Microsoft is shifting Copilot Cowork to usage-based billing and considering a cheaper, self-hosted version of DeepSeek V4 to manage rising token costs from AI agents.

What happened: Microsoft is moving Copilot Cowork away from flat-rate pricing to usage-based billing, and is weighing a fine-tuned version of DeepSeek V4 as a lower-cost model option. The product currently uses Anthropic's Claude technology, which consumes tokens quickly because it relies on agentic reasoning. A final decision on DeepSeek is expected in the coming weeks.

Why it matters: Flat-rate pricing has become unsustainable because some users perform hundreds of tasks per week, driving costs up quickly—a problem Microsoft already addressed by switching GitHub Copilot to usage-based billing. Any optional DeepSeek model would run fully on Azure and be customized with safeguards, keeping customer data within Microsoft's cloud infrastructure, though using a Chinese AI model may draw criticism in the US.

What to watch: CEO Satya Nadella has signaled support for this direction in a recent blog post arguing that companies should be able to pick and tune AI models for specific use cases and costs. The shift reflects his stated vision of AI as a consumption business built around intense users and intense usage.