Nvidia's CEO says he backs both open and closed AI models, while a startup shows a tuned open model matching a top rival at a fraction of the price.
Nvidia CEO Jensen Huang says his company does not mind whether open or closed AI models win, as long as its chips power them. Speaking Tuesday at the Fireworks Forge event in San Francisco, he said Nvidia loves closed models. He also argued the world will not run on one general-purpose AI.
Closed models are the ones sold by companies like Anthropic and OpenAI, where the inner workings stay private. Huang said there is a solid place for them. But he said most of the world's value sits in specialized intelligence, meaning AI trained on a specific field of knowledge.
His advice was to take an open frontier model, add your own domain knowledge and tools, and test it properly. Do that, he said, and you can create essentially a superintelligence. In July he had already defended open weight models against pressure from the leaders of Anthropic and OpenAI.
Fireworks.ai, the event host, runs a platform where companies train and run models tuned to their needs. Its CEO, Lin Qiao, said cost is out of control and that businesses cannot scale without borrowing a lot of money. Fireworks also just received 25,000 Nvidia GPUs, a type of AI chip, for its expansion in Europe. Huang largely avoided the cost theme himself.
A later session gave a concrete example. Genspark built Gen-1 Slides, a tool that makes slide decks, on an open model called MiniMax M3. Genspark says it delivers results comparable to Anthropic's Opus 5 at one seventeenth of the cost.
Asked what could become commonplace next year, Huang named two things. The first is Open Shell, which he said will let every company deploy AI agents safely and confidently. Agents are AI programs that carry out tasks on their own. He admitted this might be a little self-serving.
The second is what he called hardware in the loop, which basically means robotic systems working with AI. Nvidia has clear interest and investment in robotics.
If your business pays for AI, the bill may depend on which model sits underneath. The material does not say how widely the Genspark savings apply, and the comparison comes from the company's own session. Still, it is fair to ask vendors which models they run on and whether a tuned open model could do the job for less.
If you pay for AI, ask vendors which model they run on, because tuned open models may do similar work for much less.
World models, AI that learns physics from video, are being built into home robots and driverless cars, and your own videos may help train them.
Read more: AI that learns how the physical world works is coming to robots and carsA new Alibaba-led test of real coding requests found the best AI agent fully finished just 26.4 percent of tasks, so human review still matters.
Read more: AI coding helpers fully finish only about 1 in 4 real tasksA University of Chicago law professor asked Claude to fix page numbers. It deleted his syllabus and a year of notes instead. Dropbox saved him, not the AI.
Read more: An AI assistant deleted a law professor's year of lecture notes
