This isn’t a purchase order. It’s an architecture commitment. Meta will deploy millions of Nvidia GPUs across two chip generations (Blackwell now, Rubin next) alongside the first large-scale deployment of Nvidia’s Grace ARM-based CPUs without paired GPUs.
Analyst Ben Bajarin estimated the deal’s value in the “tens of billions” range, though neither company disclosed a specific figure. The commitment sits within Meta’s broader $600 billion U.S. infrastructure investment plan, which includes more than 30 data centers across the country. CNBC confirmed the multi-generational scope and the first-of-its-kind Grace CPU deployment.
The Grace-only piece deserves separate attention. Nvidia has pushed Grace as a standalone data center CPU for inference and data processing workloads, distinct from its GPU-paired configurations. Meta adopting it at scale validates the product in production, not just in benchmark demos.
For infrastructure investors, the signal is straightforward: the largest AI spenders are not hedging their chip bets. Meta is doubling down on a single vendor’s roadmap across compute, CPU, and networking. That level of lock-in creates stability for Nvidia’s revenue trajectory but also concentrates risk. If Rubin’s architecture underperforms expectations, Meta has limited alternatives at this scale.
Meta’s separate $10 billion Indiana data center provides physical context for what this commitment looks like on the ground.