Open-source AI models are handling a growing share of usage across major AI platforms, but available market data suggests that frontier model providers such as Anthropic continue to capture a large portion of overall spending. On Vercel’s AI gateway dashboard, DeepSeek recently moved into the lead by token volume, processing more than a third of tokens on the platform, while Anthropic still accounted for more than half of total AI spend there. OpenRouter showed a similar pattern, with DeepSeek V4Flash leading weekly usage and Anthropic’s Opus 4.8 trailing in token volume but commanding a far higher average price per million tokens.
That pattern has fueled an argument from Decagon CEO Jesse Zhang that open-source and frontier models are not simply displacing one another. Zhang argues they represent different stages of enterprise adoption, with expensive frontier models used to test and define new use cases before companies shift mature workloads to cheaper alternatives. By that logic, open-source models increasingly dominate production workloads while frontier labs continue to lead earlier-stage experimentation and discovery. Market data cited alongside that argument does not fully prove the theory, but it does support the view that growth in open-source adoption has not yet significantly reduced spending on leading proprietary models.
At the same time, critics are raising broader doubts about whether the economics of frontier AI are sustainable. Skeptics cited in recent debate point to heavy losses at OpenAI, capital-intensive infrastructure commitments, and warnings from executives such as Palantir CEO Alex Karp, who has criticized token-based AI business models and argued that enterprises may be paying heavily for limited value while exposing proprietary advantages. Concerns have also extended to intellectual property and governance issues, including tensions around Anthropic’s product moves affecting Figma, as well as financing and customer-payment risks tied to major data-center buildouts involving Oracle and the Stargate project.
Supporters of the frontier model business argue that revenue growth and pricing power still matter. Anthropic has been presented by backers as evidence that premium models can remain commercially valuable even as lower-cost competitors spread, while Nvidia’s results have been cited as proof that parts of the broader AI stack are generating substantial cash. The central question now is whether the model layer can remain a durable, premium business or whether it will face deeper price pressure as open-source alternatives improve. For now, the available evidence points to a two-tier AI economy: open-source models gaining scale in usage, and frontier labs still holding much of the premium end of the market.
