AMD and Oracle are ushering in the next era of artificial intelligence with an AI supercluster.

AMD announced a strategic partnership with Oracle to expand artificial intelligence (AI) infrastructure at Oracle AI World in Las Vegas on the 14th.

The collaboration will build the world's first public AI supercluster based on AMD's latest GPU, the Instinct MI450 series, on Oracle Cloud Infrastructure (OCI) starting in the third quarter of 2026.

Oracle plans to deploy 50,000 MI450 GPUs in its initial deployment, with expansion beyond 2027.

This supercluster provides a high-performance computing environment for ultra-large-scale AI model training and inference, and is based on AMD's 'Helios' rack design.

The design integrates the MI450 GPU, next-generation EPYC CPU 'Venice', and high-speed networking technology 'Vulcano' to maximize performance and energy efficiency.

The MI450 GPU offers up to 432GB of HBM4 memory and 20TB/s of memory bandwidth, enabling it to process 50% larger models in memory than its predecessor.

Also, a liquid-cooled rack with 72 GPUs packed tightly together. Delay is minimized and scalability is secured through high-speed interconnect technology based on the structure and UALink.

“Our customers are building the world’s most ambitious AI applications, and they require scalable, high-performance infrastructure,” said Mahesh Thiagarajan, senior vice president of Oracle Cloud Infrastructure. “By collaborating with AMD, we will provide customers with the best price-performance, security, and flexibility.”

“This collaboration is significant because it delivers an open, optimized, and security-centric system for the future of AI data centers,” said Forrest Norrod, vice president and general manager of AMD’s Data Center Solutions business.