In the latest mega-deal for AI compute power, the social media giant and the AI chip leader have signed a new agreement under which Meta will use even more Nvidia chips in its data centers.
This agreement marks the expansion of the companies existing partnershipWhat the vendors describe as “multigenerational and multigenerational,” spanning across on-premises, cloud, and AI infrastructure.
Meta said in a statement that the deal is designed to “support the creation of data centers optimized for AI training and inference” as well as its core business.
Although neither party provided financial details, the sums involved are likely to be huge – possibly up to tens of billions of dollars, given Meta CEO Mark Zuckerberg’s commitment last month. Spend up to $135 billion on AI this year alone.
The deal covers Meta’s purchase of Nvidia Grace central processing units (CPUs), as well as Blackwell and Next Generation. vera rubin gpu. According to the Nvidia release, the vendors also said they are collaborating on Vera CPUs, the successor to the Grace CPUs, with “the potential for large-scale deployment” in 2027.
Typically, GPUs and CPUs work together to improve efficiency, with the GPU handling the processing needed to train AI models, and the CPU focusing more on system operations.
But the deal constitutes Nvidia’s first Grace-only deployment, pointing to what could be a sign of things to come, as AI vendors rely more on standalone CPU inference better suited to handle AI workloads.
The deal doesn’t just cover chips. Meta has also adopted Nvidia’s Spectrum-X Ethernet platform into its AI infrastructure for improved performance and efficiency and the Nvidia Confidential Computing security system to enable secure use of AI features on WhatsApp.
“No one deploys AI at the scale of meta – integrating frontier research with industrial-scale infrastructure to power the world’s largest personalization and recommendation systems for billions of users,” Nvidia CEO Jensen Huang said in the release.
“We’re excited to expand our partnership with Nvidia to build leading-edge clusters using their Vera Rubin platform to provide personal superintelligence to everyone in the world,” Zuckerberg said in the release.
While the deal is clearly attractive to Nvidia and marks another step in Meta’s AI ambitions, it also raises questions about Meta’s ongoing efforts to make its own chips and become less dependent on Nvidia.
Meta has invested heavily in this area, but the Financial Times reported that its strategy has faced technical problems and delays in the rollout.
