Microsoft reportedly weighs swapping GPT and Claude for China's Kimi K3 in Copilot

According to reporting, Microsoft is evaluating whether to swap OpenAI's GPT and Anthropic's Claude for Moonshot AI's Chinese Kimi K3 model inside Copilot, a change that could save the company as much as $600 million. The rationale is straightforward cost arbitrage: Chinese open-weight and low-cost models have closed enough of the capability gap that running them for high-volume Copilot workloads could dramatically cut inference bills. The report follows earlier signals that Microsoft was interested in DeepSeek's V4 for similar reasons.

The strategic tension is geopolitical. Adopting a Chinese frontier model in a flagship U.S. enterprise product could invite scrutiny from the Trump administration, which is reportedly weighing bans on cutting-edge Chinese models. It also sits awkwardly beside Microsoft's simultaneous multibillion-dollar commitment to European sovereign-AI partner Mistral — suggesting Microsoft is hedging across multiple non-OpenAI suppliers at once.

Competitively, the report intensifies questions about the durability of Microsoft's OpenAI relationship and about model commoditization: if Copilot can run on whichever model is cheapest, frontier labs lose pricing power in their largest distribution channel. Kimi K3's popularity — Moonshot paused new subscriptions as demand swamped capacity — bolsters its credibility. Skeptics note this is an evaluation, not a committed switch, and that data-governance, trust, and political risk may block adoption regardless of cost. Watch for official confirmation and whether any Chinese model actually ships in a shipping Copilot tier.