On March 12, financial filings submitted by NVIDIA to the U.S. Securities and Exchange Commission (SEC) disclosed that the company is set to allocate $26 billion (equivalent to approximately RMB 178.8 billion) over the forthcoming five years to the research and development of open-source AI large models. The magnitude of this investment significantly surpasses the $3 billion expended by OpenAI for training GPT-4. This strategic step signifies NVIDIA's official transformation from a mere 'chip manufacturer' to a 'leading full-stack AI laboratory.' The investment will encompass model development, computing infrastructure, talent recruitment, and ecosystem establishment. The initial batch of self-developed models is anticipated to be unveiled between late 2026 and early 2027. From a technical perspective, NVIDIA has embraced an 'open weights' methodology, allowing enterprises to freely download and fine-tune key model parameters, albeit with training data and code potentially remaining partially undisclosed. Presently, NVIDIA has discreetly finished the pre-training of a model boasting 550 billion parameters and introduced a new-generation open-source large language model, Nemotron 3 Super, which supports an ultra-long context window of 1 million Tokens. NVIDIA's strategy is designed to establish the technical blueprint for AI models from scratch, advocating its own hardware architecture and software stack as industry benchmarks while stimulating demand for computing power through open-source models. Financial analysts project that if NVIDIA can capture a 10% stake in the foundational model market while reinforcing its hardware supremacy, it could potentially yield an extra $50 billion in yearly revenue within three years.