AMD will invest up to $5 billion in Anthropic and supply up to 2 gigawatts of Instinct MI450 GPUs to the AI company, the two firms announced Wednesday. The first gigawatt lands in the first half of 2027, deployed inside AMD's new Helios rack-scale systems. It is Anthropic's largest publicly disclosed silicon commitment outside its existing Nvidia and Google TPU supply, and it hands AMD its most significant frontier-lab customer to date.
The scale of the deal matters because 2 gigawatts of GPU capacity is roughly the power draw of two million US homes and puts AMD alongside Nvidia as a primary training platform for Claude. AMD has spent the past two years chasing Nvidia's dominance in AI accelerators, and the Anthropic order — combined with an earlier warrant issued to OpenAI for 160 million shares, or roughly 10% of AMD — establishes MI450 and Helios as credible alternatives at gigawatt scale.
Anthropic's compute hunger has been the defining story of its year. The company said in April that demand for Claude and Claude Code had caused what it called inevitable strain on its infrastructure, hitting reliability and performance during peak hours. It has since signed a cascade of infrastructure deals: multi-billion-dollar arrangements with Amazon in April, a multi-gigawatt agreement with Google and Broadcom the same month, an all-capacity lease of Elon Musk's Colossus 1 data center in Memphis, Tennessee via SpaceX in May, and additional capacity with TeraWulf.
“By partnering with AMD across the stack, we are securing the capacity we need and optimizing it for training and serving Claude.”— Tom Brown, Anthropic cofounder and chief compute officer
Key facts
- 01AMD will invest up to $5 billion in Anthropic and supply up to 2 gigawatts of Instinct MI450 GPUs in Helios rack-scale systems.
- 02The first gigawatt deploys in the first half of 2027, following Anthropic's earlier deals with SpaceX, TeraWulf, Google, Broadcom, and Amazon.
- 03Anthropic's run-rate revenue hit $47 billion, up from roughly $10 billion for all of last year, and it closed a round at a $965 billion valuation.
- 04AMD previously issued OpenAI a warrant for 160 million shares, roughly 10% of the company, tied to deployment and share-price milestones.
- 05Anthropic is paying SpaceX $1.25 billion per month through May 2029 for compute at the Colossus 1 data center in Memphis.
The SpaceX arrangement alone is worth $15 billion a year — Anthropic will pay $1.25 billion per month through May 2029 for Colossus 1 compute, according to SpaceX's prospectus. Preliminary talks with Meta over additional leased capacity have also been reported. AMD now joins that supplier list, and the deal includes a multi-year engineering collaboration in which AMD will use Claude across its software development, engineering, and product work.
The financial arithmetic behind the buying spree is aggressive but grounded in real revenue. Anthropic said in May that its run-rate revenue had topped $47 billion, up from the roughly $10 billion in revenue it generated over the entirety of last year. That growth, driven largely by Claude Code adoption inside developer tooling, supported a funding round at a $965 billion valuation and a confidential S-1 filing with the SEC in June for a potential IPO as soon as this year.
For AMD, the customer concentration is a feature. CEO Lisa Su has been methodically building a roster of hyperscale AI buyers to counter Nvidia's grip, and pairing OpenAI and Anthropic — the two most prominent frontier labs outside Google DeepMind — as anchor MI450 customers gives the Helios platform the deployment volume it needs to iterate against Nvidia's Blackwell and Vera roadmaps.
“We are thrilled to deepen our partnership with Anthropic and deploy AMD Helios at gigawatt scale.”— Lisa Su, AMD CEO
The structure of the investment is unusual. AMD is putting up to $5 billion into the customer that is, in turn, buying its chips — a pattern that increasingly defines the AI infrastructure economy, where chipmakers, cloud providers, and model labs invest into each other's balance sheets to lock in multi-year supply. Nvidia has used similar arrangements with CoreWeave and others. Critics of these circular structures argue they inflate perceived demand; defenders point out that the underlying compute contracts are binding and the deployed capacity is real.
There are execution risks worth noting. The first gigawatt does not deploy until the first half of 2027, meaning Anthropic's near-term compute constraint remains real, and the MI450 has to hit its performance and reliability targets against a moving Nvidia roadmap. Helios is a new rack-scale platform, and gigawatt-class deployments of first-generation systems have historically produced yield and thermal surprises. Anthropic is hedging by keeping multiple suppliers active rather than betting on any single one.
The strategic read for the AI market is that Anthropic has now bought its way out of the single-supplier problem. Every frontier lab that survives the next two years will need diversified silicon, diversified power, and diversified geography — and Anthropic is the first outside Google to plausibly claim all three. That is what a $965 billion valuation and a pending IPO buy you: the ability to place gigawatt-scale bets on a chip line that has not yet shipped, and to make AMD a first-class citizen in a market Nvidia has owned for a decade.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
