Microsoft is considering deploying the Chinese AI model to its Copilot venture, and the reason is money.
The company told Axios that it is exploring its own, fine-tuned version of DeepSeek V4 or another open-source model as a cheaper option to power Copilot Cowork, an agent assistant in the Microsoft 365 suite. It expects to introduce a cheaper model within weeks.
At the same time, Microsoft is moving Copilot Cowork to usage-based pricing, charging companies for the computing they actually burn, rather than a flat fee.
Why can’t even Microsoft account eat
The change agent is a window into the economics of AI. Tools like Copilot Cowork, Anthropic’s Claude Code, and OpenAI’s Codex call the model repeatedly while working on a task, which is powerful and expensive.
“We have users doing hundreds of tasks a week, which is great, they’re very productive, but as a result the costs can be very high,” said Charles Lamanna, executive vice president of Copilot, agents and platform at Microsoft.
Copilot Cowork is currently working on models from Anthropic and OpenAI, both of which have raised prices and you’ve backed out of all-you-can-eat plans. Microsoft has already scaled GitHub Copilot for the same reason. The cheaper open source engine below is the obvious next branch.
The cheapest option is China
This is where the calculation gets awkward. Released in April, DeepSeek V4 is open source, popular with developers and cheaper to run, which is why it’s on Microsoft’s shortlist.
He’s also Chinese, and the timing could hardly be worse politically. Washington Bans DeepSeek, Threatens Chinese AI Firms, and Gets Fair Forced Anthropic to cut top models for non-US usersa growing controversy Crisis negotiations with the Department of Commerce.
Microsoft clearly knows what that looks like. Any DeepSeek variant would be optional for customers and fully hosted in Azure, keeping data in Microsoft’s cloud under security, compliance and data residency controls, and it says it has fine-tuned the model and added safeguards, including changes intended to reduce bias.
Hedging against own suppliers
The bigger picture is that Microsoft no longer wants to depend on any one lab. With OpenAI, it takes a multi-model approach that mixes and matches engines under its own roof, freed from tight, often strained exclusivity.
For now, it’s an assessment, not a decision, and Microsoft says it will confirm its choice when the cheaper tier ships. But the fact that it’s willing to even list DeepSeek as a candidate says a lot about how difficult it is for agents to spend in this climate.
