Now that the dust has settled from its GPU Technology Conference (GTC), a new, fuller picture is emerging amidst all the many announcements. Analysts agreed that Nvidia is not just a graphics chip provider anymore, it is a full stack solution provider, of which GPUs are just one of many parts.
Nvidia is repositioning itself as an infrastructure provider and a lot of the biggest announcements fit into this overall picture, analysts said. Not only individual hardware elements like the latest GPUs, networking technology advancements like silicon photonics and even efforts in storage, but also why they laid out their roadmap so far in advance. CEO Jensen Huang announced two new generations of GPU architecture stretching into 2028. Previously he’s only given one year of advanced generational preview.
“They’re trying to help build an overall ecosystem around this infrastructure and want to make sure it’s partners have a better idea of what’s coming so that they can plan accordingly,” said Bob O’Donnell, president and chief analyst with TECHnalysis Research.
Nvidia is evolving from being just the parts provider to a full stack offering, says Anshel Sag, principal analyst with Moor Insights and Strategy. “Nvidia also gave a lot of visibility of its roadmap in a way that it previously hasn’t with Feynman [AI chips] all the way out to 2028. This is only because Nvidia needs to communicate clearly with its ecosystem and enable them to plan that far out,” he said.
His boss, Moor Insights & Strategy CEO Patrick Moorhead, added: “Nvidia has shifted from a one-off product company to a full AI infrastructure, providing compute, GPU compute, networking and now data platform and with Dynamo, an AI scheduler,” Moorhead said.
“Add that to all the horizontal and vertical developer tools, frameworks and models and I think you could also surmise that it is foundational to all tech. While there are questions on how long the growth will continue, I’m convinced we are looking at long term growth driven by enterprise AI and edge AI,” he said.
“Nvidia has evolved from a gaming chip company to an AI supercomputer company with a deep and wide software stack that covers over a dozen vertical apps, super hi-speed electro-optical inter-processor communications, and a killer processor that uses the latest HBM4 high-speed high-density memory. The company also announced GPUs would now power the latest storage systems. All in all, Nvidia has scaled out, and up and is a total computer supplier,” said Jon Peddie, president of Jon Peddie Associates.
To that end, Nvidia is focused on the bottlenecks in the AI process that it can address, and they are many: Networking with the silicon photonics switches, memory with KV cache, simulation data with Cosmos, humanoid robots with Isaac GR00T, desktop AI workstations with DGX Spark and DGX Station, and Dynamo, Jim McGregor, principal analyst with Tirias Research, said.
“Everything is getting more dense, more power efficient, and higher performance,” he said.
And as a result, everyone wants to work with Nvidia. There was a much bigger focus on partnerships as GTC demonstrated: Cisco for networking, Cisco and T-Mobile for 5G/6G, Dell, HP and Lenovo for their DGX Spark and Studio hardware platforms, Dell, HPE, Lenovo, Supermicro and others for the Blackwell Ultra server systems and more, said O’Donnell.
Source:: Network World