Meta Buys Millions of Nvidia GPUs: $135B AI Infrastructure Deal

Mark Zuckerberg has just signed the largest check in the history of artificial intelligence. Meta will purchase millions of Nvidia processors over the coming years under a multi-year agreement whose estimated value runs into the "tens of billions of dollars," according to analyst Ben Bajarin of Creative Strategies. The move cements Meta as the anchor customer Nvidia needed to silence market doubts about competition from in-house chips.

The deal includes the current Blackwell Ultra (B300) chips and the forthcoming Rubin (R100) chips, which will begin mass production in the second half of 2026. Additionally, for the first time, Meta will deploy Nvidia's Grace central processing units independently, not merely as GPU complements. "There is no one who deploys AI at the scale of Meta," said Nvidia CEO Jensen Huang in a joint appearance with Zuckerberg.

Nvidia shares rose more than 5% following the announcement on February 17, adding hundreds of billions in market capitalization. Meta accounts for approximately 9% of Nvidia's total revenue, but this deal deepens a relationship spanning more than a decade. AMD fell 4% on the same day, signaling that Meta, which had been exploring Google's TPU chips, chose to double down on Nvidia.

The investment falls within Meta's capital budget of $115–$135 billion announced in January for 2026, allocated almost entirely to AI infrastructure. Zuckerberg explained that the goal is to "deliver personal superintelligence to everyone in the world," a vision he unveiled last July. The chips will power the training of Llama 5 and the deployment of AI agents across WhatsApp, Instagram, and Facebook.

Building data centers to house millions of chips is a monumental logistical feat. Supply chain constraints, particularly around HBM4 memory and advanced liquid cooling systems, could delay timelines. Moreover, the industry faces a paradox: massive AI investment with economic returns still uncertain. If Meta cannot monetize these chips through assistants that users actually pay for, the market faces a brutal correction.

The deal marks a decisive moment: AI is no longer experimental, it is critical infrastructure comparable to the internet or electricity. But it also creates systemic risks. If a handful of companies control key hardware, they can influence pricing, innovation, and availability. Nvidia has sold its entire future capacity to Meta, Google, and Microsoft. The question remains: what happens to the rest of the market?