On January 31th, NVIDIA announced that it had released a preview version of the "NVIDIA NIM Microservices" using DeepSeek R1 671b on its website, publicly stating that DeepSeek-R1's reasoning capabilities were "the most advanced." NVIDIA stated that the DeepSeek R1 NIM microservice can process up to 3872 tokens per second on a single NVIDIA HGX H200 system; NVIDIA advertised that developers can now test and experiment with the API .

The NVIDIA AI Enterprise software platform will soon incorporate relevant APIs and be officially launched in the form of downloadable "NIM microservices". NVIDIA's use of DeepSeek to allow NIM microservices to transform hardware advantages into AI service capabilities is an important move for NVIDIA in the enterprise-level AI market. It not only accelerates the commercialization of high-performance models, but may also reshape the AI ​​supply chain pattern, promoting the industry to shift from competition at the infrastructure level to integration at the application ecosystem level. Developers and enterprises should pay attention to the evolution of NIM and evaluate whether it can become a key lever to reduce the complexity of AI deployment.

The release of DeepSeek R1 671b announced by NVIDIA as a preview version of NIM microservices is of great significance. It shows great potential in terms of technical performance, development convenience, business prospects, etc., and has a positive layout in data privacy and security as well as future development.