Microsoft weighs Chinese AI model for Copilot amid US restrictions

Microsoft is evaluating a switch from OpenAI‑based models to the open‑source DeepSeek V4, a Chinese‑developed large‑language model, for its Copilot corporate assistant. The move is driven by rising costs of running AI agents on long‑running, repetitive tasks and by recent U.S. government actions that limited access to Anthropic’s advanced models. Charles Lamanna, Microsoft’s vice‑president for Copilot, said the company wants to keep data within Azure if it adopts DeepSeek, ensuring that customer information does not leave its servers.

The assessment reflects a broader industry trend where American firms turn to cheaper Chinese open‑source AI models as U.S. export controls and security concerns push up the price of Western offerings. NIST notes that DeepSeek lags U.S. models by about eight months, but its cost can be up to 95 % lower, making it attractive for large‑scale deployment.

At the same time, Microsoft’s Secure Future Initiative is expanding the use of AI agents within software development pipelines. These agents automatically review code, flag vulnerable dependencies, and detect credential leaks before deployment, aiming to maintain speed without compromising security.