Teaching Everyone to Fish for Tokens
Nvidia plans to invest $26 billion in developing open-source models, aiming to enable global construction of “token machines” through open data and training code to prevent smart monopolies. This model is analogous to the Linux operating system, providing complete training recipes for enterprises to modify and use. Nvidia’s Nemotron series has already released some data and code to encourage community contributions. Although building high-performance models is capital-intensive and highly competitive, the current withdrawal of companies like Databricks and 01.ai is considered unusual, and the industry still relies on Nvidia’s financing. There are two possible paths in the future: one is for the open-source ecosystem to become profitable through performance or strong market demand, making training benefits comparable to those of APIs from Anthropic and OpenAI; the other is that if it cannot become profitable, open-source models will shift towards an independent development path focused on efficiency, modifiability, and specialization.