Teaching Everyone Token Fishing

Open-source AI models are similar to the Linux operating system, but they are categorized into models that include complete training recipes and data, and open-weight models that only contain inference code. Nvidia is releasing training code and data through its Nemotron models, encouraging enterprises to build their own models. This is a strategy that prevents the monopoly of intelligence while simultaneously driving multiple companies to purchase Nvidia's inference chips.

Currently, Nvidia is investing $26 billion in this open-source model business. Model training is capital-intensive, but if open-source recipes can maintain competitiveness, Nvidia's chip demand and profits will far exceed the cost of building models. While companies like Databricks and 01.ai are stepping back from the training game, there are still no clear signs that the open-source ecosystem is collapsing due to capital intensity.