Microsoft Could Turn to DeepSeek V4 to Cut Copilot Cowork Costs

Microsoft is reportedly considering using a self-hosted version of DeepSeek’s V4 model for Copilot Cowork, according to a report from Axios. The move comes just weeks after Microsoft expanded Copilot Cowork availability to all Microsoft 365 users and could signal a broader shift in how the company manages the growing costs of enterprise AI.

According to Axios, Microsoft is exploring alternatives as the cost of using advanced AI models from OpenAI and Anthropic continues to rise. The company has also reportedly restricted the use of Anthropic’s Claude Fable 5 model because of concerns related to data retention policies.

As enterprise AI adoption accelerates, token consumption has become a major financial challenge. More advanced agentic workflows, coding assistants, and automation tools require significantly more processing than traditional chatbot interactions, increasing operational expenses for AI providers and customers alike.

AI Token Costs Are Becoming a Growing Concern

The report notes that several companies have already encountered unexpectedly high AI bills as employees and automated systems increase their usage of large language models.

At the same time, OpenAI and Anthropic are reportedly introducing higher enterprise pricing while placing greater limits on token consumption. These changes have pushed some organizations to reevaluate which AI models they use and how they deploy them internally.

Copilot Cowork combines Microsoft’s enterprise Copilot capabilities with advanced AI models designed for workplace automation, collaboration, and agentic workflows. As these workloads become more demanding, maintaining flat-rate subscription plans becomes increasingly difficult.

Microsoft is reportedly moving Copilot Cowork toward a metered pricing structure based on total token usage, aligning costs more closely with actual consumption.

Why DeepSeek V4 Could Appeal to Microsoft

A self-hosted deployment of DeepSeek V4 could offer Microsoft a way to lower inference costs while retaining control over infrastructure, deployment, and data handling.

The approach would not necessarily eliminate political concerns, however. DeepSeek is widely viewed as a China-linked AI company, and any significant Microsoft partnership involving its technology would likely attract attention from lawmakers and regulators in Washington amid ongoing scrutiny of Chinese AI platforms.

Enterprise AI Economics Are Changing

The reported discussions highlight a broader trend across the AI industry. As model capabilities improve, operating costs continue to rise, forcing both providers and customers to rethink pricing structures.

For Microsoft, finding lower-cost alternatives could become increasingly important as enterprise customers scale AI-powered automation across large organizations.

If Microsoft ultimately adopts DeepSeek V4 for Copilot Cowork, it would demonstrate that even one of OpenAI’s largest partners is actively exploring cheaper AI options as enterprise demand grows and token costs continue to climb.

The development also follows other recent AI-related controversies. Anthropic recently restricted access to its Mythos-class Fable 5 model for non-US citizens following concerns about advanced cyber capabilities, while Microsoft continues to face scrutiny over the billions of dollars it has invested in AI infrastructure and services, including a recent report that the company faces a shareholder lawsuit over AI spending and Azure growth.

Via Wccftech