Nvidia's CEO chats about the future of AI: 'We're going to need three computers... one to create the AI… one to simulate the AI… and one to run the AI'

At this year's Siggraph event, Nvidia's Jen-Hsun Huang sat down with Wired for an hour-long chat about all things Nvidia, RTX, and AI. Among the varied topics touched upon, including a recognition that AI training and inference have huge energy demands, was Huang's assertion that more computers are going to be needed for AI systems in the future—specifically, three times more.

Siggraph is an annual conference normally about computer graphics and technology involved with interactivity (think AR and VR, that kind of thing) but it was only a matter of time before AI would become the main topic of discussion. To that end, Nvidia's CEO was interviewed by Wired's Lauren Goode for an hour-long streamed discussion that covered GPUs, RTX, and ray tracing, but mostly AI.

If you've been keeping up to date with Nvidia's push for generative AI to be everywhere, then there's nothing in the discussion that will really pique your interest. However, at one point, Huang mentioned how the world of AI is now moving away from its pioneering phase and moving toward the next one, which Nvidia's CEO called the «enterprise wave.»

After that comes the «physical wave», which, according to Huang, is «really, really quite extraordinary.» He clarified that statement by saying three computers will be required: one computer to create the AI, another to simulate and refine the AI, and finally a third computer to run the AI itself.

«It's a three computer problem. You know, a three body problem and it's so incredibly complicated and we created three computers to do that.»

Jen-Hsun is, of course, talking about Nvidia's raft of hardware and software packages, from its DGX H100 servers to create the AI, Jetsen embedded computers to simulate the AI, and then workstations and servers using Omniverse and RTX GPUs to run the AI.

Siggraph is one of my favourite tech events and I've been watching presentations and reading research papers presented at the conference for years. I have to say that it's a bit of