AI Pulse by Inblix

AMD bets $5B on Anthropic with massive 2-gigawatt GPU deal for 2027

The Verge AI · Jul 22, 2026 · 2 min read · Read original article →

Curated by the Inblix editorial team


Featured image for article: AMD bets $5B on Anthropic with massive 2-gigawatt GPU deal for 2027

Forget the cloud credit scraps Big Tech usually throws around. AMD just put its chips — literally — on the table with a deal to invest up to $5 billion in Anthropic. But the money is arguably the less interesting part. The real headline is scale: Anthropic plans to deploy up to 2 gigawatts of AMD’s Instinct MI450 GPUs, using the chipmaker’s new Helios rack-scale system. To put that in perspective, a single gigawatt can power roughly 750,000 homes. This isn’t a pilot program; it’s a foundational bet on the hardware that will train the next generation of Claude.

The first gigawatt of that capacity is targeted for the first half of 2027. That timeline signals just how desperate the AI industry is for compute, and how long it takes to actually build it out. Anthropic has been on a data center signing spree lately, inking deals with SpaceX and TeraWulf, and it already has deep infrastructure ties with Google, Broadcom, and Amazon. This AMD partnership, however, is different. It gives Anthropic a direct line to a GPU architecture that isn’t Nvidia’s. In a market suffocating under Nvidia’s supply constraints, that’s not just diversification — it’s a lifeline.

The deal is a two-way street. While Anthropic gets the compute, AMD gets a flagship customer to prove its silicon is more than a budget alternative. The companies are launching a multi-year engineering collaboration where AMD will use Claude across its own software and product development. Tom Brown, Anthropic’s cofounder and chief compute officer, framed it bluntly: “By partnering with AMD across the stack, we are securing the capacity we need and optimizing it for training and serving Claude.” The subtext: Nvidia can’t meet all the demand, and AMD is finally ready to be a credible alternative at the very highest end.

Skepticism is warranted. AMD has promised data center dominance before, only to watch Nvidia tighten its grip. But the raw numbers here — $5 billion, 2 gigawatts — force a recalculation. This isn’t a press release about a new chip. It’s a material shift in the supply chain for frontier AI models. If AMD can deliver on that 2027 timeline without the usual driver headaches and ecosystem gaps, the CUDA moat might finally start to look a little shallower. The question isn’t whether Anthropic needs the compute. It’s whether AMD can actually build it.

💡 Key Takeaways

  1. The 2-gigawatt deployment target is a historic scale for AI inference and training hardware, dwarfing typical cloud GPU contracts and signaling an industry-wide compute crisis.
  2. Anthropic is aggressively building a multi-vendor hardware strategy to break its dependence on Nvidia, with AMD's Instinct MI450 and rack-scale Helios system now central to that plan.
  3. AMD isn't just selling chips; it's locking in a marquee AI customer to co-develop its software stack, using Claude internally to close the ecosystem gap with Nvidia's CUDA.

Keep reading: See related articles below for more coverage on this topic.

Get smarter about AI

The sharpest AI news, curated daily. Delivered free to your inbox.

Learn more

Glossary terms

← Back to all articles