Key Points:
- Meta will use millions of Nvidia chips to power its AI data centers.
- The deal includes the first large-scale use of Nvidia’s standalone Grace CPUs.
- Meta plans to deploy next-generation Vera Rubin systems starting in 2027.
- Nvidia and Meta stocks rose, while competitor AMD’s stock dropped 4%.
Meta and Nvidia announced a massive expansion of their partnership on Tuesday, solidifying a deal that will see millions of Nvidia chips power Meta’s artificial intelligence ambitions. While the companies kept the specific financial numbers private, analysts estimate the agreement is worth tens of billions of dollars. This move is a central part of Meta’s plan to spend up to $135 billion on AI infrastructure in 2026 alone.
The most significant technical shift in this agreement involves CPUs. For the first time, Meta will deploy Nvidia’s Grace central processing units as standalone chips in its data centers.
Previously, companies mostly paired these CPUs with graphics processors (GPUs) in the same server. By using them independently, Meta aims to better handle specific AI workloads, such as “inference”—the process where the AI actually answers user questions or performs tasks.
Mark Zuckerberg, Meta’s CEO, framed the deal as a necessary step toward his vision of “personal superintelligence” for everyone. To achieve this, Meta is securing a steady supply of Nvidia’s current Blackwell GPUs, which have been hard to find due to high demand. The company also locked in access to the upcoming Vera Rubin systems, which are scheduled for deployment in 2027.
This partnership strikes a blow to competitors. Following the news, shares of Advanced Micro Devices (AMD) sank about 4%, as investors worry the chipmaker is losing ground in the data center war.
Meanwhile, stocks for both Meta and Nvidia climbed in extended trading. The deal also includes Nvidia’s networking technology, which helps link thousands of chips together to work as a single supercomputer.
Meta is currently building two enormous facilities to house this hardware: the “Prometheus” site in Ohio and the “Hyperion” site in Louisiana. These projects are part of a broader $600 billion investment plan for U.S. infrastructure through 2028. While Meta still designs some of its own chips and buys from other suppliers, this agreement cements Nvidia’s role as the primary engine behind Facebook and Instagram’s future AI features.
