Meta Platforms has entered a multiyear agreement to purchase up to six gigawatts of AI computing power from Advanced Micro Devices (AMD), in a deal valued at over $100 billion (£74.5 billion). This arrangement could result in Meta acquiring up to a 10% stake in the American chip manufacturer.
Announced on Tuesday, the agreement involves Meta procuring substantial quantities of AMD’s latest GPUs and AI CPUs to support a new generation of data centres. The companies anticipate shipments to commence later this year, with the initial gigawatt of capacity expected to be operational in the latter half of 2026.
The magnitude of Meta’s infrastructure expansion is underscored by the six gigawatts of AI compute capacity, which equates to the electricity required to power over five million homes. As part of the announcement, AMD has granted Meta performance-based warrants to acquire up to 160 million AMD shares at a nominal price of $0.01 each, contingent upon AMD’s stock reaching $600.
Following the announcement, AMD’s shares increased by approximately 7% at the US market opening on Tuesday, while Nvidia’s stock experienced a slight decline.
The agreement represents a significant victory for AMD CEO Lisa Su, as she seeks to challenge Nvidia’s dominance in the AI chip market. Nvidia is estimated to control around 90% of the market for GPUs used in AI training and inference, making it the world’s most valuable publicly traded company, with a market capitalisation exceeding $4.6 trillion last year.
Nvidia has long been one of Meta’s major suppliers and recently announced a “multigenerational” agreement with Meta to construct data centres powered by millions of Nvidia’s chips. Meta has indicated plans to increase its capital expenditure to as much as $135 billion this year, up from $72 billion last year, to expand its AI infrastructure. CEO Mark Zuckerberg has stated that the company aims to deploy “tens of gigawatts” of compute power this decade and “hundreds of gigawatts or more over time.”
In a statement, Zuckerberg emphasised the importance of diversifying Meta’s compute resources, expressing confidence in AMD as a long-term partner. Lisa Su remarked that the agreement positions AMD “at the centre of the global AI buildout,” highlighting the company’s commitment to delivering “high-performance, energy-efficient infrastructure optimised for Meta’s workloads.”
This deal marks one of the first instances of AMD supplying custom AI chips tailored for a specific customer’s needs. The chips utilise a modular ‘chiplet’ architecture, designed to facilitate performance optimisation for particular tasks. The structure of this agreement is reminiscent of a similar deal AMD made with OpenAI in October, which also included warrants for up to 160 million AMD shares linked to purchase volumes and share price targets.
Some critics have described these negotiations as a form of “circular financing,” where equity incentives are tied to substantial investments. With a limited number of hyperscale customers capable of purchasing chips in such volumes, both AMD and Nvidia have employed long-term supply agreements and equity incentives to secure demand.
The announcement occurs amid increasing investor scrutiny regarding the scale of AI spending by major tech companies. Despite concerns about capital intensity, Meta recently credited record quarterly revenue to AI-driven improvements. AMD reported a 34% growth in fourth-quarter sales to $10.27 billion, while Nvidia is expected to report a year-on-year revenue increase of around 68% when it releases its results this week.
For AMD, valued at approximately $320 billion, this agreement with Meta signifies a pivotal step in its efforts to become a credible alternative to Nvidia in AI infrastructure.
You must be logged in to post a comment.