Nvidia is happy to take your money whether you're using open or closed models

Nvidia CEO Jensen Huang argues that the company can prosper whether AI is built from proprietary frontier models or open-weight alternatives, since both depend on computing infrastructure. He defended the role of closed models while emphasizing that specialized systems may deliver much of AI’s practical value. In his view, organizations can adapt an open model with their own domain knowledge, tools, and evaluations to create highly capable systems for specific tasks. That outlook matches Fireworks.ai’s business of helping companies customize and run models; the company has received 25,000 Nvidia GPUs to expand in Europe.

The discussion also highlighted the pressure to rein in AI costs. Fireworks CEO Lin Qiao warned that heavy token spending makes business growth difficult, while Genspark described a slide-generation model based on MiniMax M3 that aims to match Anthropic’s Opus 5 at about one-seventeenth the price. Huang identified agent systems as this year’s major breakthrough and predicted that Nvidia’s Open Shell could help companies deploy agents safely. He also forecast that “hardware in the loop” - AI systems interacting with physical machinery, especially robots - would become a major industrial development.