Meta Commits $60bn to AMD for AI Chips Despite Industry Bubble Worries
The owner of Facebook, Meta, has finalised a monumental $60 billion (£44.5 billion) agreement with the US semiconductor firm Advanced Micro Devices (AMD) to purchase artificial intelligence chips. This deal, spanning five years, comes at a time when concerns are mounting over the enormous financial investments flooding the AI sector.
Details of the Strategic Partnership
Under the terms of the pact, Meta will acquire a 10% ownership stake in AMD, reinforcing the partnership's significance. AMD's chief executive, Lisa Su, announced that the company will supply 6GW worth of chips to Meta, commencing with 1GW of the upcoming MI450 hardware in the latter half of this year.
In addition to AMD's flagship graphics processing units (GPUs), Meta plans to purchase central processing units (CPUs), including a customised variant tailored specifically for the social media platform's requirements. Su emphasised that this custom CPU will be optimised to deliver high performance while minimising energy consumption, with the deal encompassing two generations of AMD's CPU technology.
AMD's Growing Influence in the AI Market
This agreement follows a similar partnership AMD secured with OpenAI last year, which was widely regarded as a strong endorsement of AMD's chips and software, leading to a substantial increase in its stock value. The recent series of chip supply deals highlights the AI industry's insatiable demand for advanced processors.
Meta has also independently negotiated a deal with AMD's larger competitor, Nvidia, to procure millions of AI chips, indicating a multi-vendor strategy to meet its extensive computational needs.
Meta's Ambitious AI Infrastructure Plans
Referring to Meta's chief executive, Mark Zuckerberg, Su stated, "So no question Mark is very, very ambitious in what he wants to accomplish, and we want to use every aspect of our technology to really help Meta to accomplish that." She added, "Meta is making a big bet on AMD."
Meta contributed to the design of the MI450 hardware, which is specially optimised for inference computing—the process where AI models like ChatGPT generate responses to user queries. Industry analysts predict that the market for inference hardware will far surpass that for training equipment used in developing large AI models.
Diversified Chip Sourcing Strategy
Santosh Janardhan, Meta's infrastructure head, clarified that the company intends to continue purchasing chips from other suppliers while simultaneously advancing its in-house processor development. Reports from Reuters suggest Meta has been in discussions with Google regarding the use of its tensor processors (TPUs) for AI workloads.
Janardhan explained, "The scale at which Meta is building datacentres and infrastructure requires multiple chip vendors and approaches." He further noted, "All of the chipmakers end up having sort of a seat at the table."
This diversified approach underscores Meta's commitment to leveraging a broad spectrum of technological solutions to support its expansive AI ambitions, even as the industry grapples with fears of an investment bubble.
