Meta’s AI ambitions are taking a bold new direction. After years of betting heavily on AMD’s Instinct MI300X and Helios platforms, the social media giant is now aligning with NVIDIA for a long-term partnership that includes millions of next-generation Blackwell GPUs and the upcoming Vera Rubin AI accelerators. This move signals a major realignment in Meta’s data center strategy, one that could reshape the competitive landscape for AI infrastructure.

What changed? Until recently, Meta was AMD’s largest AI customer, accounting for over 40% of the company’s AI revenue last year. The partnership centered on AMD’s Instinct MI300X and custom Helios server racks designed specifically for Meta’s Open Rack Workload (ORW) specifications. However, the new NVIDIA deal introduces a multi-year, multi-generational commitment that extends beyond GPUs to include NVIDIA’s standalone Grace CPUs and the yet-to-be-released Vera Rubin platform.

NVIDIA Secures Meta as Major AI Partner, Shifting Strategy with Next-Gen Vera Rubin and Blackwell Chips
  • Blackwell and Blackwell Ultra GPUs: Meta will deploy millions of these chips, which are already proving critical for AI workloads like generative models and large-scale inference. The Blackwell Ultra variant, in particular, is optimized for agentic AI workloads, offering up to 50 times better performance per watt compared to previous generations.
  • Vera Rubin accelerators: While details remain scarce, Vera Rubin is expected to combine GPU and CPU capabilities into a single, highly efficient chip, potentially bridging the gap between NVIDIA’s traditional GPU-focused approach and Meta’s need for scalable, heterogeneous computing.
  • Grace CPUs: NVIDIA’s standalone CPU offering will complement the GPUs, enabling Meta to build more flexible and power-efficient data center clusters.
  • Networking and security: The partnership also includes NVIDIA’s Spectrum-X Ethernet for high-speed, scalable networking and Confidential Computing for secure AI workloads, particularly on platforms like WhatsApp.

This isn’t just another hardware deal—it’s a strategic pivot. By integrating NVIDIA’s Blackwell and Vera Rubin platforms, Meta is positioning itself to accelerate the development of personal superintelligence, as outlined by CEO Mark Zuckerberg. The move also underscores NVIDIA’s dominance in the AI infrastructure space, further solidifying its lead over competitors like AMD and Intel.

For Meta, the shift reduces dependency on a single vendor while gaining access to NVIDIA’s latest advancements in AI acceleration. For NVIDIA, the partnership adds a major enterprise customer—one that rivals Microsoft and OpenAI in scale. The financial terms remain undisclosed, but the agreement’s longevity suggests a commitment that could stretch well into the next decade.

As Meta’s AI roadmap evolves, so too does the battle for control over the next generation of data center hardware. With Blackwell and Vera Rubin at the forefront, NVIDIA has just made a high-stakes play to keep Meta—and by extension, its AI workloads—closely tied to its ecosystem.