AMD closes $5 billion partnership with Anthropic, takes aim at Nvidia

In a historic and monumental market move that shakes the semiconductor duopoly and redefines competition in supercomputer infrastructure for artificial intelligence, US chipmaker AMD and renowned AI developer Anthropic have announced a long-term strategic alliance. The transaction involves large-scale technology supply and capital investment.

Strategic $5 billion capability investment

As a cornerstone of the multibillion-dollar agreement, the semiconductor manufacturer has committed to a strategic capital investment of up to $5 billion in the developer. The funds will be released in installments, directly tied to the achievement of specific technical milestones and physical deployment targets for logical servers in processing centers.

The strategic alliance provides cash stability for the creator of the Claude family and enables massive expansion of its proprietary cloud capacity. The move is part of the startup's multi-vendor compute strategy, establishing strong alternatives in the global market amid the critical shortage of traditional graphics chips that limits datacenter expansion in several countries.

Deployment of two Gigawatts and MI450 accelerators

The practical operational arm of the agreement involves the colossal deployment of up to two gigawatts of AMD Instinct MI450 series graphics accelerators, focusing on the ultra-high-performance MI455X model. The supercomputing infrastructure will begin installation in the first half of 2027 (UTC) and will integrally use Helios rack systems, specifically designed to support massive agentic workloads.

To ensure fast traffic and stable connections, the complex will use next-generation AMD EPYC processors codenamed Venice and the Pensando physical network architecture. The companies estimate that the sale and technical maintenance of server clusters could generate up to $27 billion for the manufacturer over the entire strategic supply cycle.

Technical co-development and ROCm software optimization

The partnership also establishes a broad and deep software engineering collaboration between the teams. Engineers from the creator of Claude will use their AI models to automate testing and optimize workloads on the partner's graphics accelerators. The joint effort aims to accelerate the maturity and stability of the open-source ROCm software ecosystem, a direct competitor to Nvidia's proprietary CUDA platform.

In turn, the manufacturer will broadly and unrestrictedly integrate the Claude model into its workflows and internal engineering to design the next generation of microchips. This unification of capabilities directly challenges the technological sovereignty of centralized infrastructures in the United States operated by brands such as Microsoft, search engine Google, and the closed neural networks maintained under the control of rival OpenAI.

High-speed network switches and stable buses in datacenters

Interconnecting and securely orchestrating a massive infrastructure of two gigawatts of Instinct chips requires efficient datacenters with stable physical buses and fast optical network switches. Data traffic and internal networks of the complex depend on optical switches designed by specialized physical network designers, such as semiconductor manufacturer Broadcom. The lowering cost of local optical switches optimizes the centers and reduces the electrical consumption of the partner companies' computers in the IT market.