Oracle Corporation has announced a revolutionary partnership, which is soon to completely reshape the future of artificial intelligence. The technology giant is soon to harness a record number of AMD GPUs, which is the largest single commitment to anything other than NVIDIA technology ever to take place within the realms of cloud computing. This is soon to signal a massive shift regarding the allocation of AI workloads.
Oracle is the first hyperscale cloud to provide massive AMD GPU resources
Oracle Cloud Infrastructure is set to become the first hyperscale cloud to provide a publicly accessible AI supercomputer supported by 50,000 AMD Instinct MI450 Series GPUs. These will go live starting calendar Q3 2026 and are set to continue into 2027 and beyond due to growing demand. This is the biggest commitment to AMD infrastructure by a major cloud vendor and is a first in industry history.
The new partnership is based on an existing partnership between Oracle and AMD, which started with the AMD Instinct MI300X service offered in 2024. Oracle has already deployed AMD Instinct MI355X GPUs to the zettascale OCI Supercluster, which can support up to 131,072 GPUs. This will enable seamless integration and optimal performance when the Supercluster is deployed with MI450.
The AMD MI450 Series GPUs support up to 432 GB of HBM4 memory and 20 TB/s of memory bandwidth. According to AMD, the devices enable customers to perform machine learning tasks, including model training and inference, on models 50 percent larger than the previously supported models, completely within the memory. These devices enhance machine learning workloads due to their efficient architecture.
Cloud providers broaden chip sources due to increasing AI demand
This is contrary to the growing trend among cloud companies to source their chip supplies from a variety of manufacturers due to increased demand for AI computing power and the dominance of NVIDIA. As it is, NVIDIA has more than 90% market share when it comes to AI accelerators. With the massive commitment by Oracle to AMD, customers are guaranteed alternative computing platforms to support their AI applications.
“We think customers are going to take to AMD technology very, very well, particularly in the inferencing domain,” said Karan Batta, senior VP of Oracle Cloud Infrastructure.
Steven Dickens, CEO of HyperFrame Research, also highlighted the relevance of market diversification to the future of the industry. This is coming on the heels of OpenAI’s move to integrate AMD chips into its data centre, which is a sign of growing industry support and validation of AMD’s role as a credible alternative to Nvidia’s industry-dominating products.
Oracle positions itself by achieving massive AI compute work
Oracle has immense performance requirements and backlogs when it comes to AI compute, and such an alliance between Oracle and AMD is extremely important to cater to increasing demands. The company announced during its AI World conference, which took place in Las Vegas, that demonstrate the latest technology advancements in the arena of AI. Oracle will start the provision of cloud services supported by new AMD processors starting the second half of next year.
AMD “Helios” rack designs also support 72 GPU configurations along with liquid-cooled systems to provide optimal performance and power efficiency. AMD Pensando “Vulcano” AI-NICs, with speeds of up to 800 Gbps, can be stacked on each GPU to support high-speed connectivity. With such architecture, GPUs are able to communicate and share resources effectively, reducing latency and maximizing GPU throughput due to hardware coherence.
Oracle’s pledge to utilize 50,000 AMD GPUs marks a turning point in the competition and diversification of AI infrastructure. Such a huge investment not only gives customers strong competing choices against NVIDIA-led initiatives but also solves Oracle’s considerable AI compute backlog. The tie-up of the two companies places them aptly on the next frontier of AI infrastructure developments.
