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

Open source AI models are not displacing frontier labs like Anthropic in enterprise spend, despite surging token volumes. According to a new analysis by Decagon CEO Jesse Zhang, the two categories serve different phases of the same lifecycle: frontier models prove out new use cases, then lighter open source models take over production as those use cases mature.

What happened

Jesse Zhang published a theory arguing that open source AI and frontier models are not competitors but two phases of the same lifecycle. Expensive frontier models are used to validate new use cases, and as those use cases mature, they shift to cheaper open source alternatives. The overall spend on frontier models barely declines because new use cases keep emerging.

Data from Vercel's AI gateway dashboard supports this view. DeepSeek now processes just over a third of tokens on the platform, while Anthropic still accounts for more than half of total AI spend. On OpenRouter, DeepSeek V4 Flash processes 5.3 trillion tokens weekly, while the frontier model Opus 4.8 handles just over 2 trillion. However, Opus 4.8 costs roughly 23x more per token ($1.37 per million tokens vs. $0.06 for DeepSeek), meaning frontier models still capture the lion's share of spending.

Why AI builders should care

For teams building AI products, this two-tier economy has direct implications. If you are prototyping a new feature or agent, frontier models like Anthropic's Opus 4.8 give you the best chance of proving the concept works. Once the use case is stable and you need to scale, switching to an open source model like DeepSeek V4 Flash can cut token costs by over 20x.

This pattern means you should plan for model migration from the start. Build your architecture with a model abstraction layer so you can swap out the inference provider as your use case matures. The frontier labs will keep owning discovery, while open source will increasingly own production.

Practical implications

The data also shows that the market is growing fast enough to sustain both tiers. New use cases keep appearing, so frontier labs maintain their premium pricing even as open source models eat into production workloads. Nvidia's Nemotron is poised to leap to the front of the pack, which could shift the balance further.

For builders, the key metric to watch is not just token volume but total spend. Even if open source models handle more tokens, frontier models may still generate more revenue per token. That dynamic could persist as long as new, hard problems keep emerging.

Caveats

The evidence comes from specific platforms (Vercel and OpenRouter) and may not represent the entire enterprise AI market. Pricing and model leadership can change quickly, especially with new entrants like Nemotron. The two-phase lifecycle is a useful framework, but it is not a guaranteed prediction. Builders should monitor their own cost and performance data rather than relying solely on industry averages.

FAQs

Why isn't open source AI hurting Anthropic yet?

Industry data shows frontier labs like Anthropic still command a large share of AI spend, even as open source models grow in token volume. The two-phase lifecycle explains this: frontier models are used for discovery and new use cases, while open source models handle mature production workloads. As new use cases keep emerging, frontier spend remains high.

What are frontier models and how do they differ from open source models?

Frontier models are state-of-the-art, expensive models used to prove out new use cases. Open source models are lighter, cheaper alternatives that take over production once a use case is validated. They serve different phases of the same lifecycle rather than competing directly.

Which open source AI models lead in production vs. discovery?

On OpenRouter, DeepSeek V4 Flash leads in token volume with 5.3 trillion weekly, while the frontier model Opus 4.8 handles about 2 trillion. However, Opus 4.8 costs roughly 23x more per token, so it still captures the majority of spending. Nvidia's Nemotron is expected to shift the balance soon.

What is the two-phase lifecycle of enterprise AI?

The two-phase lifecycle describes how frontier models are used for discovery (proving new use cases) and open source models are used for production (scaling validated use cases). This creates a stable two-tier economy where both categories thrive without directly competing.