Oracle to use 50,000 AMD chips in supercluster and unveils 'largest' supercomputer with OpenAI

Deal seen to help AMD catch up with Nvidia in AI race

Oracle chief executive Mike Sicilia addresses Oracle AI World in Las Vegas. Alvin R Cabral / The National

Oracle has teamed up with chip maker AMD for a new artificial intelligence supercluster service powered by tens of thousands of graphics processing units, and unveiled what it says is the world's "largest" AI supercomputer in the cloud.

The Texas-based company's Oracle Cloud Infrastructure platform will be a launch partner for an AI supercluster run by AMD MI450 GPUs, Oracle said on the sidelines of its rebranded AI World conference in Las Vegas.

The partnership will involve initial use of 50,000 GPUs in the third quarter of 2026 and will expand from there, Oracle said.

GPUs perform technical calculations faster and with greater energy efficiency than traditional central processing units, resulting in leading performance for AI training and other applications that require accelerated computing.

They were initially created to handle 3D graphics, most notably in games, but have evolved to become one of the most important types of computing technology.

With the AMD chips, OCI users would "gain powerful new capabilities for training, fine-tuning and deploying the next generation of AI", said Forrest Norrod, an executive vice president and general manager at AMD.

The deal is another win for California-based AMD, after last week's multi-year deal with OpenAI, in which it will supply chips to the generative AI leader and give the creator of ChatGPT the option to buy about 10 per cent in AMD.

It is, however, still lagging behind the offerings of Nvidia, the world's most valuable company, which offers fully integrated systems that combine GPUs and central processing units, which have become key in the AI arms race.

Meanwhile, Oracle also said that it has linked up with California-based OpenAI to launch Zettascale10, an AI supercomputer that connects hundreds of thousands of Nvidia GPUs.

OCI Zettascale10 was built with "extremely low" latency, competitive price performance and "reliability" required for large-scale AI workloads, the companies said.

In its first phase, OCI Zettascale10 aims to use up to 800,000 Nvidia GPUs.

OCI Zettascale10 is also the main component of a supercluster built in collaboration with OpenAI in Abilene, Texas, as part of the Stargate mega data centre.

"Customers can build, train and deploy their largest AI models into production using less power per unit of performance and achieving high reliability," said Mahesh Thiagarajan, executive vice president of OCI.

Shares of AMD jumped mo re than 3.4 per cent early on Tuesday. Oracle, meanwhile, was down 1.8 per cent.