AMD puts $5 billion on the table for Anthropic
Advanced Micro Devices (AMD) is making a bold move. The chipmaker has agreed to invest up to $5 billion in Anthropic, the startup behind the Claude AI assistant. This isn’t just a financial bet — it’s a sprawling infrastructure deal that could reshape how AI models are trained and deployed.
The agreement covers tens of billions of dollars’ worth of AI systems. Anthropic will deploy up to two gigawatts of computing capacity using AMD’s next-generation Instinct MI450-series accelerators. The first gigawatt is expected to come online in the first half of 2027.
Here’s the twist: AMD’s investment is tied to deployment milestones. If Anthropic hits certain targets, the money flows. Neither company has disclosed what those conditions are, or what happens if delays creep in.
How the deal works — and how it’s different
This isn’t a simple check-writing exercise. The transaction combines an equity investment from AMD with a separate, large hardware order from Anthropic. The companies haven’t said Anthropic must use AMD’s investment to pay for the systems.
It’s a structure AMD has used before — but with a twist. In October, AMD agreed to supply OpenAI with systems supporting up to six gigawatts of capacity. That deal included warrants allowing the ChatGPT developer to acquire roughly 10% of AMD’s stock, vesting in stages based on deployment, commercial, and share-price conditions.
AMD struck a similar deal with Meta in February, covering up to six gigawatts of GPU capacity, with performance-based warrants tied to shipment and purchase targets.
The Anthropic agreement is different. AMD is making a direct equity investment, not issuing warrants. That puts more skin in the game for the chipmaker — and gives Anthropic cash it can use however it sees fit, within the broader partnership.
What Anthropic is actually buying
Anthropic will deploy AMD Helios systems. These aren’t just GPUs — they’re integrated rack-scale platforms combining MI455X accelerators from the Instinct MI450 series, EPYC “Venice” processors, Pensando networking hardware, and ROCm software.
This matters because it’s a full-stack approach. Instead of buying chips and figuring out the rest themselves, Anthropic gets a pre-integrated system designed to work together. AMD chair and CEO Lisa Su called it a “deepened partnership” that brings together Anthropic’s AI leadership with AMD’s high-performance computing.
Anthropic has already been testing AMD’s earlier MI355X accelerators. The new deal scales that relationship dramatically — from evaluation to gigawatt-scale deployment.
Gigawatts: What that number really means
Two gigawatts of power capacity is enormous. For context, Anthropic has separately secured more than 300 megawatts through SpaceX’s Colossus 1 facility in Memphis, which houses over 220,000 Nvidia GPUs. The AMD deployment represents more than six times that power capacity.
But power capacity isn’t the same as computing performance. The two facilities use different hardware and deployment models. AMD executives have said that building one gigawatt of AI infrastructure can cost tens of billions of dollars, depending on the equipment and facilities involved.
Some of the AMD systems will go into Anthropic’s own data centres. Others will be hosted by cloud providers and specialist AI infrastructure companies. AMD and Anthropic are also hunting for operators capable of hosting these massive systems.
Lisa Su noted that gigawatt-scale capacity requires 12 to 24 months of planning before deployment. That’s why the first systems won’t arrive until 2027.
Anthropic’s multi-supplier strategy takes shape
Anthropic is building a diverse compute network. The AMD systems will run alongside infrastructure based on Nvidia GPUs, Amazon Trainium processors, and Google tensor processing units.
Amazon remains Anthropic’s primary cloud and training partner. The startup has secured up to five gigawatts of capacity from Amazon and uses more than one million Trainium2 processors. Trainium3 deployments are planned for 2026.
Anthropic has also locked in multiple gigawatts of next-generation TPU capacity from Google and Broadcom, with deployments expected in 2027. Claude is available through Amazon Web Services, Google Cloud, and Microsoft’s Azure-based Foundry platform.
Anthropic co-founder and chief compute officer Tom Brown explained the logic: “Running across a diversified range of hardware lets us map the right workloads to the right hardware.” He said the AMD partnership gives Anthropic additional capacity while allowing the companies to optimise systems for training and serving Claude.
The company has also agreed to use computing capacity at SpaceX’s Colossus facilities and has discussed leasing infrastructure from Meta. Reuters reported that potential Meta deal could be worth up to $10 billion over two years.
Engineering collaboration and software development
The AMD agreement includes a multi-year engineering programme. The companies will use Claude to optimise workloads for Instinct accelerators and support development of AMD’s ROCm software platform. AMD plans to deploy Claude across its engineering and product-development teams.
This gives Anthropic a dual role: customer of AMD’s infrastructure and participant in developing the software that runs workloads on that infrastructure. It’s a tight feedback loop that could help AMD close the software gap with Nvidia’s CUDA ecosystem.
Market reaction and what comes next
AMD shares rose 2.4% after the agreement was announced. The company’s stock has more than doubled since the start of the year, outperforming the broader Philadelphia Semiconductor Index.
The deal positions AMD as a serious player in the AI infrastructure race, alongside Nvidia’s reported talks about investing up to $30 billion in OpenAI. Chip suppliers are increasingly pairing investments or equity incentives with commercial agreements involving major AI developers.
For Anthropic, the AMD deal provides another pillar in its compute strategy — and signals that the startup is thinking long-term about hardware diversity. “Access to compute is central to keeping Claude at the frontier and meeting demand from our customers,” Brown said. “By partnering with AMD across the stack, we are securing the capacity we need and optimising it for training and serving Claude.”
The real test will come in 2027, when the first gigawatt of AMD-powered infrastructure is supposed to go live. Until then, the industry will be watching — and counting the milestones.
