Su said the warrant structure would help “make sure that we are always a clear seat at the table when [Meta] are thinking about what they need next.”
Meta’s chief executive Mark Zuckerberg said he expected AMD to be “an important partner for many years to come.”
Meta has said that it will almost double its AI infrastructure spending this year to as much as $135 billion, as US tech giants rush to build the data centers to train and run AI software. It is already one of AMD’s biggest AI chip customers.
“We don’t believe that a single silicon solution will work for all of our workloads,” said Santosh Janardhan, Meta’s head of infrastructure. “There’s a place for Nvidia, there’s a place for AMD and… there’s a place for our own custom silicon as well. We need all three.”
Under the deal, AMD will build a custom version of its MI450 AI chips for Meta. They will be used primarily for “inference” workloads, the process of running models after they have been trained.
The chips need 6 gigawatts of power—equivalent to the amount required by 5 million US households for a year.
Increasingly creative funding arrangements to support massive AI infrastructure build-outs have emerged in recent years, leading to warnings about circular financing.
AMD has, for example, helped data center builder Crusoe secure a $300 million loan from Goldman Sachs by offering a backstop guaranteeing the use of its chips if Crusoe is unable to find customers after installing them in an Ohio facility.
Tech giants such as Meta, historically flush with cash, are meanwhile facing the prospect of tapping bond and equity markets or stemming capital returns to shareholders to help fund their unprecedented infrastructure plans. The Facebook and Instagram parent raised $30 billion in October, marking its biggest bond sale to date.
© 2026 The Financial Times Ltd. All rights reserved. Not to be redistributed, copied, or modified in any way.
