Anthropic, OpenAI tighten API access as $11.5B revenue fuels vertical push

Anthropic and OpenAI are tightening external API access for enterprise clients, restricting model capabilities and staging releases as they build competing vertical applications — a shift that threatens the $11.5 billion API business model.

Anthropic's catastrophic risk report warned that "if distilled models inherit risk-related capabilities from the original model but lack sufficient safeguards during deployment, they may pose downstream risks to the world even if the original model has strict anti-abuse mechanisms."

Anthropic reported second-quarter revenue of approximately $11.5 billion, up roughly 14 times from $787 million a year earlier, and reached adjusted operating profit, according to Bloomberg. The company's Fable model has already had cybersecurity-related capabilities deliberately weakened, and the Trump administration has pressured both labs to adopt staged-release strategies for new models, requiring individual client approval for access.

Developers including Canva, Harvey, and Cursor now face a supplier-as-competitor dilemma: their core products depend on APIs from labs that are building competing applications. Harvey and Cursor have begun training internal models to reduce dependency, and more developers are expected to follow, reshaping the AI value chain.

Safety and Distillation Drive the Squeeze

The API tightening is driven by two forces beyond commercial competition. First, safety: the Trump administration has pushed Anthropic and OpenAI to adopt staged releases for new models over concerns about cybercriminals and malicious actors, and has previewed a pre-release testing process for major AI companies. Even after public release, both labs may deliberately reduce model performance on specific tasks for security reasons — Anthropic's Fable model already has weakened cybersecurity capabilities.

Second, distillation: competitors can use outputs from commercially available models to train similar models at a fraction of the cost. Anthropic's risk report explicitly warns that distilled models could inherit dangerous capabilities without the original's safeguards. A former Anthropic researcher told The Information that fully preventing distillation is "basically impossible."

The commercial calculus is equally direct. Anthropic's API business generated roughly $11.5 billion in Q2 revenue with adjusted operating profit, but AI-driven vertical applications could offer even higher margins. Some investors have warned developers that Anthropic may reserve its most advanced technology for its own competitive applications rather than exposing it through APIs. Anthropic has already entered AI-driven drug discovery and other vertical domains. If the company — the dominant player in the AI API market — restricts access to limit competition, it could face significant antitrust scrutiny.

The stakes are visible in Anthropic's broader financial trajectory. Reuters reported the company's revenue run rate has topped $65 billion, and its IPO valuation hinges on a projected $190-200 billion revenue forecast for 2028.

Developers Build Their Own Models

The response from developers is already underway. Legal AI company Harvey and code editor Cursor have both begun training proprietary internal models to reduce reliance on Anthropic and OpenAI. The Information expects more developers to follow this path.

The logic is straightforward: when the supplier of core infrastructure can become a competitor or unilaterally change service terms, building independent technical capability becomes a strategic necessity. This trend could fragment the AI application layer, with more companies investing in proprietary model training rather than relying on third-party APIs.

For investors, the implications cut both ways. Anthropic's API revenue growth — 14 times year-over-year to $11.5 billion — demonstrates the strength of the frontier model market. But if the labs' vertical ambitions push developers toward self-training, the API growth trajectory could slow. Companies like Harvey and Cursor that invest in proprietary models face higher upfront costs but gain strategic independence. The market is watching whether Anthropic's $65 billion revenue run rate can hold if its largest API customers become competitors.

This article is for informational purposes only and does not constitute investment advice.