Meta Bets Billions on Nvidia Chips to Power Its AI Ambitions

In a massive expansion of a long-running partnership, Meta has committed to purchasing millions of Nvidia's latest AI chips for its global data center network. The deal, announced Tuesday, includes Nvidia's new Grace standalone central processing units (CPUs) and its next-generation Vera Rubin AI systems.

Meta CEO Mark Zuckerberg framed the move as a continuation of the company's push to develop advanced AI, reiterating a goal he set last summer. While the financial details were not disclosed, industry analysts estimate the multiyear agreement is valued in the tens of billions of dollars. This aligns with Meta's previously stated plan to invest up to $135 billion in artificial intelligence during 2026 alone.

A key element of the agreement is Meta's large-scale adoption of Nvidia's Grace CPUs as standalone processors, a first for the chipmaker's data center technology. These CPUs are engineered to handle specific AI tasks independently, complementing the more powerful graphics processors (GPUs) that drive model training.

The partnership also secures Meta's supply of Nvidia's coveted Blackwell and upcoming Rubin GPUs, which have faced high demand and backorders. The deal encompasses Nvidia's networking hardware and security technology, some of which will be integrated into features on WhatsApp.

Despite this significant commitment to Nvidia, Meta continues to diversify its chip suppliers. The company develops its own silicon and uses processors from rivals like AMD, a common strategy among tech giants to ensure supply and manage costs. Meta's overall U.S. infrastructure plan calls for $600 billion in spending by 2028, including 30 data centers, with 26 planned for American sites.

The announcement buoyed shares of both Meta and Nvidia in after-hours trading Tuesday, while shares of AMD declined.

Source: CNBC