In a nutshell: Meta has signed a multi-year agreement with Nvidia to purchase millions of Blackwell and Rubin GPUs, a deal reportedly worth tens of billions of dollars. The company has also committed to buying standalone Grace CPUs from the chipmaker for use across its AI data centers in the US, India, and other global locations.
According to separate press releases from Meta and Nvidia, the social media giant will use the chips to power its planned hyperscale data centers optimized for AI training and inference. The company will also deploy Nvidia's new Spectrum-X Ethernet switches as part of the Facebook Open Switching System (FBOSS) software stack used to control and manage network infrastructure.
As part of the agreement, Meta will adopt Nvidia's Confidential Computing platform to deliver AI features within WhatsApp Messenger while ensuring data security and confidentiality. According to Nvidia, the technology enables data to be processed on remote edge or public cloud servers within a protected enclave of the processor, preventing unauthorized access to sensitive information.
Meta said the secured platform will ensure privacy during computation, since end-to-end encryption only protects data while it is being transmitted to and from the company's servers. The new technology is also expected to help third-party software providers and AI companies safeguard their intellectual property, allowing users to collaborate securely without exposing sensitive data.
.@Meta is deploying AI at scale through deep codesign across the full NVIDIA platform.
– NVIDIA Newsroom (@nvidianewsroom) February 17, 2026
This expanded partnership will enable the large-scale deployment of NVIDIA CPUs and millions of NVIDIA Blackwell and Rubin GPUs.
Learn more: https://t.co/UTNEkXGN9V pic.twitter.com/jcIm63Loar
The two companies plan to expand the use of the Confidential Computing platform beyond WhatsApp to "emerging use cases" across Meta's portfolio. However, neither disclosed what those use cases might be or provided a timeline for large-scale deployment. Both companies said the technology will enable supported apps and services to deliver "privacy-enhanced AI" to consumers and enterprises.
Meta is also developing its own custom AI chips, known as the Meta Training and Inference Accelerator (MTIA). Designed specifically for inference workloads in the company's hyperscale data centers, the chips feature processor cores based on a customized version of the open-source RISC-V architecture and are manufactured by TSMC using its 5nm process.
A Financial Times report claims Meta is facing "technical challenges and rollout delays" with the MTIA chips, making it unclear when they will go live. However, the new chips are intended to supplement – not replace – Nvidia's accelerators, meaning the partnership is likely to continue even after MTIA is fully deployed.
Meta is reportedly planning to spend up to $135 billion this year on its AI-focused Superintelligence Labs division, as well as its core social media business. A significant portion of that budget was expected to go toward purchasing Google's Tensor chips, though it remains unclear whether those plans will proceed following the new deal with Nvidia.
Meta signs major Nvidia deal to power AI data centers with Blackwell and Rubin GPUs
