Meta Platforms Inc. has finalized a significant agreement with Nvidia Corp. to acquire millions of next-generation Vera Rubin graphics processing units and Grace central processing units. This multi-billion dollar deal is designed to accelerate Meta's ambitious artificial intelligence development.
Mark Zuckerberg, Meta's CEO, stated the expanded partnership is critical for achieving the company's goal of delivering "personal superintelligence to everyone." While Meta has been a major Nvidia customer, this deal signifies a deeper collaboration, including Meta's commitment to deploying Nvidia's Grace CPUs as standalone chips. This deviates from the typical practice of combining CPUs with GPUs for enhanced AI processing efficiency.
Analysts estimate the deal's value in the tens of billions of dollars. This investment aligns with Meta's previously announced plans to spend $135 billion on AI infrastructure this year, and a total of $600 billion by 2028. These plans include constructing numerous data centers globally.
The agreement also encompasses Nvidia's Spectrum-X Ethernet switches and InfiniBand interconnect technologies to link GPU clusters, along with Nvidia's security products for Meta's social media platforms.
This substantial commitment to Nvidia's hardware raises questions about Meta's progress in developing its own AI processors, such as the MTIA chip. Despite efforts to reduce reliance on external suppliers, reports suggest Meta may have encountered "technical challenges" with its in-house training chips, potentially delaying their rollout. The company is also reportedly working on optimizing Nvidia's GPUs for its upcoming frontier model, Avocado, the successor to its flagship Llama 4 LLM.