Nvidia used its GTC conference to take a position on one of the more persistent debates in AI: whether open or proprietary models will dominate. Its answer is that the question itself is wrong.
"Proprietary versus open is not a thing. It's proprietary and open," Nvidia founder and CEO Jensen Huang said during a panel at the event.
The case for coexistence
The argument, made across a back-to-back panel featuring leaders from LangChain, Thinking Machines Lab, Perplexity, Cursor, Reflection AI and Mistral, is that a diverse model ecosystem is not a transitional state on the way to consolidation. It is a permanent requirement.
The reasoning is practical. Large and small models serve different purposes. Generalist models handle breadth; specialist models handle depth. Open models accelerate research and lower the barrier to access. Proprietary systems deliver domain-specific value that open frameworks cannot always replicate. Long-running agent workflows, which may run for days and handle complex, multi-step tasks, need all of the above.
Nvidia's open-source footprint
Nvidia used the panel to highlight its own open-source contributions. The company says it is now the largest organisation on Hugging Face with nearly 4,000 team members, and that its Nemotron models have been downloaded more than 45 million times from the platform.
For a company whose business depends on selling chips to train and run both open and proprietary models, the "and not or" framing is coherent. Nvidia benefits from ecosystem growth regardless of which model type wins any given deployment.
Agents as coworkers
The panel's forward-looking thread centred on agents. Cursor CEO and co-founder Michael Truell put it plainly: "We're soon going to see agents really be coworkers that can take on tasks that take many hours or many days, and do incredibly complex workloads."
Related reading
- From Vera Rubins, Blackwell, Nemotron, NemoClaw, to Isaac and Cosmos - here's a whistlestop guide to NVIDIA's…
- Nvidia and energy partners unveil flexible AI data centres to ease grid pressure
- Meta and Arm to co-develop data centre chips built for AI workloads
That framing, agents as colleagues rather than tools, is becoming the default register for how the industry describes where this is heading. Whether enterprises are ready to manage that transition is a separate question, and one the panel did not linger on.
Nvidia encouraged attendees to watch GTC session highlights on YouTube and to begin building with its Nemotron open models.
The recap
- NVIDIA says open and proprietary models must coexist
- Nemotron models downloaded more than 45 million times
- Readers invited to watch GTC highlights and start building
