Amazon Web Services, the cloud computing division of Amazon.com, revealed plans on Tuesday to integrate key Nvidia technology into upcoming versions of its artificial intelligence chips, in a renewed push to attract major AI clients.
At AWS’ flagship cloud conference in Las Vegas, attended by around 60,000 people, the company announced it will incorporate Nvidia’s “NVLink Fusion” technology in a forthcoming chip, dubbed Trainium4. While AWS did not provide a timeline for Trainium4’s launch, NVLink Fusion is renowned for enabling rapid connections between disparate chip types, an asset long considered a central innovation by Nvidia.
NVLink Fusion will allow AWS to develop larger-scale AI servers capable of high-speed mutual recognition and data exchange, a vital capability as training large language models increasingly requires interlinking thousands of machines. The collaboration will also provide AWS customers with what the company is branding as “AI Factories”—dedicated high-performance AI infrastructure inside clients’ private data centers. This setup promises greater speed and readiness for machine learning projects, according to AWS.
Separately at the event, AWS debuted its Trainium3 UltraServer, featuring the latest-generation 3-nanometer Trainium3 chip and proprietary networking solutions. Citing internal figures, AWS claims the third-generation platform delivers a fourfold improvement in speed and memory—both for AI training workloads and high-demand application deployments—compared with its predecessor.
The Trainium3 UltraServer reportedly supports up to 144 chips per server, and scale-ups allow networks of thousands of UltraServers to deliver up to 1 million Trainium3 chips to support a single application, ten times the previous generation’s capacity.
In addition to performance gains, AWS highlighted efficiency improvements, asserting that the new chips and systems are 40% more energy efficient than the earlier versions. As global demand for data centers soars—driven by AI development and their significant electricity needs—AWS said it is focusing on lowering power usage per server even as capacity grows, a statement attributed to AWS.
