The Subsidized Era of AI Ends: Frontier Labs Double Prices Ahead of IPOs

Summary

Both OpenAI and Anthropic have doubled their flagship model prices as they prepare for public market debuts, signaling the end of a three-year period of subsidized AI compute. OpenAI's GPT-5.5 and Anthropic's Claude Fable 5 launched at 2x the pricing of their predecessors, with additional premiums for priority processing and long-context windows. This repricing extends across the entire AI ecosystem: GitHub Copilot abandoned its flat premium tier for usage-based billing in June, Cursor switched to consumption-based pricing in June 2025, and Replit moved to effort-based pricing—all citing unsustainable economics of absorbing frontier model costs under fixed subscriptions. The shift forces enterprises to distinguish between AI spending that produces value and AI spending that generates waste, making infrastructure investment in token efficiency tracking essential for the next phase of AI adoption.

• Enterprises must now invest in infrastructure to track token efficiency and distinguish high-value AI projects from those that simply consume compute without ROI

Editorial Opinion

This transition marks a critical inflection point for enterprise AI adoption. The subsidy era enabled rapid experimentation but also masked inefficiencies—poorly optimized agents and bloated tool definitions cost the same as lean, well-engineered systems. The repricing will impose discipline on the market, but it also creates a competitive moat: companies with mature MLOps infrastructure to monitor and optimize token usage will thrive, while those that treated AI as a commodity will face painful budget surprises. This is painful short-term but healthy long-term for the industry's maturation.