A week after allocating to deploy millions of NVIDIA GPUs, Meta has final another AI infrastructure agreement—this time with AMD. The long-term deal alerts Meta’s purpose to broaden its compute stack as it extends one of the largest AI deployments in the industry.
Meta stated the agreement includes deploying up to 6 gigawatts of AMD graphics operational units across its AI data centers. The deployment also includes AI-driven CPUs. Advance delivery of MI450 GPUs, incorporated into AMD’s Helios rack-scale servers, will start later this year.
AMD CEO Lisa Su explained the partnership as a significant milestone for the corporation’s AI goals. “High-performance, energy-efficient infrastructure strengthened for Meta’s workloads” will boost up “one of the industry’s largest AI deployments,” she stated, positioning AMD on the center of the global AI buildout.
Performance-Based Equity Structure
The agreement consists of a overall performance-based warrant permitting Meta to obtain 160 million AMD shares, around 10% of the corporation. The first tranche vests once 1GW of Instinct GPUs ship. Also tranches vest as Meta scales buys in the direction of the entire 6GW target.
Vesting is likewise tied to AMD stock prices thresholds and technical and commercial milestones. This structure mirrors AMD’s October agreement with OpenAI, which additionally included warrants connected to deployment benchmarks. The technique strengthen AMD’s strategy of connecting long-term equity incentives with hyperscaler adoption.
Strategic framework: Competing With Nvidia
The deal comes as AMD seeks to narrow the gap with NVIDIA, which controls around 90% of the AI chip market. NVIDIA, now worth approximately $4.66 trillion, has seen surging demand for its Grace Blackwell systems since their 2024 release. AMD, by contrast, incorporates a market capitalization of around $320 billion.
Industry analyst Ben Bajarin of Creative Strategies referred to that Meta’s scale offers it leverage. “Meta is in a unique position to manage the overall stack and they could use whoever’s compute they need,” he stated. He explained that the settlement as “a punctuation point at the fact that we are compute-restrained, and deals will be performed across the board.”
Bajarin estimates the deal should attain tens of billions of dollars over at least 4 years, given the time needed to install 6GW of infrastructure. He also emphasized a differentiator: the initial deployment includes customized GPUs. “We don’t have any indication Nvidia is doing that,” he stated. “It’s a effective way for them to win some of these deals.”
Scaling Toward a 30-Data Center Footprint
Meta lately devoted to as a whole lot as $135 billion in capital expenses this year because it competes with megacap peers and AI leaders which include OpenAI and Anthropic. The corporation plans to construct 30 data centers, consisting of 26 in the United States.
Meta has long worked with each AMD and NVIDIA and continues to layout in-house silicon. Reports also indicate discussions with Google regarding the ability deployment of its tensor processing devices in future Meta data centers.
