Three moves from Microsoft that tell you exactly where enterprise AI economics are heading: Copilot Cowork goes usage-based. DeepSeek explored as a cheaper model. And Nadella says “we do a lot of tokenmaxxing internally — it’s almost addictive.”
Why This Matters
This is the tokenminimizing trend becoming product strategy. Microsoft faced a choice:
Charles Lamanna (EVP, Copilot) told Axios: users doing “hundreds of tasks a week” is great for productivity, but “the costs can go very high.” Translation: power users on flat-rate pricing were destroying margins.
The SaaS → AgaaS Transition In Real Time
This is the Mutation Map in action. Microsoft is executing Archetype #3 (Reprice) — shifting from seat-based SaaS to outcome/usage-based pricing without fully rebuilding the stack. The AI Supercycle predicts this buys time but doesn’t solve the structural problem — because the architecture underneath is still SaaS.
The DeepSeek Signal
Microsoft exploring DeepSeek as a cheaper Copilot model is extraordinary for three reasons:
Microsoft invested billions in OpenAI
And is now exploring a Chinese open-source alternative to reduce costs. The partnership isn’t exclusive when margins are at stake.
The Bottom Line
Microsoft killed per-seat AI pricing, explored a Chinese model to cut costs, and its CEO admitted to token addiction — all in the same breath. The SaaS pricing model can’t survive agentic workloads. The model layer is commoditizing so fast that Microsoft is willing to swap its biggest partner’s models for open-source alternatives. And the capex crossover chart explains why: when you’re spending more than you earn, everything is on the table.
Source: Axios









