Meta signs multiyear deal for NVIDIA GPUs

NVIDIA announced a multiyear strategic partnership with Meta on Tuesday covering the large-scale deployment of NVIDIA CPUs, millions of Blackwell and Rubin GPUs, and Spectrum-X Ethernet networking across Meta's hyperscale data centers. ^[1]

Financial terms were not disclosed. CNBC reported the deal is "certainly in the tens of billions of dollars," citing chip analyst Ben Bajarin of Creative Strategies, who said a "good portion" of Meta's planned capex would go toward the NVIDIA build-out. ^[2] Meta announced in January that it expects to spend between $115 billion and $135 billion on capital expenditures in 2026, nearly double the $72 billion it spent in 2025. ^[3]

The deal, which spans on-premises infrastructure and cloud deployments, will support Meta's data centers optimized for AI training and inference, as well as what Meta described as "our core business" — a reference to its advertising operations, which generated $58 billion in revenue last quarter alone. ^[2][4]

The partnership includes what NVIDIA says is the first large-scale deployment of its Arm-based Grace CPUs as standalone chips — not paired with GPUs in a server, but running independently for inference and agentic workloads. ^[1] The companies are also collaborating on deploying NVIDIA's next-generation Vera CPUs, with potential large-scale rollout in 2027. ^[1]

"No one deploys AI at Meta's scale — integrating frontier research with industrial-scale infrastructure to power the world's largest personalization and recommendation systems for billions of users," said Jensen Huang, NVIDIA's CEO. ^[1]

Mark Zuckerberg framed the ambition more broadly: "We're excited to expand our partnership with NVIDIA to build leading-edge clusters using their Vera Rubin platform to deliver personal superintelligence to everyone in the world." ^[1]

What the deal covers

Meta will deploy GB300-based systems and build a unified architecture across its own data centers and NVIDIA Cloud Partner deployments. ^[1]

The partnership has four main components:

GPUs. Millions of current-generation Blackwell and next-generation Rubin GPUs for training and inference. CNBC noted that Blackwell GPUs have been on back-order for months, and Rubin chips recently entered production — meaning Meta has secured supply across both generations. ^[1][2]

CPUs. Large-scale standalone Grace CPU deployment for production workloads, with NVIDIA claiming significant performance-per-watt improvements. This is the first time Grace chips have been deployed at scale independently of GPUs. Vera CPU collaboration targets 2027. ^[1]

Networking. Spectrum-X Ethernet switches integrated into Meta's Facebook Open Switching System platform for low-latency AI-scale networking. ^[1]

Confidential Computing. NVIDIA Confidential Computing adopted for WhatsApp's private processing, enabling AI features while maintaining end-to-end data confidentiality. The companies plan to expand this beyond WhatsApp to other Meta products. ^[1]

Context

The partnership is not new (Meta has used NVIDIA GPUs for a while now) but the scope is significantly broader than previous arrangements. It now spans the full NVIDIA stack from silicon to networking to security, with joint engineering teams co-designing model optimization across Meta's core workloads.

The deal arrives as Meta pursues an aggressive AI infrastructure strategy. The company has committed to spending $600 billion in the U.S. by 2028 on data centers and related infrastructure, with 30 planned facilities including a 5-gigawatt site in Louisiana. ^[2] Meta is also developing a new frontier model code-named Avocado as a successor to its Llama family, led by former Scale AI CEO Alexandr Wang, who joined Meta after a $14.3 billion acquisition last year. ^[2][3]

For NVIDIA, the deal validates its product roadmap from Blackwell through Rubin and into the Vera generation, providing demand visibility well into 2027 and beyond. NVIDIA reports its fourth-quarter financial results on February 25. ^[1]