Qualcomm to make data center processors that connect to Nvidia chips

Audio By Carbonatix

Enlarge

By Stephen Nellis and Max A. Cherney

SAN FRANCISCO/TAIPEI (Reuters) -Qualcomm on Monday said it will make custom data center central processing units, or CPUs, that use technology from Nvidia to connect to Nvidia’s artificial intelligence chips.

Nvidia’s chips are dominant in the AI market but always paired with CPUs, a market traditionally dominated by Intel and Advanced Micro Devices. Nvidia has jumped into the CPU market itself, designing a chip using technology from Arm Holdings to develop its own “Grace” CPU.

On Monday, Qualcomm said it would return to the data center CPU market. In the 2010s, Qualcomm began developing an Arm-based CPU that it tested with Meta Platforms, but curtailed those efforts amid cost cuts and legal challenges.

But after acquiring a team of ex-Apple chip designers in 2021, Qualcomm has quietly revived those efforts, again holding discussions with Meta about a data center CPU. Qualcomm last week confirmed that it has a letter of understanding with Saudi Arabian AI firm Humain to develop a custom data center CPU.

On Monday, Qualcomm said that its future chips would use technology from Nvidia that will help them communicate quickly with Nvidia’s graphics processors (GPUs), which are the mainstay of its AI chip portfolio.

“With the ability to connect our custom processors to Nvidia’s rack-scale architecture, we’re advancing a shared vision of high-performance energy-efficient computing to the data center,” Cristiano Amon said on Monday.

(Reporting by Stephen Nellis in San Francisco and Max A. Cherney in Taipei; Editing by Sam Holmes)