Jesse Zhang, CEO of Decagon, recently unveiled a compelling new theory titled “Everyone is wrong about open source AI in the enterprise.” Published on Monday, his post delves into a significant paradox within the contemporary AI landscape: while more established AI implementations are migrating towards more efficient, lighter models – a trend observed even within his own organization – the aggregate expenditure on costly, state-of-the-art models remains largely undiminished.
This perspective offers a novel interpretation of the relationship between frontier and open-source models. Zhang posits that these are not competing entities, nor is the success of open-source solutions necessarily at the expense of frontier laboratories. Instead, he views them as distinct phases within a unified lifecycle, where expensive frontier models initially serve to validate use cases, which are then transitioned to more economical open-source alternatives as they mature.
Consequently, as mature applications shift to lighter models, a continuous emergence of new use cases ensures that the overall spending on frontier models experiences only minimal decline.
While Zhang’s post does not provide extensive data, corroborating evidence is readily available. Vercel’s AI gateway dashboard, for instance, reveals that DeepSeek has rapidly ascended to leadership in token volumes over the past week, now processing just over a third of all tokens traversing the company’s infrastructure. Similarly, Z.ai, the lab behind the widely recognized GLM-5.2 model, secured a notable fourth position during the same timeframe.
However, an examination of overall token spend paints a different picture, with Anthropic still commanding more than half of the total AI expenditure on the platform. Although Anthropic’s share has decreased slightly over the last month, largely due to its own price adjustments, the drop has not been substantial.
OpenRouter reports a comparable trend, albeit within a broader, slightly less enterprise-focused market segment. Deepseek V4Flash stands out in overall usage, processing an impressive 5.3 trillion tokens weekly. In contrast, the most prevalent frontier model, Opus 4.8, handles just over 2 trillion. While OpenRouter does not rank models by total spend, it indicates that the average token cost for Opus 4.8 is approximately 23 times higher than that of V4Flash ($1.37 per million tokens versus just 6 cents), suggesting Opus likely still captures the dominant share of spending.
These figures do not even account for Nvidia’s latest offering, Nemotron, which is anticipated to quickly rise to prominence, bolstered by Nvidia’s extensive industry connections and the model’s inherent adaptability.
While these statistics do not definitively prove Zhang’s complete thesis on AI lifecycles, they strongly suggest that frontier labs like Anthropic are not significantly suffering from the proliferation of open-source alternatives—at least not yet. One plausible explanation is the exponential growth of AI-addressable tasks, enabling top-tier models to maintain their strong market position by dominating early-stage deployments. As Zhang articulates, “The frontier labs will keep owning discovery. Open source will increasingly own production.” Another factor could be that many complex use cases, even as clients explore open-source options, remain too challenging to be entirely replaced by cheaper alternatives.
Regardless of the underlying reasons, this emergent two-tiered economy of AI models appears poised to become a relatively stable characteristic of the broader AI landscape.
As recently as last September, I contemplated the possibility that foundation labs might ultimately become commodity suppliers, akin to "selling coffee beans to Starbucks"—providing core inputs while the application layer garnered the primary benefits. Parts of that prediction have indeed materialized: specialized AI applications have shifted towards lighter models, and the economic viability of "GPT wrapper" startups has largely remained consistent.
Yet, we are also observing that, on a per-token basis, frontier providers have adeptly retained control over the most lucrative segment of the marketplace: the premium token price. This dynamic shows little indication of changing in the foreseeable future.
The Editorial Staff at AIChief is a team of professional content writers with extensive experience in AI and marketing. Founded in 2025, AIChief has quickly grown into the largest free AI resource hub in the industry.