Nvidia ramps up the AI arms race with Nemotron 3
Nvidia’s decision to release a new generation of open artificial intelligence models marks a notable shift for a company best known as the dominant supplier of chips powering the AI boom. With the launch of its Nemotron 3 family, the US group is moving more decisively into the software layer of the industry at a moment when its biggest customers are also becoming potential competitors.
The chipmaker said it was making available a suite of advanced models, training data and tools designed to help developers customise and deploy AI systems. Nvidia argues that open models remain central to experimentation and product development, even as leading US groups such as OpenAI, Google and Anthropic focus increasingly on proprietary systems. Industry data show that Chinese technology companies, which regularly release and update open models, now command a large share of global developer attention.
Nemotron 3 is offered in three versions, ranging from 30bn to 500bn parameters, placing the larger models among the most computationally demanding systems available outside closed commercial offerings. Nvidia also said it would disclose the data used to train the models and provide software to fine-tune them, a level of transparency that contrasts with the growing secrecy of many US rivals.
Jensen Huang, Nvidia’s chief executive, said “open innovation is the foundation of AI progress”, presenting the release as an effort to give developers “the transparency and efficiency they need to build agentic systems at scale”. Company executives argue that open models are easier to adapt to specific tasks, can be combined more flexibly and benefit from advanced training techniques that improve reasoning.
The move can also be read as a strategic hedge. Major AI groups are investing in their own chips, potentially reducing long-term reliance on Nvidia hardware. At the same time, geopolitical pressures are reshaping the market. China has pushed domestic companies to use home-grown silicon, raising the prospect that Chinese open models become optimised for non-Nvidia chips.
By strengthening its role in open AI software, Nvidia is seeking to anchor developers more firmly to its ecosystem. Whether this offsets the risks of customers turning into rivals will depend on how durable openness proves in an industry increasingly defined by scale, secrecy and strategic rivalry.
