Microsoft Eyes Chinese DeepSeek Model to Cut Copilot Costs

Microsoft is exploring the integration of a self-hosted, fine-tuned version of DeepSeek-V4, an open-source model from the Chinese developer DeepSeek, to power a lower-cost tier of its Copilot Cowork agentic AI. The company aims to launch this cheaper version within several weeks to manage soaring computing costs associated with high-volume enterprise users who execute hundreds of tasks weekly.

This strategic pivot follows pricing increases and a shift from flat-rate to usage-based models by current providers OpenAI and Anthropic. CEO Satya Nadella has recently argued against the dominance of a few major AI players, suggesting that such market concentration is harmful to the economy. To mitigate security and political concerns regarding the model's Chinese origin, Microsoft intends to make the option optional for customers and host the model entirely on Azure under its own security and compliance controls.

The move occurs as the United States government, under the Trump administration, increases scrutiny of Chinese AI firms and pushes for restrictions on foreign nationals' access to advanced American models. By transitioning to a multi-model strategy, Microsoft seeks to reduce dependency on single suppliers while adjusting Copilot Cowork to a usage-based pricing structure to ensure long-term operational sustainability.