Amazon unveils plans to build AI supercomputer

Developed in collaboration with Anthropic, an AI startup, the new supercomputer, codenamed Project Rainer, will leverage AWS’s latest Trainium2 AI chips to drive advancements in AI model training and scalability.

The supercomputer, expected to be five times larger than the cluster that trained Anthropic’s most advanced AI models, will be housed in a single data centre. It will feature hundreds of thousands of Amazon's Trainium2 chips, connected via the newly unveiled ‘Ultraserver’ architecture. Once completed, Amazon anticipates that the Ultracluster will set a new benchmark as the largest AI machine in the world.

Matt Garman, CEO of Amazon Web Services, introduced Project Rainer during his keynote, stating: “Trainium2 delivers 30 to 40% better price performance compared to current GPU-powered instances.”

Anthropic: the first Ultracluster customer

Anthropic, backed by an $4 billion investment from Amazon, will be the first customer for the Ultracluster. The AI firm has been at the forefront of exploring AI’s potential and will leverage the new infrastructure to push the boundaries of AI innovation further.

Additionally, prominent names such as Apple, Adobe, and Databricks are already using Amazon's Trainium2 chips, further solidifying Amazon’s growing influence in the AI chip market.

Trainium2 and beyond: Amazon’s strategic vision in AI

Amazon also announced the general availability of its Trainium2 AI chips, alongside plans to launch an even more advanced Trainium3 chip in late 2025. Designed as a lower-cost alternative to GPU-based solutions, Trainium2 reinforces Amazon's dual role as both a customer and competitor to Nvidia, the market leader in AI GPUs.

The new Trainium-powered Ultraservers, which connect 64 Trainium chips into a single architecture, are pivotal to AWS’s AI strategy. These innovations align with Amazon’s broader goal of providing enterprises with accessible and scalable AI solutions through its Cloud platform.

As the largest Cloud provider by market share, Amazon is intensifying competition in the AI chip market, challenging Nvidia's dominance. While AWS continues to rely on Nvidia GPUs to meet customer demand, the Trainium2 chip positions Amazon to offer more cost-efficient alternatives tailored to enterprise needs.