Anthropic to Deploy Up to 2 Gigawatts of AMD GPUs in Compute Deal

Anthropic will deploy up to 2 gigawatts of AMD Instinct MI450 Series GPUs, with the first gigawatt starting in the first half of 2027, the two companies announced on July 29. AMD also committed to a future strategic equity investment of up to $5 billion in Anthropic, according to the AMD announcement published by StorageNewsletter. For enterprises that run Claude through cloud platforms or directly, the deal is about supply: more of the compute behind a widely used business model will come from a second chip vendor, not only from Nvidia or cloud providers' in-house chips.

What did AMD and Anthropic agree to?

The deployment uses AMD's Helios rack-scale systems built around MI455X GPUs, part of the MI450 Series, with AMD EPYC "Venice" CPUs, AMD Pensando networking and AMD's ROCm software. It extends Anthropic's existing use of AMD's current MI355X GPUs, the announcement said.

The agreement also has an engineering side. Over several years, Anthropic teams will use Claude to tune workloads for AMD Instinct GPUs and to speed up development of ROCm, AMD's answer to Nvidia's CUDA software platform. In the other direction, AMD will roll out Claude broadly across its own engineering and product teams.

Anthropic co-founder and chief compute officer Tom Brown tied the deal to customer demand, saying, "Access to compute is central to keeping Claude at the frontier." The announcement did not say where the systems will be located or give other financial terms.

Why does a 2-gigawatt GPU deal matter to enterprise AI buyers?

Model providers' capacity has become a practical constraint on enterprise AI. The same week, Microsoft said demand for its AI capacity still exceeds supply after it added about a gigawatt in a single quarter, Directions on Microsoft reported. Amazon CEO Andy Jassy said that even after raising 2026 capital spending to about $220 billion, Amazon will not have enough capacity to meet all of this year's demand, according to Stocktwits.

When a model provider has more compute, customers tend to see fewer rate limits, more room for long-context and agent workloads, and more stable service during demand spikes. A second hardware supplier also gives the provider some leverage on price and delivery dates, which matters when GPU supply is the bottleneck.

The deal fits a broader pattern of model providers spreading their bets. Reuters, via The Star, reported that Microsoft is adding Anthropic models alongside its own and OpenAI's. And Amazon's second-quarter profit included a large non-operating gain, mainly from its Anthropic investment, Shelf Awareness reported.

What else did AMD announce on July 29?

The Anthropic agreement was one of several AMD partnership announcements that day. AMD said OpenAI has had access to Helios systems for several months and expects to bring them online through multiple deployment partners in the second half of 2026, the first phase of the 6-gigawatt deployment the two announced in October 2025, according to StorageNewsletter. AMD and Meta said Meta plans to deploy up to 6 gigawatts of AMD GPU-based infrastructure, per a separate release. These are company statements; the timelines depend on hardware that is not yet shipping at scale.

What should enterprise leaders watch next?

Most enterprises will never buy an MI450 directly. But they do buy services that run on them, so a few questions are worth adding to vendor reviews.

Ask where your model runs. Model providers increasingly serve the same model from several chip platforms. Ask providers how they test for consistent output quality and latency across that hardware, and how they report incidents.

Plan for more than one provider. Capacity deals announced this week run into 2027 and beyond, and the timing is the vendors' estimate. Skyflow's Joe McCarron made the case for building around that risk during a AI Connect webinar: "You don't want their SLAs to be your SLA." Our coverage of how to build AI systems that grow reliably covers model-agnostic interfaces and graceful degradation.

Keep exit costs low. A AI Connect panel on self-hosted LLMs, governance and vendor lock-in recommended open frameworks and multi-vendor proofs of concept for the same reason: the supplier landscape is moving quickly.

The first hard checkpoint is the first half of 2027, when Anthropic's first gigawatt on AMD hardware is due to start. Until then, the deal is a signal of how far model providers will go to secure compute, and of how much of enterprise AI's reliability now depends on chip supply chains.

Frequently asked questions

What did AMD and Anthropic announce?

On July 29, 2026, they announced a partnership for Anthropic to deploy up to 2 gigawatts of AMD Instinct MI450 Series GPUs, plus a multiyear engineering collaboration and up to $5 billion in future AMD equity investment.

When will Anthropic's AMD GPUs come online?

The first gigawatt is scheduled to begin in the first half of 2027, according to the companies.

Why does this matter to enterprises using Claude?

More compute for a model provider can mean more capacity and stability for its customers, and a second chip supplier reduces dependence on any one vendor.

Disclosure: AI Enterprise Magazine uses AI tools in its editorial process, including Anthropic's Claude models. Anthropic had no role in this article.

Filed under: Data & AI Infrastructure, AI Strategy & ROI