Meta Just Locked In Millions of NVIDIA Chips — Here’s Why It Matters
Meta just made one of the biggest bets in AI hardware history — and it tells you everything you need to know about where the real money in artificial intelligence is heading.
On Tuesday, Meta and NVIDIA announced a “multigenerational” partnership that will see Meta build hyperscale data centers powered by millions of NVIDIA’s Blackwell and next-generation Rubin GPUs. But here’s the part most people are missing: Meta isn’t just buying graphics processors. For the first time, a major tech giant is making a large-scale purchase of NVIDIA’s Grace CPU as a standalone chip. That’s NVIDIA muscling directly into Intel and AMD’s turf — and it signals a fundamental shift in how AI infrastructure gets built.
The numbers behind this deal are staggering. Meta previously purchased around 350,000 H100 chips by the end of 2024 and scaled to 1.3 million GPUs by the end of 2025. Now it’s ordering “millions” of next-gen chips. Meanwhile, Meta has guided AI infrastructure spending of $115 billion to $135 billion for 2026 — nearly double last year’s $72.2 billion. That’s not a company hedging its bets. That’s a company going all-in.
Why the sudden hunger for CPUs alongside GPUs? One word: inference. As AI shifts from training massive models to actually running them at scale — think agentic AI, real-time recommendations for 3 billion users, AI features inside WhatsApp — CPUs become critical. They handle the general-purpose computing tasks that feed data to and from the GPUs. Without enough CPU horsepower, those expensive GPUs sit idle. Analysts at Semianalysis recently noted that one of Microsoft’s OpenAI data centers now requires “tens of thousands of CPUs” just to process the petabytes of data generated by its GPUs.
For investors, this deal has several implications worth watching. First, it’s a massive validation for NVIDIA ahead of its February 25th earnings report. The stock popped 1.6% on Wednesday and has been one of the market’s primary sentiment drivers. Second, this “one-throat-to-choke” approach — where Meta sources GPUs, CPUs, and networking equipment from a single vendor — could pressure competitors like AMD, Intel, and even Google’s TPU business. Third, it suggests the AI capex boom isn’t slowing down. If anything, it’s accelerating as companies move from the training phase to deploying AI at consumer scale.
Mark Zuckerberg himself framed the deal in unmistakable terms, saying Meta will use NVIDIA’s Vera Rubin platform to “deliver personal superintelligence to everyone in the world.” Whether or not you buy the superintelligence timeline, the spending is real, the chips are real, and the competitive implications across the semiconductor industry are enormous. With NVIDIA earnings just days away, this partnership just raised the stakes — and the expectations — for what Jensen Huang delivers next.