AMD chips to power the US “sovereign AI” supercomputers at Oak Ridge
AMD is partnering with the US Department of Energy to build two flagship sovereign AI supercomputers at Oak Ridge National Laboratory: Lux and Discovery. Backed by roughly $1 billion in public and private investment, these systems are intended to accelerate research in energy, medicine, national security and other scientific fields using homegrown infrastructure.
According to the announcement, Lux will be powered by AMD Instinct MI355X GPUs, AMD EPYC CPUs and AMD Pensando networking technologies, and is slated for deployment in 2026 as the first US “AI Factory” supercomputer. Discovery will follow and is planned to use next‑generation AMD EPYC “Venice” CPUs and MI430X accelerators (part of a new MI400 series designed for sovereign AI), with an operational target around 2029.
Key points
- Lux (2026): MI355X GPUs + EPYC CPUs + Pensando networking — positioned as the DoE’s early AI Factory capability.
- Discovery (2029): Next‑gen EPYC “Venice” CPUs + MI430X accelerators from the MI400 family for sustained high‑performance AI and scientific workloads.
- Funding: About $1 billion combined public/private investment to support development and deployment.
- Goals: Speed scientific discovery in energy, healthcare, and national security and strengthen domestic AI infrastructure and supply chains.
Why this matters
“Sovereign AI” emphasizes national control over the compute, data and workforce needed for advanced AI. Building major AI supercomputers with domestic hardware aims to reduce dependence on foreign supply chains, give researchers prioritized access to cutting‑edge infrastructure, and support mission‑critical scientific work. Oak Ridge — already home to some of the most powerful supercomputers in the world — will play a central role in this initiative.
What to watch next
- Progress on Lux’s 2026 deployment and performance benchmarks once it comes online.
- Technical details and benchmarks for the MI430X/MI400 series accelerators and EPYC “Venice” CPUs as Discovery moves toward 2029.
- How the DOE, AMD and partners (HPE) allocate access for researchers and agencies, and how these systems integrate with broader national AI strategies.
For more on Oak Ridge and its computing programs, see the Oak Ridge National Laboratory site and AMD’s product pages for EPYC and Instinct (vendor pages provide additional technical details).
Discussion: Do you think investing in sovereign AI supercomputers is the right approach to accelerate research and secure AI supply chains — or are there alternative strategies governments should prioritise?