AMD and Anthropic have announced a landmark strategic partnership that could reshape the AI infrastructure landscape. Under the agreement, Anthropic will deploy up to 2 gigawatts of AMD Instinct MI450 Series GPUs, while AMD commits to a strategic equity investment of up to $5 billion in Anthropic — marking AMD's first investment in an AI company and one of the largest chip-and-capital deals in the industry's history.
Key Highlights
- Anthropic will deploy up to 2 gigawatts of AMD Instinct MI455X (MI450 Series) GPUs across its infrastructure
- The first gigawatt of deployment is scheduled to begin in H1 2027
- AMD will invest up to $5 billion in Anthropic equity, contingent on deployment milestones
- The partnership includes a multi-year software engineering collaboration using Claude to accelerate AMD's ROCm ecosystem
- AMD will adopt Claude models across its own engineering and product development teams
Deal Details
The partnership spans both hardware deployment and financial investment. Anthropic's compute strategy under the deal splits between direct ownership and rented capacity — the company will acquire some AMD chips outright for its own facilities while securing the remainder through cloud providers and neoclouds.
AMD's investment of up to $5 billion is structured as a letter of intent tied to computing power deployment milestones, meaning AMD's capital flows into Anthropic as each phase of the 2-gigawatt buildout reaches operational status. AMD is also reportedly in discussions to provide a financial backstop for Anthropic's future data center leases, further deepening the companies' infrastructure interdependence.
The hardware at the center of the deal is the AMD Instinct MI455X, the latest-generation chip in the MI450 Series, deployed within AMD's Helios rack-scale platform. That full-stack system integrates EPYC "Venice" CPUs, Pensando networking silicon, and AMD's ROCm software stack — positioning AMD as a complete AI infrastructure provider rather than a standalone GPU vendor.
Engineering Collaboration
Beyond the financial terms, the partnership's technical dimension may prove equally significant. The two companies will undertake a multi-year software collaboration aimed at optimizing AI workloads for AMD Instinct GPUs and accelerating development of the ROCm software ecosystem — historically AMD's weakest point in its competition with NVIDIA.
Under the arrangement, Anthropic's Claude models will be used to help write, test, and optimize ROCm code, directly applying frontier AI to the software gap that has long limited AMD's appeal to AI researchers and developers. AMD will also deploy Claude internally across its engineering and product teams.
Lisa Su, AMD's CEO, described the scope: "This collaboration combines Anthropic's leadership in frontier AI with AMD's full strength in computing — from silicon to systems to software."
Competitive Implications for NVIDIA
The deal is AMD's most direct challenge yet to NVIDIA's dominant position in AI accelerator hardware. NVIDIA's H100 and Blackwell GPUs currently supply the vast majority of AI training and inference compute globally, supported by the CUDA software ecosystem that has built a decade-long lead.
By using Claude to strengthen ROCm, AMD and Anthropic are targeting the precise weakness that has kept many AI teams locked in to NVIDIA hardware even when AMD's chips offered competitive raw performance. A 2-gigawatt commitment from one of the world's most prominent AI labs sends a signal to the broader market — hyperscalers, neoclouds, and enterprise buyers — that AMD's platform is production-ready at scale.
The news follows AMD's earlier announcement of the Helios rack system, which attracted major customers including Microsoft Azure, Meta, Oracle, and OpenAI as early adopters.
Background
Anthropic, founded in 2021 by former OpenAI researchers Dario and Daniela Amodei, develops the Claude family of AI models and has grown into one of the world's most valuable AI companies. It had previously relied primarily on custom Google TPUs and NVIDIA GPUs for its training and inference workloads.
AMD's Instinct GPU line has grown rapidly as a challenger to NVIDIA, gaining traction at major cloud providers seeking hardware diversity. The MI300X generation saw significant adoption in 2024 and 2025, and the MI450 series represents AMD's most competitive offering to date.
What's Next
The first gigawatt of AMD MI450 deployment at Anthropic is targeted for H1 2027, giving both companies roughly 12 months to finalize infrastructure buildout, cloud provider agreements, and software stack validation. AMD's $5 billion investment will flow incrementally as those milestones are reached.
For the broader AI chip market, the deal raises pressure on NVIDIA to defend its supply relationships with Anthropic and other frontier labs, while signaling that the era of near-exclusive NVIDIA dominance in frontier AI compute may be drawing to a close.
Source: Bloomberg