Jakarta, INTI - Chinese AI startup Moonshot recently launched a new open-source artificial intelligence (AI) model called Kimi K3. Shortly after its release, the model attracted the attention of Microsoft, which is reportedly considering integrating the AI Model into its Azure cloud computing service.
According to a report by The Information, Microsoft is testing the effectiveness of Kimi K3 and could potentially use it to power its AI Copilot service. If realized, Kimi K3 has the potential to become one of the Chinese-made AI models Microsoft uses to support its AI products and reduce Microsoft’s reliance on OpenAI and Anthropic models.
Potential Cost Efficiency
Microsoft's interest in Kimi K3 is partly driven by the operational cost efficiencies it offers. Based on internal calculations, the use of Kimi K3 has the potential to save Microsoft up to USD 600 million in inference costs. Moonshot also claims that Kimi K3 has competitive performance compared to other AI models and is designed to be more efficient for running in data centers.
This efficiency is supported by a smaller KV cache size and optimized memory usage per token. Kimi K3 also distributes 896 experts across multiple GPUs to reduce computational requirements without sacrificing performance.
However, Kimi K3's inference costs are still relatively high for an open-source model. The average cost is USD 0.94 per Intelligence Index task. This cost is higher than GPT-5.6 Terra's USD 0.55 per task, but slightly lower than GPT-5.6 Sol's USD 1.04 in maximum reasoning mode.
Conclusion
Chinese AI startup Moonshot has launched the open-source Kimi K3 model, which Microsoft is reportedly evaluating for potentially integration into its Azure cloud service and AI Copilot. Microsoft's interest was sparked by Kimi K3's competitive capabilities and its inference cost efficiency, with potential savings of up to USD 600 million.
Read more: Indonesia Urged to Develop AI Regulation Through Dedicated AI Law