Jensen Huang argues that open-weight AI models are expanding Nvidia's dominance as the only platform capable of running all frontier models
Jensen Huang wants to make one thing very clear: if you’re running an open AI model, you’re almost certainly running it on Nvidia hardware. And if you want to run every frontier model, open or closed, there’s only one platform that does it all.
The open model playbook
On July 24, 2026, Huang made his inaugural post on X, sharing an open letter co-signed by Nvidia and roughly 25 other organizations. The letter called for fostering both open and closed AI model development to drive competition and innovation, while pushing back against regulatory approaches that might stifle progress.
Nvidia has been putting its own silicon where its mouth is. The company released the Nemotron language model family and Cosmos for physical AI under permissive licenses, giving developers access to both weights and training resources. Then on August 11, 2026, Nvidia unveiled Nemotron 3.5 Lightning as another open-source offering, further expanding its portfolio of freely available models.
Why open models matter for Nvidia’s bottom line
During earnings calls and keynotes, Huang has pointed to open-source AI adoption as a key driver of enterprise GPU server deployments. Companies in SaaS, robotics, and industrial applications are increasingly building custom AI solutions rather than simply consuming API calls from closed model providers. These enterprises want models they can modify, deploy on their own infrastructure, and run at the edge, not just in the cloud.
Huang has been careful to frame this as a balance rather than a binary. He maintains that both open and closed models should coexist, and that open models actually benefit companies building closed ones by advancing the overall state of AI research. He maintains that the right balance between open and closed models enhances AI capabilities and amplifies Nvidia’s total addressable market for GPU sales.
