AWS has collaborated with leading AI company Anthropic on Project Rainier, which delivers Trainium2 chips in record time. Anthropic is aiming to scale to more than one million chips by the end of 2025

Less than one year after Project Rainier was announced, AWS has deployed the project as Anthropic already runs workloads on its leading AI model, Claude. This is more than five times the compute power Anthropic used to train its previous AI models.

A groundbreaking project, Project Rainier represents a major milestone in AWS’s commitment to advancing its AI infrastructure on a significantly large scale.

The AI compute power is being used to build and deploy future versions of Claude. AWS says the more compute that is dedicated to training it, the smarter and more accurate it will become.

Project Rainier is already spread across multiple data centres in the US and is one of the largest projects AWS has ever attempted.

“Project Rainier is one of AWS’s most ambitious undertakings to date,” said Ron Diamant, an AWS distinguished engineer and head architect of Trainium. “It’s a massive, one-of-its-kind infrastructure project that will usher in the next generation of AI models.”

Trainium2 is being used to deliver the company’s ambitions. It is a custom-designed AWS AI chip built specifically for training AI systems. It is also specialised for processing the significantly large amounts of data needed to teach AI models how to complete a broad range of complex tasks in a fast manner.

As part of Project Rainier, AWS has built Trainium2 infrastructure that’s 70% larger than any other AI computing platform in AWS history, the company said. A single chip is capable of completing trillions of calculations a second.

And AWS is using thousands.

The solution is UltraServers – a new type of compute solution that combines four physical Trainium2 servers each with 16 Trainium2 chips. This is designed to reduce network latency, as data can move much faster within the system and accelerate complex calculations.

“When you connect tens of thousands of these UltraServers and point them all at the same problem, you get Project Rainier – a mega UltraCluster,” AWS explained.

Operating such a large compute cluster is complicated. To ensure uptime and capacity is available, reliability is very important. To combat this, AWS builds its own hardware so it can control every aspect of the technology stack. With this level of visibility, AWS said it’s able to optimise the stack further.

As a global leader in digital transformation, AWS is also investing heavily into nuclear power and battery storage – financing renewable energy projects to power its large operations in a more sustainable way. For the past five years, Amazon has been the largest corporate purchaser of renewable energy in the world and remains on its net zero by 2040 pathway.

When it comes to water, AWS is eager to use little to none. It cites its data centres in St. Joseph County, India – one of the Project Rainier sites – as maximising the use of outside air for cooling.

“From October to March the data centres won’t use any water for cooling at all, while on an average day from April to September they’ll only use cooling water for a few hours per day,” the company said.

It is hoped that Project Rainier will become a model for pushing AI further, enabling it to tackle broader and more complex global challenges.

All images credited to Amazon

RELATED STORIES

Nvidia and OpenAI sign $100bn deal to power 10GW of AI data centres

Nvidia, Microsoft and BlackRock-backed consortium to buy Aligned Data Centers in $40bn deal

Nscale raises $1.1bn funding to expand AI thanks to Nvidia, Dell & Nokia partnerships