Microsoft weighs DeepSeek for Copilot amid rising AI costs

At a glance:

• Microsoft is evaluating a self-hosted, fine-tuned version of DeepSeek V4 or another open-source model to power Copilot Cowork, aiming to reduce operational expenses.
• Copilot Cowork is transitioning to usage-based pricing, charging enterprises for actual compute consumption instead of flat fees.
• The move highlights tensions between cost efficiency and geopolitical concerns, as DeepSeek is Chinese amid U.S. restrictions on Chinese AI technologies.

Strategic cost reduction in agentic AI

Microsoft is exploring the integration of DeepSeek V4, a Chinese open-source AI model, into its enterprise Copilot suite to address escalating costs associated with agentic AI tools. The company confirmed to Axios that it is testing a self-hosted, fine-tuned version of DeepSeek V4 or alternative open-source models to power Copilot Cowork, an agentic assistant within Microsoft 365. This initiative is part of a broader effort to introduce a lower-cost tier within weeks, reflecting the financial strain of running AI agents that repeatedly invoke models during task execution.

Charles Lamanna, Microsoft’s executive vice-president for Copilot, agents and platform, highlighted the challenge: "We have users who do hundreds of tasks a week, which is great, they’re way productive, but the consequence is the costs can go very high." The shift mirrors Microsoft's prior decision to meter GitHub Copilot, which also faced rising costs due to model usage. Copilot Cowork currently relies on Anthropic and OpenAI models, both of which have increased prices and scaled back unlimited usage plans, pushing Microsoft to seek more economical alternatives.

Geopolitical and technical tensions

DeepSeek V4, released in April, has gained traction among developers for its cost-effectiveness and open-source accessibility. However, its Chinese origin complicates Microsoft's strategy amid escalating U.S.-China tensions over AI technologies. Washington has floated banning DeepSeek and pressured Chinese AI firms, while recently forcing Anthropic to restrict access to its top models for non-U.S. users—a dispute that led to crisis talks with the Commerce Department. Microsoft emphasized that any DeepSeek integration would be optional for customers and fully hosted on Azure, ensuring data residency and compliance under its cloud infrastructure. The company also noted that the model has been fine-tuned and augmented with safeguards to mitigate bias.

Multi-model diversification strategy

Microsoft's consideration of DeepSeek signals a strategic pivot away from over-reliance on single AI providers. Historically tied to OpenAI through an exclusivity agreement, Microsoft is now embracing a multi-model approach, blending engines from various labs under its own control. This diversification reduces dependency risks and provides flexibility to optimize costs and performance. While the evaluation remains preliminary, Microsoft's willingness to publicly acknowledge DeepSeek as a candidate underscores the urgency of managing AI expenses in an increasingly competitive and geopolitically fraught landscape.

Implications for enterprise AI adoption

The move could reshape how enterprises approach AI adoption, balancing cost efficiency with regulatory and ethical considerations. Usage-based pricing aligns with industry trends toward pay-per-use models, offering transparency but potentially increasing expenses for heavy users. Microsoft's emphasis on Azure hosting and compliance controls may ease some enterprise concerns about data security, though political scrutiny of Chinese technologies remains a hurdle. The outcome of this evaluation will likely influence how other tech giants navigate the trade-offs between cost, performance, and geopolitical risks in their AI strategies.

What to watch next

Microsoft has yet to finalize its model selection, with a decision expected alongside the launch of its lower-cost Copilot tier. The company's handling of DeepSeek integration could set a precedent for how Western firms engage with Chinese AI technologies amid regulatory pressures. Meanwhile, competitors like Anthropic and OpenAI may face renewed pressure to justify their pricing models as enterprises seek more affordable alternatives. The broader AI industry will also monitor how this shift impacts open-source adoption and the balance between innovation and cost management.