economy of models may become a tremendously stable feature of the AI economy.
As currently as last September, I was writing about the possibility that foundation labs could end up to be selling coffee beans to Starbucks — that is, serving as commodity inputs at the same time as the application layer reaped the advantages. Some parts of that prediction came true: Vertical AI plays switched to lighter models, for one, and the economics of “GPT wrapper” startups have stayed mostly stable.
But we’re also when you consider that, token for token, frontier providers had been able to hold on to the most appropriate part of the market — the premium token price. And that doesn’t seem likely to change any time soon.
