Nvidia Allocates $26B to AI, Launches Nemotron 3 Super
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Nvidia announced a $26 billion, five‑year investment in open‑weight AI models, disclosed in a 2025 filing and confirmed to WIRED.
- Nvidia released Nemotron 3 Super, a 128‑billion‑parameter open‑weight model that it says outperforms OpenAI’s GPT‑OSS on the Artificial Intelligence Index (score 37 vs 33) and on the PinchBench benchmark.
- Nvidia said most of the $26 billion will fund open‑source model development, with a smaller portion earmarked for AI algorithm infrastructure, autonomous‑vehicle software, and other research.
- Bryan Catanzaro, Nvidia’s VP of applied deep learning research, emphasized that the company is “taking open model development much more seriously” and highlighted recent technical innovations that improve reasoning, long‑context handling, and reinforcement‑learning responsiveness.
- Nvidia noted that Chinese open‑weight models (e.g., DeepSeek, Alibaba’s Qwen) are gaining traction in the West, and that its US‑made open models could serve as a strategic alternative in the US‑China AI competition.
- Nvidia previously pretrained a 550‑billion‑parameter model and has released specialized models for robotics, climate modelling, and protein folding, indicating a broader ecosystem focus beyond the flagship Nemotron 3 Super.
- Nathan Lambert (AI2) praised the investment as a significant signal for openness and called for additional US government funding for open models, echoing comments from Andy Konwinski of the Laude Institute.
Why it matters: The $26 billion spend gives Nvidia a direct role in shaping the open‑model ecosystem, strengthening its chip sales by tying models to its hardware, while providing researchers and startups a US‑based alternative to Chinese open models that could otherwise erode Nvidia’s market share.
