Nvidia launches new open models to run agents in the enterprise and on PCs
I find the PC angle more interesting than the data center one. Enterprises running open models on their own GPU clusters is already a familiar story, but Nvidia explicitly targeting local agents on workstations suggests they see a real market for AI that never leaves the machine. That could matter for legal, medical, or financial work where sending a prompt to an API is a nonstarter.
The models themselves are only half the story. Nvidia's real business is selling the chips that run them, and every new open model that needs a beefy GPU is a reason to buy another card. I'd want to know how these models compare to the ones from Meta or Mistral on actual agentic tasks, because open weights alone won't convince anyone if the tool-calling and reasoning are clunky.
Still, it's a sensible bet. The enterprise appetite for private AI is real, and Nvidia is positioned to profit no matter which model wins, as long as it runs on their hardware.