The AI Cost Crisis.

This week MSFT quietly canceled its Claude Code licenses. Token-based billing made the cost untenable — even for a company with effectively infinite resources. 

In addition, Uber’s CTO sent an internal memo warning they had already burned through their entire 2026 AI budget in just four months. Enterprise AI prices have jumped 20-37% and GitHub (owned by Microsoft) is dropping flat-rate plans for usage-based billing across the board.

The latest benchmark makes the problem obvious:

• Opus 4.7 Max (Anthropic) leads at 64.8% score… but costs $11.02 per task.

• Composor 2.5 sits right behind it at 63.2% for only $0.55 — roughly 20× cheaper.

• Gemini 3.5 Flash delivers 49.8% at $1.94.

This is why Cursor + cheaper high-performance models (including xAI options) are winning. You get 95%+ of the coding performance at a small fraction of the cost. The AI subsidy era is ending in real time. Anthropic’s premium pricing strategy is hitting a wall — and this could genuinely hurt them more than most people expect. Either enterprises slash usage (slowing the revenue the labs need) or the labs absorb bigger losses. The numbers stop working either way.