AMD is powering Lux and Discovery, which will serve as the country’s first AI factory for science and one of its flagship systems for national research.
The Lux system will roll out in 2026. It is being developed in collaboration with Oak Ridge, AMD, Oracle Cloud Infrastructure and the usual grey box-pushers at HPE. It will use AMD’s Instinct MI355X GPUs, EPYC processors and Pensando networking gear, which the chipmaker says is tailored for model-heavy, data-hungry AI workloads.
Lux is being positioned as the US Department of Energy’s answer to high-end AI computing needs. It's expected to support research into materials, manufacturing, energy and health, all while waving the stars and stripes for "sovereign AI".
AMD chair and CEO Lisa Su said: “We are proud and honoured to partner with the Department of Energy and Oak Ridge National Laboratory to accelerate America’s foundation for science and innovation.” She added that the systems will help push forward priorities in medicine, science and energy.
Discovery is the bigger play. It is slated to arrive in 2028 with operations starting the following year. It's built using AMD’s new “Venice” EPYC CPUs and the MI430X GPU, a fresh entry in its MI400 accelerator series. This silicon is pitched at keeping the US competitive while locking its AI efforts inside its own borders.
Energy Secretary Chris Wright said: “Winning the AI race requires new and creative partnerships that will bring together the brightest minds and industries American technology and science has to offer.” He described Lux as a shift towards a “common-sense approach” to national computing, with Discovery to follow through a competitive procurement process.
The price tag for both systems will hit around $1 billion, with the funds coming from public coffers and AMD’s own wallet. It's a joint effort to build up the sort of infrastructure that lets the US say it’s serious about AI.
ORNL director Stephen Streiffer claimed the new gear would speed up innovation, saying: “The Discovery system will drive scientific innovation faster and farther than ever before.”
He added that Oak Ridge’s history in supercomputing had already slashed the time it takes to solve real-world problems.
HPE president and CEO Antonio Neri added: “HPE’s newest supercomputing solutions harness converged AI and HPC architectures, enabling the lab to achieve unprecedented productivity and scale.”
Mahesh Thiagarajan, executive vice president of Oracle Cloud Infrastructure said: “Oracle will deliver sovereign, high-performance AI infrastructure that will support the co-development of the Lux AI cluster.”
Discovery builds on the exascale work done by Frontier, and will use HPE’s Cray Supercomputing GX5000 platform. It will let researchers work on batteries, catalysts, next-gen reactors, new materials and all the other tech the US would like to say it invented first.
The system will have more memory, faster networking and higher throughput than first-gen exascale machines. Power draw will remain roughly the same, which is supposed to make it look green while producing more compute.
It runs on open-source software and standards. The idea is to make sure that American researchers are not relying on closed or foreign-built infrastructure when it comes to AI.
