AMD signs Anthropic AI server deal as it expands rack-scale AI push
Anthropic plans to procure up to 2 gigawatts of AMD AI chips from 2027 under a deal reportedly worth tens of billions, while AMD also highlighted Helios rack systems and other gigawatt-scale customer agreements.
The WSJ exclusive directly confirms every specific detail in the claim body: the AMD-Anthropic deal, Anthropic buying up to 2 GW of Instinct MI450 chips from H1 2027, and AMD investing up to $5 billion tied to deployment milestones. BlockBeats independently corroborates the $5 billion investment and 2 GW figure, both citing WSJ. Timing coincides with AMD's Advancing AI 2026 event, where such customer announcements were anticipated (Yahoo/Jefferies). The headline phrase 'hundreds of billions' is looser than WSJ's 'tens of billions,' but the precise deal terms stated in the claim match the primary source, supporting a likely_true assessment.
Advanced Micro Devices has signed a multibillion-dollar AI infrastructure agreement with Anthropic under which the AI startup plans to procure up to 2 gigawatts of AMD chips for data centers starting in the first half of 2027, with The Wall Street Journal reporting the deal is worth tens of billions of dollars. AMD may also invest up to $5 billion, subject to deployment milestones. The agreement was highlighted as AMD unveiled its new Helios rack system at its Advancing AI 2026 event in San Francisco on July 23, part of a broader strategy to move from selling standalone chips to full rack-scale systems and to compete more directly with Nvidia. AMD also said it has gigawatt-scale GPU supply agreements with Meta and OpenAI, as model developers seek more AI compute capacity and diversify hardware suppliers.
- Instinct MI450 chips: AMD's next-generation AI accelerators designed for data center workloads.
- Deployment milestones: Predefined operational targets that trigger funding, investment or other contractual commitments.
- Rack system: A hardware system that integrates GPUs, CPUs and networking components into a unit used to build AI servers and data centers.