Frontier AI Models Hold Revenue Lead Despite Open Source Surge
New data shows expensive models from Anthropic and OpenAI still capture majority of enterprise spending even as token volumes shift to cheaper alternatives.
The explosive growth of open source AI models hasn't dented revenue at frontier labs like Anthropic, according to new infrastructure data that reveals a surprising split in the enterprise AI market.
Decagon CEO Jesse Zhang recently published analysis arguing that frontier and open source models occupy different phases of the same deployment lifecycle rather than competing directly. His thesis: companies use expensive state-of-the-art models to prove new use cases, then migrate mature workflows to cheaper alternatives—but the overall spend on premium models remains steady as new experimental use cases continuously emerge.
The token volume versus spending gap
Infrastructure metrics support this division. Vercel's AI gateway dashboard shows DeepSeek now processes over one-third of all tokens flowing through its platform, with China's GLM-5.2 model claiming fourth place in volume. Yet Anthropic still accounts for more than half of total AI spending on the same infrastructure.
OpenRouter data tells a parallel story across a broader market segment. DeepSeek V4 Flash processes 5.3 trillion tokens weekly compared to just over 2 trillion for Anthropic's Opus 4.8. But with Opus costing roughly 23 times more per token—$1.37 per million tokens versus 6 cents—the frontier model likely captures the majority of revenue despite lower volume.
Why it matters
This emerging two-tiered structure challenges assumptions that open source models would commoditize foundation model providers. Instead, frontier labs appear to have secured the high-margin discovery phase of AI deployment while open source handles production workloads. The pattern suggests foundation model economics may prove more durable than earlier predictions of a race to zero, particularly if the addressable market for AI tasks continues expanding faster than mature use cases migrate to cheaper alternatives.
Production versus discovery economics
Zhang frames the division as "frontier labs will keep owning discovery, open source will increasingly own production." The rapid market expansion may allow premium providers to maintain position by continuously capturing early-stage deployments, even as yesterday's cutting-edge applications move to budget alternatives.
Another factor: many enterprise use cases may prove too complex for complete migration to lighter models, creating sustained demand for frontier capabilities regardless of open source progress.
Nvidia's newly released Nemotron model—positioned for strong enterprise adoption through the company's hardware relationships—suggests this premium tier will remain competitive and potentially expand.
The infrastructure data indicates that token-for-token, frontier providers have retained control of the most valuable market segment: the premium price point. That positioning appears stable for now, contradicting earlier predictions that foundation labs would become mere commodity inputs while application layers captured value.
These details were first reported by TechCrunch.
This is an original analysis by the Omega editorial team. Source reporting: AI Watch.
Want systems like this working for your business?
Book a Call
