Project Rainier AWS Builds the World’s Most Powerful AI Computer

Project Rainier: AWS’s Ambitious AI Compute Cluster

If you’ve ever been in Seattle on a clear day, you might hear locals say, “the mountain is out,” referring to Mount Rainier. This 14,410-foot (4,392-meter) stratovolcano dominates the landscape, and its name now represents one of Amazon Web Services’ (AWS) biggest projects: Project Rainier, an AI compute cluster expected to be the world’s most powerful for training artificial intelligence models.

Project Rainier is a massive, multi-data center machine designed to push the boundaries of AI training. Announced late last year and already in progress, it stands apart from any previous AWS project in both scale and ambition.

A Mountain of Compute

The AI safety and research company Anthropic, an AWS customer, will use this cluster to develop future versions of its AI model, Claude. According to Gadi Hutt, director of product and customer engineering at Annapurna Labs (AWS’s chip design division), Rainier offers five times the compute power of Anthropic's largest existing training cluster. More compute means smarter and more accurate AI models, and this cluster delivers that at an unprecedented scale and speed.

Chips, Chips, Chips

At the core of Project Rainier is the “EC2 UltraCluster of Trainium2 UltraServers.” EC2 refers to Amazon Elastic Compute Cloud, AWS’s virtual server rental service. The standout is Trainium2, AWS’s custom chip designed specifically for AI training workloads. Unlike common chips in laptops or phones, Trainium2 handles massive data at incredible speeds—capable of trillions of calculations per second. To put that in perspective: counting to one trillion would take a person over 31,700 years, yet Trainium2 achieves that in milliseconds.

From Traditional to Ultra

Project Rainier doesn’t rely on a handful of chips. Instead, it leverages UltraServers—units combining four physical Trainium2 servers, each with 16 chips, linked via high-speed connections called NeuronLinks. These connections, marked by distinctive blue cables, act like express lanes for data, reducing latency and speeding up complex calculations across all 64 chips in an UltraServer.

Connecting tens of thousands of UltraServers forms the UltraCluster that is Project Rainier. This architecture dramatically cuts down communication delays common in traditional data centers, where servers communicate through slower network switches. That’s why this system is affectionately called a “friendly giant” by AWS engineers.

No Room for Failure

Communication happens at two levels: NeuronLinks within UltraServers and Elastic Fabric Adapter (EFA) technology between UltraServers and across data centers. EFA, visible by yellow cables, ensures fast, flexible connections that scale across multiple buildings.

Reliability is critical. AWS builds its own hardware, giving it full control over the technology stack—from chip design to software and data center infrastructure. This vertical integration allows rapid troubleshooting and optimization, ensuring the enormous compute capacity stays available.

Controlling the Stack

Having oversight of the entire stack is a strategic advantage. As Rami Sinno, Annapurna's engineering director, explains, knowing every level—from chip components to power delivery and software—enables AWS to optimize performance and innovate quickly. Sometimes improvements come from redesigning power systems or rewriting coordination software; often, it’s a combination of these efforts.

Sustainability at Scale

High performance doesn’t come at the cost of sustainability. AWS continuously improves energy efficiency in its data centers, focusing on rack layouts, electrical distribution, and cooling methods. In 2023, all electricity used by Amazon’s operations, including data centers, was matched by 100% renewable energy resources.

The company invests heavily in nuclear power, battery storage, and large-scale renewable projects globally. AWS has been the largest corporate purchaser of renewable energy worldwide for five years and aims to reach net-zero carbon by 2040.

New data center designs supporting Project Rainier include innovations in power, cooling, and hardware to reduce mechanical energy use by up to 46% and embodied carbon in concrete by 35%. Water stewardship is another focus: many data centers, like those in St. Joseph County, Indiana, minimize or eliminate cooling water use for most of the year by relying on outside air.

Thanks to these efforts, AWS leads the industry in water efficiency. A recent Lawrence Berkeley National Laboratory report shows AWS uses just 0.15 liters of water per kilowatt-hour in data centers—more than twice as efficient as the 0.375 liters per kilowatt-hour industry average and a 40% improvement since 2021.

The Future of AI

Project Rainier is more than just a technical feat. It sets a new standard for the computational resources available for AI, enabling models like Claude to become more sophisticated. Beyond that, this cluster opens doors for AI to address complex challenges in medicine, climate science, and more.

Much like Mount Rainier serves as a landmark in the Pacific Northwest, Project Rainier marks a turning point in computing—ushering in new capabilities chip by chip.

For those interested in expanding their expertise in AI and cloud technologies, exploring comprehensive training options can be a valuable step. Check out Complete AI Training for courses that cover the latest in AI development and deployment.