NVIDIA's Open Source Model Strategy Officially Emerges, Delving into the Model Layer from Selling Shovels to Mining Gold

BlockBeats News, August 23rd. Analyst qinbafrank posted an article interpreting the latest deal between NVIDIA and Poolside: NVIDIA paid $6 billion to obtain a non-exclusive license for Poolside's "Model Factory" technology, which is the core software system for Poolside's Laguna series of open-source encoding models. At the same time, NVIDIA extended invitations to over 100 Poolside engineers to join NVIDIA's focus on the Nemotron open-source weight model project. Since its launch in 2023, this project has been developing a larger and more advanced version, with rumored parameter scales reaching the trillions, along with approximately $1 billion in equity investment. The analyst believes this is a landmark move for NVIDIA, transitioning from selling shovels to mining gold. Huang Renxun is taking this opportunity to delve into the model layer, building an "AI Factory" full-stack capability deeply integrated with the NVIDIA ecosystem.

The analyst outlined four key points:

Firstly, NVIDIA aims to rebuild one of the world's strongest open-source weight models, benchmarking against Chinese open-source models, while directly challenging leading U.S. closed-source companies like OpenAI and Anthropic. Open-source weight models have lower operating costs and are more customizable.

Secondly, extending from selling GPU hardware to software and the model layer, NVIDIA is strengthening its control over the entire upstream and downstream processes.

Thirdly, while enhancing its moat, it also intensifies competition with customers: the improvement in open-source model capabilities will drive GPU demand, but Nemotron's direct competition with large customers' closed-source models may strain relationships.

Fourthly, the trend can be summarized as major model manufacturers developing in-house chips, and chip manufacturers developing in-house models, further intensifying competition.

The analyst pointed out that Huang Renxun publicly endorsed open source at the end of July, which was actually paving the way for NVIDIA's in-house open-source models. This deal is a key step for NVIDIA to strengthen its position in the open-source competition between China and the U.S. If the new version of Nemotron matches in performance, it will further benefit cloud providers and software companies with access to customer workflow data.