Free Audit

Home / Latest / Research

Research

AMD Locks In Anthropic for 2 Gigawatts of AI Chips, Backs It With a $5 Billion Bet

AMD and Anthropic announced a multi-year hardware and investment agreement on July 22, 2026. Anthropic gets up to 2 gigawatts of MI450-series compute; AMD gets a stake in one of the two companies still setting the pace in frontier AI.

Published — claps

AMD and Anthropic announced a strategic partnership on July 22, 2026, under which Anthropic will deploy up to 2 gigawatts of AMD’s Instinct MI450-series GPUs in AMD’s Helios rack-scale systems. AMD, in turn, committed to a strategic equity investment of up to $5 billion in Anthropic, with the money released in stages as Anthropic hits deployment milestones.

The first gigawatt of that capacity starts landing in the first half of 2027. AMD CEO Lisa Su has said full deployment of the two-gigawatt commitment needs 12 to 24 months of planning before hardware goes live, so this is a multi-year buildout, not a single quarter’s order.

TL;DR

AMD and Anthropic signed a deal on July 22, 2026 pairing a hardware supply agreement (up to 2GW of MI450-series Instinct GPUs in Helios racks, starting H1 2027) with an AMD equity investment of up to $5 billion in Anthropic, paid out against deployment milestones rather than upfront. Unlike AMD’s earlier deals with OpenAI and Meta, which used AMD stock warrants to compensate the customer, this one runs money the other direction: AMD is buying into Anthropic, not paying Anthropic in AMD shares. The deal also includes a software collaboration where Claude helps optimize workloads for AMD’s ROCm stack, and AMD adopts Claude internally. AMD stock rose modestly on the news (reports put it around 2.4%), a smaller pop than some expected, because much of the announcement had already been anticipated ahead of AMD’s Advancing AI 2026 event. Anthropic now runs production workloads across four chip architectures: Nvidia GPUs, Google TPUs, AWS Trainium, and now AMD Instinct.

What was announced

The core of the deal, per AMD’s own announcement and its SEC filing counterpart on the investor relations site:

  • Anthropic will deploy AMD Helios rack-scale solutions built on AMD Instinct MI455X GPUs (part of the MI450 series), paired with AMD EPYC “Venice” CPUs and AMD Pensando networking, running on AMD ROCm software.
  • Total scale: up to 2 gigawatts of compute capacity.
  • Timeline: the first gigawatt begins deploying in the first half of 2027.
  • This builds on Anthropic’s existing use of AMD’s current-generation MI355X GPUs, so it is an expansion of an existing relationship, not a cold start.
  • AMD and Anthropic will collaborate to use Claude to optimize AI workloads specifically for AMD Instinct GPUs, and to accelerate development of AMD’s ROCm software stack.
  • AMD will broadly adopt Claude across its own engineering and product development teams.

AMD CEO Lisa Su: “This collaboration brings together Anthropic’s leadership in frontier AI with the full strength of AMD high-performance computing.” Anthropic co-founder and Chief Compute Officer Tom Brown: “Running across a diversified range of hardware lets us map the right workloads to the right hardware.”

Deal terms: the equity part is the interesting part

The chip order is large. The money is what makes this deal stand out from the usual AI infrastructure press release.

AMD committed to a strategic equity investment of up to $5 billion in Anthropic. That investment isn’t a single wire transfer. It’s structured as staged, milestone-based drawdowns tied to Anthropic actually deploying the hardware on schedule. If deployment slips, the investment presumably slips with it.

Compare that to how AMD structured its two other gigawatt-scale AI deals this cycle:

  • OpenAI (October 2025): a 6-gigawatt hardware commitment, paired with AMD issuing OpenAI warrants for up to 160 million AMD shares at a penny each, vesting in tranches through 2030.
  • Meta (February 2026): another 6-gigawatt commitment, with a similar performance-based warrant structure giving Meta the right to buy up to 160 million AMD shares.

In both of those deals, AMD paid its customers in AMD stock to secure the order. With Anthropic, the capital flows the other way: AMD is buying a stake in Anthropic, not handing Anthropic a stake in AMD.

One read on why: AMD didn’t need to sweeten this one with warrants. Anthropic’s revenue has been climbing fast enough, and compute is scarce enough, that AMD could structure the deal as an investment instead of a discount. Whether that’s a sign of AMD’s negotiating strength or Anthropic’s, depends who you ask.

Why it matters

Nvidia still runs the AI chip market. Nobody serious argues otherwise. But this deal is a data point in a trend that’s been building since 2025: frontier labs don’t want to build their entire stack on one vendor.

Anthropic’s compute setup was already unusual. Its primary training substrate runs on Google TPUs and AWS Trainium, not Nvidia GPUs, a choice that traces back to Amazon’s roughly 22% stake in Anthropic and to price and availability advantages Anthropic has found in Google’s TPU pods. Anthropic also has a separate agreement with Google and Broadcom for multiple gigawatts of next-generation TPU capacity starting in 2027, part of a $50 billion U.S. infrastructure buildout the company announced in November 2025.

Add AMD Instinct and Anthropic now runs production workloads across four chip architectures. That’s not a hedge. That’s a company treating compute sourcing the way a hyperscaler does: match the workload to whichever silicon does the job best.

For AMD, the number that matters is the running total. Stack this against OpenAI’s 6 gigawatts and Meta’s 6 gigawatts and AMD’s publicly disclosed commitments now top 14 gigawatts of future AI accelerator deployment, before counting Oracle’s and Microsoft’s smaller, unquantified Helios commitments announced the same week at AMD’s Advancing AI 2026 event.

AMD shares rose only modestly on the announcement, reports put it around 2.4%, a muted move given the size of the deal. Analysts at Jefferies had already flagged an Anthropic announcement as likely ahead of the Advancing AI event, so a chunk of the good news was priced in before the ink dried. AMD stock had already roughly doubled over the course of 2026.

Anthropic’s compute appetite tracks its growth. Run-rate revenue reportedly surpassed $30 billion, up from roughly $9 billion in late 2025, with more than 1,000 business customers now spending $1 million or more a year on Claude. That kind of demand doesn’t get satisfied by one chip vendor’s supply chain.

Compute economics eventually show up downstream in product quality, including how AI systems answer questions. More training and inference capacity, spread across cheaper and more available hardware, is part of what lets a lab keep serving fast, well-grounded answers instead of rationing quality during demand spikes. Anyone watching how their content performs when Claude answers a question is, indirectly, watching deals like this one play out.

Frequently asked questions

What exactly did AMD and Anthropic agree to? Anthropic will deploy up to 2 gigawatts of AMD’s Instinct MI450-series GPUs (specifically MI455X chips) in AMD’s Helios rack systems, starting with the first gigawatt in the first half of 2027. AMD separately committed to invest up to $5 billion in Anthropic, released in stages tied to deployment milestones.

Is this the same as AMD’s deals with OpenAI and Meta? No. The hardware commitment is smaller (2GW versus 6GW each for OpenAI and Meta), and the financial structure is reversed. OpenAI and Meta received AMD stock warrants as part of their deals. Anthropic did not; instead AMD is putting cash equity into Anthropic.

Does this mean Anthropic is moving away from Nvidia? No. Anthropic still uses Nvidia GPUs alongside Google TPUs, AWS Trainium, and now AMD Instinct. The company has been explicit that it runs a multi-vendor stack on purpose, to match specific workloads to whichever hardware suits them and to avoid depending on a single supplier.

How did AMD’s stock react to the news? Modestly. Reports put the move at around 2.4% on the day, smaller than some expected given the deal’s size. Analysts had partly anticipated an Anthropic announcement ahead of AMD’s Advancing AI 2026 event, so some of the reaction was already priced in.

What is AMD Helios? Helios is AMD’s rack-scale AI infrastructure platform, combining Instinct GPUs, EPYC “Venice” CPUs, and Pensando networking hardware into a single deployable system, running on AMD’s ROCm software stack. It’s AMD’s answer to Nvidia’s rack-scale systems for large AI deployments.

Primary sources and further reading

Continue Reading

Research

AI Agent Startups Raised $1.8 Billion in July as Google and Anthropic Pushed Agents Further Into Products

 Research

DeepMind, OpenAI, and Anthropic CEOs Agree AI Needs Independent Testing, Not on Who Runs It

 Research

Anthropic Discloses Claude Models Breached Three Companies During Security Tests