If regulators on both sides of the Atlantic let him close the $40 billion acquisition of Arm Holdings announced Sunday, Nvidia co-founder and CEO Jensen Huang will have the three essential components he needs to build his data center empire.
Nvidia is already the dominant supplier of accelerators used in supercomputers for machine learning and traditional high-performance computing workloads. Last year it gobbled up Mellanox, the leading supplier of networking technology for high-performance computing. With Arm under its wing, Nvidia would also be one of the top challengers to Intel in the data center CPU market, and there would be little it wouldn’t be able to design in terms of computing infrastructure solutions.
“The amount of computer science horsepower … inside this company would be quite extraordinary,” Huang said on a call with reporters and analysts Sunday, following the announcement that SoftBank had agreed to sell most of its stake in Arm to Santa Clara, California-based Nvidia. It would be the biggest acquisition in the chip industry’s history.
For Arm, which designs processors and licenses its intellectual property to others, data centers are a relatively small business today. The company’s chips power most of the world’s smartphones and tablets, and that’s where it makes most of its money. But it’s been pouring a lot of resources into expanding in the data center market, and an acquisition by Nvidia – whose data center revenue recently surpassed its gaming revenue – would undoubtedly accelerate that expansion.
“I’m super excited to focus a lot of energy around turning Arm into a world-class data center CPU,” Huang said.
The industry met the announcement of the deal with a lot of skepticism about the lengths to which Nvidia was prepared to go to ensure Arm continues serving its customers without influence from Nvidia that would disadvantage its competitors. Some have suggested that the deal would push more vendors toward RISC-V, the open source chip architecture that’s been growing in popularity in recent years and gaining momentum in the data center market.
Nvidia executives have maintained that they are committed to Arm’s current business model, that it would operate as an independent subsidiary, and that its headquarters would remain in Cambridge, UK.
What makes Huang particularly confident in Nvidia’s ability to boost Arm’s presence in the data center market are his company’s broad capabilities across hardware and software. A “broadly successful” data center platform starts with the CPU, he said, but it also requires “system software, all of the various engines and libraries and application frameworks on top…
“We’re rather unique in being able to take a CPU (and one that has been really, really refined over time) and turn it into a computing platform from end to end: the system, the software, all of the algorithms, and all of the frameworks.”
On the Sunday call Huang highlighted Arm’s recent high-profile wins in the data center: AWS’s first and second-generation Arm-based Graviton processors and Fujitsu’s Fugaku supercomputer in Kobe, Japan, which in June became the first Arm CPU-powered system to reach the number-one spot on the Top500 list of the world’s fastest supercomputers.
But the deal isn’t only about Nvidia making Arm part of its product portfolio and integrating it with its own hardware. Huang envisions leveraging Arm’s intellectual property licensing infrastructure as a sales channel for Nvidia IP. That means we could eventually see Nvidia GPU architecture and designs, or Mellanox networking technology, available for other vendors to license and make their own products with, paying Nvidia royalties from the parts sold.
Nvidia may also ship Nvidia-branded hardware based on Arm, or license architectures that combine Arm and Nvidia IP. “Any of those options are available,” Huang said.
The chief executive expects proliferation of AI to drive massive demand for computing infrastructure everywhere. That includes IoT devices (where Arm is also a dominant player) that connect to edge data centers and to core cloud data centers.
“It is impossible for one company to build all of those solutions,” he said. “But it is possible for us to come with some architectures that every company in the world could benefit from building amazing solutions with.”