Nvidia Corp.’s graphics processing units sit at the heart of almost every artificial intelligence system in use today, and they’ve become especially critical for the most powerful large language models that power generative AI technologies such as ChatGPT. The company has emerged from almost nowhere just a few years ago to dominate the AI industry, and that dominant status means its annual GTC event has become one of the most prominent AI events on the annual tech calendar.

This year’s edition was perhaps the most hotly anticipated of all, as evidenced by the more than 16,000 people who attended in person. Nvidia underlined its status as the leading AI chip company when it debuted its most powerful GPU architecture yet, called Blackwell. The Blackwell GPU architecture embodies what theCUBE Research chief analyst Dave Vellante (pictured, left) describes as a “monolithic approach” that plays into every aspect of the AI market.

“Jensen Huang is basically saying that the same infrastructure is going to do both training and inference, and there’s definitely a market for that,” Vellante said. “There are real benefits to monolithic, and he spoke about that, how every part of the chip talks to each other. Nvidia had to create a new communications system to do that. It’s taking the high end of the market and there’s really nobody that is going to touch them.”

Vellante spoke with theCUBE host and research analyst John Furrier (right) and ZK Research analyst Zeus Kerravala (center) at the end of day two of Nvidia’s GPU Technology Conference, in an exclusive segment for theCUBE, SiliconANGLE Media’s roving news desk. They discussed Nvidia’s plans to corner the entire AI market, and also dissected Huang’s vision of generative AI as a new era of generated information and content.

Nvidia’s AI data center

Nvidia became the top dog in AI because of its willingness to evolve, going from a simple chipmaker to the creator of an entire back-end infrastructure for AI. Kerravala explained that the company’s chips could have been held back by any number of limitations, such as networking bandwidth and inefficient memory transfers.

“But these things didn’t hold them back because they solved them,” the analyst said. “We talked about NVLink yesterday and I think it’s the most underappreciated innovation, because it solved the problem of inter-chip communications.”

That has enabled Nvidia to build what is essentially a data center for AI, Kerravala said, starting with a chip, then a pair of chips, before going to a motherboard, a server, a system and then a cluster of systems. “It’s really a data center, and they built all of that together,” he said.

Generative AI vision

Nvidia has ambitious plans to leverage this AI data center infrastructure it has built, with Furrier noting Huang’s comment that the company intends to “manufacture AI and LLMs at scale,” and build the operating system for the future of AI.

According to Furrier, this can be translated as Nvidia making a “land grab for the token market,” extending its control to data, especially vector embeddings and similar capabilities that will sit at the heart of this comprehensive AI operating system. This links back to Huang’s new industrial revolution pitch, where he outlined a broader vision for a new generative AI era.

“The old world is pre-recorded,” Furrier explained, saying it’s about generation rather than retrieval. “You’ve got the news, videos, everything is prerecorded and that means you’ve got to fetch it. Generative AI is the seeds of innovation, where data is seeds.”

A new data center supercycle

The analysts agreed that one of the major takeaways from this year’s GTC is that Nvidia isn’t about to run out of gas anytime soon. Many have wondered if Nvidia’s rapidly-growing stock is overvalued, but Vellante said it’s clear that we’re in the early days of a “data center supercycle” that is going to morph the way mobile and cloud computing changed our lives.

“I think its going to be completely radical, and within eight years we’re going to have functional robots,” Vellante said. “We’re going to be talking to robots and they’re going to be doing all of these menial tasks around our homes, and it’s going to change the way people live and work. We’re in the early innings of a 10-year data center supercycle.”

Here’s the complete video analysis, part of SiliconANGLE’s and theCUBE’s ongoing coverage of the GPU Technology Conference:

Photo: SiliconANGLE