AMD and Anthropic are joining forces—and on a massive scale. On July 22, 2026, the two companies announced a strategic partnership under which Anthropic will deploy up to 2 gigawatts of AMD Instinct MI450 GPUs. The first phase of the deployment, totaling one gigawatt, is set to begin in the first half of 2027. And AMD is taking it a step further: The chipmaker plans to invest up to $5 billion in Anthropic.
Key Points at a Glance
- Anthropic will deploy up to 2 gigawatts of AMD Instinct MI450 (model MI455X) in AMD Helios rack-scale systems.
- The first gigawatt will go into operation in the first half of 2027.
- AMD plans an equity investment of up to $5 billion in Anthropic.
- Both companies are using Claude to optimize workloads for Instinct GPUs and the ROCm software.
- The deal builds on Anthropic’s existing use of the older Instinct MI355X.
What AMD and Anthropic Have Specifically Agreed Upon
Anthropic needs computing power—and plenty of it. The AI provider is scaling its infrastructure to meet the demand for its chatbot, Claude. This is exactly where AMD comes in: Anthropic will rely on AMD Helios Rackscale systems featuring the Instinct MI455X, the top-of-the-line model in the MI450 series. This will be complemented by AMD EPYC processors codenamed “Venice,” Pensando networking technology, and the company’s proprietary ROCm software.
AMD is therefore not supplying individual accelerators, but rather a complete rack system comprising GPUs, CPUs, networking, and software. Helios is designed for training and inference of large AI models. Incidentally, this collaboration isn’t new: Anthropic already uses AMD’s Instinct MI355X; the current deal represents a significantly larger next step.
| Component | Detail |
|---|---|
| GPU | AMD Instinct MI455X (MI450 series) |
| CPU | AMD EPYC “Venice” |
| Network | AMD Pensando |
| Software | AMD ROCm |
| Platform | AMD Helios (Rackscale) |
| Total Capacity | up to 2 gigawatts |
| Launch of first gigawatt | first half of 2027 |
Claude will be developed on AMD hardware in the future
The second part of the agreement is particularly exciting. AMD and Anthropic are launching a multi-year engineering collaboration in which Claude itself will help accelerate the development of AMD’s software. Specifically, the teams will use the AI model to optimize workloads for Instinct GPUs and accelerate the development of ROCm software. AMD, in turn, plans to deploy Claude extensively across its engineering and product teams.
This is more than just a nice side note. For years, ROCm has been considered AMD’s biggest weakness compared to NVIDIA’s CUDA ecosystem. If a leading AI model, of all things, helps close this software gap, it could give AMD a significant boost in the race for AI data centers. Qualcomm’s billion-dollar acquisition of Modular recently demonstrated that software is the real game-changer.
$5 billion: the investment with an asterisk
AMD has committed to investing up to $5 billion in Anthropic as a strategic equity investment in the future. The wording is important: “up to” and “in the future.” The investment is therefore a letter of intent and not yet a finalized deal. How much will actually be invested in the end—and under what terms—has not yet been publicly determined.
This pattern is now familiar in the industry: chip manufacturers secure data center customers not only through supply contracts but also through equity stakes. AMD had already demonstrated exactly this approach with OpenAI through a 6-gigawatt alliance.
Context: AMD is currently racking up AI deals one after another
The Anthropic deal is part of a whole series of announcements. Just two days earlier, AMD had announced that Microsoft is using AMD Helios on Azure for AI inference. Added to this is the aforementioned OpenAI partnership. AMD is thus gradually moving closer to the center of the global AI infrastructure expansion, which has so far been clearly dominated by Nvidia.
For Anthropic, the move also makes sense. Chief Compute Officer Tom Brown emphasizes that they run Claude on various hardware to distribute the appropriate workloads to the right hardware in each case. This diversification reduces dependence on a single supplier. At the same time, Anthropic continues to expand its product offerings centered around Claude, as seen most recently with the fixed-rate access to Claude Fable 5.
“This collaboration combines Anthropic’s leadership in frontier AI with AMD’s full strength in high-performance computing,” said AMD CEO Lisa Su.
FAQ
What have AMD and Anthropic agreed upon?
Anthropic will deploy up to 2 gigawatts of AMD Instinct MI450 GPUs in AMD Helios systems. In addition, AMD plans to invest up to $5 billion in Anthropic and collaborate on software optimization.
When will the deployment begin?
The first gigawatt of Instinct MI450 GPUs is scheduled to go live in the first half of 2027.
What is AMD Helios?
Helios is AMD’s rack-scale platform for AI data centers. It combines Instinct MI455X GPUs, EPYC “Venice” CPUs, Pensando networking technology, and ROCm software into a complete rack system.
Is AMD really investing $5 billion in Anthropic?
AMD has announced an investment of up to $5 billion. This is a letter of intent for the future, not a deal that has already been finalized.
Will Anthropic continue to use other hardware?
Yes. According to Anthropic, Claude deliberately runs on a broad range of hardware to appropriately allocate different workloads. The AMD deal is part of this diversification strategy.
Conclusion
For AMD, the Anthropic deal is another major coup—and the second one this week, following the Microsoft deal. Two gigawatts of Instinct MI450 plus a potential billion-dollar investment show that AMD is serious about catching up to Nvidia in the AI business. I find the software aspect of this almost more interesting than the sheer volume of units: If Claude actually helps make ROCm more competitive with CUDA, AMD is addressing the very issue that has stood in the company’s way for years. Caution is still warranted, however. Much of this is described as “up to” and “in the future”; the lion’s share of the rollout is scheduled for 2027 and beyond. The announcement is impressive, but the company still has to deliver.
