Moonshot AI Kimi K3 Model Cuts Microsoft Costs

Moonshot AI’s Kimi K3 model enables rapid chip design and massive savings for Microsoft as NVIDIA H200 GPUs reportedly head to China.

Moonshot AI has unveiled Kimi K3, a frontier model capable of designing a chip within 48 hours that achieves speeds exceeding 8,700 tokens per second. Leveraging lower compute costs to ascend benchmark rankings, the model has reportedly saved Microsoft $600 million. During the opening of a new Wistron plant in Dallas, NVIDIA CEO Jensen Huang emphasized the growing demand for AI computing power as Kimi K3 adoption increases. Huang noted the industry's need for both closed models from OpenAI and Anthropic and open models from Moonshot AI. In a significant shift, NVIDIA H200 AI GPUs, which were previously banned in China, are now reportedly being shipped to Chinese AI firms to support these advancing technologies.