Nvidia (NVDA) CEO Jensen Huang has big plans for the AI data center industry. Through the company’s GTC conference, the chip giant’s executive and co-founder debuted Nvidia’s Spectrum-X and Quantum-X silicon photonics networking systems.
Networking switches connect multiple Nvidia graphics chips together to act as one giant system. Huang says Spectrum-X and Quantum-X photonics can join lots of and even hundreds of GPUs and can enable data center corporations to deploy as much as 1 million GPU clusters.
That’s a large jump from the present largest AI supercomputer, xAI’s Colossus, which is made up of 100,000 Nvidia Hopper GPUs. And based on Huang, those million cluster data centers might be connected to other million-dollar data centers nearby to form massive data center facilities.
“Over the subsequent several years, we will be constructing giant AI factories. Not normal AI factories … ones you see from space,” Huang explained.
The CEO spent a superb deal of his time on stage during his GTC keynote explaining how Nvidia’s latest advancements, starting from Spectrum-X and Quantum-X photonics to its Blackwell Ultra and Vera Rubin superchips, will help the corporate’s customers, including cloud service providers like Microsoft and Google, operate ever-more-powerful AI systems to drive higher revenue opportunities.
“AI factories are directly related to revenue, and if the throughput just isn’t good, your revenue is hurt. When you haven’t got enough capability, your revenue is hurt. When you’re not producing something of great value … your revenue is hurt,” Huang said.
Read more: How does Nvidia make cash?
The CEO equated AI factories to auto plants, explaining how if automakers don’t have the fitting factories and products on offer, their revenues will suffer.
”This might apply to a automobile company. In case your capability just isn’t high enough, your revenue is hurt. In case your utilization just isn’t good, your revenue is hurt. in case you construct a automobile, no one desires to buy it, your revenue is hurt,” he explained.
“It is a factory. Except it is a factory of tokens. It is a recent world. An industrial revolution.”
That industrial revolution won’t just happen inside of information centers though. Along with its chips for large server systems, Nvidia also showed off a handful of smaller AI-centered computers, including its mini desktop-sized DGX Spark and full-size desktop-sized DGX Station.
Full of Nvidia’s Grace Blackwell super chips, the computers are designed for AI developers and researchers. And when paired with the corporate’s other AI platforms, they make up a recent approach to enterprise IT that revolves around AI.