Meta’s 5-gigawatt Hyperion knowledge heart underneath development in Richland Parish, Louisiana, Jan. 9, 2026.
Courtesy of Meta
Meta will use tens of millions of Nvidia chips in its synthetic intelligence knowledge facilities, together with Nvidia’s new standalone CPUs and next-generation Vera Rubin programs, in a sweeping new deal introduced Tuesday.
Meta CEO Mark Zuckerberg mentioned in an announcement that the expanded partnership continues his firm’s push “to ship private superintelligence to everybody on this planet,” a imaginative and prescient he introduced in July.
Monetary phrases of the deal weren’t supplied.
Shares of Meta and Nvidia climbed throughout prolonged buying and selling on Tuesday. Superior Micro Units inventory sank about 4% on the information.
In January, Meta introduced plans to spend as much as $135 billion on AI in 2026. “The deal is actually within the tens of billions of {dollars},” mentioned chip analyst Ben Bajarin of Artistic Methods. “We do anticipate a great portion of Meta’s capex to go towards this Nvidia build-out.”
The partnership is nothing new, as Meta has been utilizing Nvidia graphics processing models for at the very least a decade, however the deal marks a considerably broader expertise partnership between the 2 Silicon Valley-based giants.
Standalone CPUs are the largest new factor within the deal, with Meta turning into the primary to deploy Nvidia’s Grace central processing models as standalone chips in its knowledge facilities, versus integrated alongside GPUs in a server. Nvidia mentioned it is the primary large-scale deployment of Grace CPUs on their very own.
“They’re actually designed to run these inference workloads, run these agentic workloads, as a companion to a Grace Blackwell/Vera Rubin rack,” Bajarin mentioned. “Meta doing this at scale is affirmation of the soup-to-nuts technique that Nvidia’s placing throughout each units of infrastructure: CPU and GPU.”
The following-generation Vera CPUs are deliberate to be deployed by Meta in 2027.
The multiyear deal is a part of Meta’s total dedication to spend $600 billion within the U.S. by 2028 on knowledge facilities and the infrastructure the services require.
Meta has plans for 30 knowledge facilities, 26 of which can be based mostly within the U.S. Its two largest AI knowledge facilities are underneath development now: the Prometheus 1-gigawatt website in New Albany, Ohio, and the 5-gigawatt Hyperion website in Richland Parish, Louisiana.
Additionally included within the deal is Nvidia’s networking expertise, Spectrum-X Ethernet switches, that are used to hyperlink GPUs collectively inside large-scale AI knowledge facilities. Meta can even use Nvidia’s safety capabilities as a part of AI options on WhatsApp.
The social media large would not solely depend on the highest chipmaker. In November, Nvidia inventory fell 4% on stories that Meta was contemplating utilizing Google‘s tensor processing models in its knowledge facilities in 2027.
Meta additionally develops in-house silicon processors and makes use of chips from AMD, which received a notable take care of OpenAI in October as AI giants search a second supply to Nvidia amid constrained provide.
Nvidia’s present Blackwell GPUs have been on back-order for months, and the next-generation Rubin GPUs just lately went into manufacturing. With the deal, Meta has secured a wholesome provide of each.
Engineering groups from Nvidia and Meta will work collectively “in deep codesign to optimize and speed up state-of-the-art AI fashions” for the social media large.
Meta has been creating a brand new frontier mannequin dubbed Avocado as a successor to its Llama AI expertise. The newest model launched final spring did not excite builders, CNBC beforehand reported.
Meta’s inventory has been on a curler coaster in latest months, and its AI technique particularly has puzzled Wall Road.
The inventory noticed its worst day in three years in October after the corporate introduced formidable AI spending, then popped 10% in January after reporting stronger-than-expected gross sales steering.
CNBC’s Jonathan Vanian and Kristina Partsinevelos contributed to this report.
