Meta and AMD have entered into an agreement worth at least 100 billion dollars. For that amount, Meta is acquiring a massive quantity of chips as well as shares in AMD.
Meta is set to purchase at least 100 billion dollars worth of chips from AMD. This involves 6 gigawatts of AMD MI450 accelerators. The MI450 is the planned successor to the MI355X and a competitor to Nvidia’s Blackwell and Rubin chips, among others. These accelerators can both train new complex models and provide inference at scale.
In the coming years, Meta aims to acquire so many chips that the power required to run them is equivalent to the consumption of about five million households. With this investment, Meta intends to power its further development of AI models.
This investment is of the type directly responsible for the DRAM and SSD shortages in the PC market today. The scale is so vast that it exceeds the capacity of component manufacturers.
Following in OpenAI’s footsteps
Meta is not the only one looking to buy six gigawatts of MI450s. Last year, OpenAI closed a very similar deal with AMD. The timing of this agreement is notable. Just a few days ago, Nvidia and OpenAI backed away from a chip agreement, also worth 100 billion dollars. That deal was replaced by a smaller agreement, driven in part by fears that the size of the investment might not be immediately recouped by future revenues. AMD and Meta are not concerned about this with their new agreement.
read also
Nvidia won’t invest $100 billion in OpenAI after all: more caution, less money, but no bubble
With its billions, Meta is buying more than just chips from AMD. Within the framework of the agreement, Meta can also acquire ten percent of the chip specialist’s shares.
A larger role for AMD
The approach where OpenAI and Meta invest heavily in AMD, and also gain a financial stake, fits into CEO Lisa Su’s strategy. She is fighting from an underdog position against the Nvidia of her distant cousin Jensen Huang, which controls more than 90 percent of the AI chip market. By bringing major AI developers on board, Su can boost the relevance of her own chips and help the surrounding ecosystem mature.
For its part, Meta does not want to be exclusively dependent on Nvidia. The investment is intended to ensure more diversity in the technology behind the AI data center infrastructure. The first chip deliveries are expected to take place later this year.