AI Startups Shift to Open-Weight Models, Pressuring OpenAI and Anthropic Revenue Models

AI startups are increasingly adopting open-weight models instead of relying on OpenAI and Anthropic, as model costs surge and usage scales. The shift threatens the revenue growth and business models of the two leading closed-model providers as they prepare for public markets. Legal-tech startup Harvey, valued at $15.6 billion, saw gross margins collapse from about 50% in early 2026 to negative 50% by June after token usage rose 20-fold. Harvey launched its own model in August 2026, built on Moonshot AI's Kimi K3, and returned margins to positive. Healthcare startup Abridge is building custom clinical models on NVIDIA's open models; AI customer-service firm Decagon now handles 80% of queries through proprietary models. Fintech Ramp, which raised $750 million in June, and Rogo are evaluating self-trained models. Venture firms including Sequoia Capital and General Catalyst back the trend. Anthropic's annualized revenue topped $65 billion by end-July 2026, while OpenAI's exceeded $40 billion in July. Both companies have introduced cheaper models to address cost concerns.