A provocative new theory from Decagon CEO Jesse Zhang is reshaping how the industry understands the relationship between cutting-edge AI models and their open-source counterparts. Writing on Monday, Zhang argued that open-source models are not stealing market share from frontier labs such as Anthropic. Instead, they serve two distinct phases of a single lifecycle: expensive frontier models prove out new use cases, which are then handed off to cheaper open-source alternatives once mature.
Data from Vercel's AI gateway supports this view. In the past week, DeepSeek surged to process just over a third of all tokens on the platform, while GLM-5.2 from Z.ai jumped to fourth place. Yet when measured by total spend, Anthropic still accounts for more than half of all AI expenditure on Vercel, despite recent price increases. OpenRouter tells a similar story: DeepSeek V4 Flash handles 5.3 trillion tokens weekly, versus 2 trillion for Anthropic's Opus 4.8, but Opus costs roughly 23 times more per token, meaning it still captures the lion's share of spending.
The pattern suggests that as the AI-addressable market expands rapidly, frontier labs maintain their position by dominating early-stage deployments. 'The frontier labs will keep owning discovery. Open source will increasingly own production,' Zhang said. New models like Nvidia's Nemotron are poised to enter the fray, leveraging Nvidia's deep industry connections to potentially leapfrog existing players, but the overall spend on premium models has barely budged.
For UK businesses, this two-tier economy has significant implications. Companies exploring AI for the first time may find frontier models essential for prototyping, but can later migrate to cheaper open-source versions for scaled production. The UK Information Commissioner's Office (ICO) has yet to issue specific guidance on this bifurcated market, but the EU AI Act's risk-based framework could shape how UK firms adopt each tier. Frontier models, often more opaque, may face stricter regulatory scrutiny, while open-source models offer greater transparency but require careful governance around data privacy and bias.
Experts caution that the stability of this model may depend on how quickly new use cases emerge. 'If the pipeline of novel applications dries up, frontier labs could lose their premium pricing power,' said one AI analyst who asked not to be named. 'But for now, the market is growing fast enough to sustain both tiers.' For UK consumers, the trend could mean more affordable AI-powered services in the long run, as mature applications shift to cheaper models, while still benefiting from state-of-the-art innovation in areas like healthcare diagnostics and financial modelling.