Amazon unveils world’s largest AI compute cluster to challenge Nvidia’s dominance

AWS unveils Project Rainier, the world’s largest AI compute cluster, alongside new Ultraservers and expanded partnerships in AI chip innovation.

Photo: Amazon Web Services

Key Takeaways

• Amazon's Project Rainier will utilize the world's largest AI compute cluster with Trainium chips.
• Nvidia's dominance is being challenged as Amazon offers significant cost savings and competition.

<?xml encoding="UTF-8">

Amazon Web Services announced plans for “Ultracluster,” a massive AI supercomputer comprising hundreds of thousands of its in-house Trainium chips, to be operational by 2025.

In its annual AWS re:Invent conference being held today, AWS revealed Project Rainier, described as the world’s largest AI compute cluster, which will be used by AI startup Anthropic, in which Amazon recently invested $4 billion.

AWS’s Trainium chips are designed by Annapurna Labs, an Amazon-owned company based in Austin, Texas.

Acquired in 2015, Annapurna drives Amazon’s AI chip development, aiming to reduce reliance on Nvidia.

The company also introduced a new Ultraserver featuring 64 interconnected Trainium chips during the event.

These servers, which leverage Amazon’s proprietary NeuronLink technology, can reach 20.8 petaflops of compute power per server.

AWS also announced Apple as one of its newest chip customers.

Apple’s senior director of machine learning and AI, Benoit Dupin, noted that they are testing Trainium2 chips and anticipate cost savings of approximately 50%.

The AI semiconductor market is valued at $117.5 billion in 2024 and is expected to reach $193.3 billion by 2027, according to a report by Wall Street Journal.

Nvidia currently holds around 95% of the market share, but Amazon’s push to develop its own chips—designed by Annapurna Labs and fabricated through Taiwan Semiconductor Manufacturing Co.—is aimed at challenging this dominance.

Disclaimer