Multiyear GPU deployment for data centers
Just days after outlining plans to dramatically expand its use of Nvidia processors, Meta has reached another large-scale agreement, this time with Advanced Micro Devices, deepening its commitment to artificial intelligence infrastructure.
The social media company said it will deploy up to 6 gigawatts of AMD graphics processing units to power AI-focused data centers under a multiyear arrangement. The deal also includes the integration of AI-optimized central processing units into its systems.
Initial shipments of AMD’s MI450 GPUs, embedded in Helios rack-scale servers, are expected to begin later this year.
Equity component tied to performance milestones
The agreement features a performance-based warrant allowing Meta to purchase 160 million AMD shares, representing roughly 10% of the chipmaker. The first tranche of warrants will vest once 1 gigawatt of AMD’s Instinct GPUs are shipped, with additional tranches contingent on purchases reaching the full 6-gigawatt target.
Vesting conditions are also linked to AMD share price levels as well as technical and commercial milestones.
AMD CEO Lisa Su described the structure as aligning incentives between the two companies and investors, calling it one of the most transformative transactions in AMD’s AI expansion strategy.
Strategic implications for the AI chip race
For AMD, the partnership represents a significant opportunity to narrow the gap with Nvidia, which holds an estimated 90% share of the AI accelerator market and carries a multi-trillion-dollar market valuation. AMD, by contrast, has a market capitalization under $350 billion and has been working to strengthen its position with its Helios systems, designed to compete with Nvidia’s Grace Blackwell platform.
Industry analysts estimate the Meta agreement could be worth tens of billions of dollars over several years, given the scale and duration required to deploy 6 gigawatts of AI compute capacity.
One distinguishing element of the arrangement is that Meta’s initial deployment reportedly involves customized GPUs, underscoring its ability to tailor infrastructure at the stack level.
Meta’s broader AI investment strategy
The deal follows Meta’s recent commitment to invest as much as $135 billion in capital expenditures this year. The company is building out a global AI footprint that includes plans for 30 data centers, with the majority located in the United States.
AMD previously struck a comparable agreement with OpenAI involving similar share warrant provisions tied to deployment milestones, reinforcing its role as a secondary supplier to hyperscalers seeking alternatives to Nvidia.
Meta continues to work with both AMD and Nvidia and is also developing proprietary chips internally. Reports late last year suggested discussions with Google regarding potential deployment of its tensor processing units in future data centers.
