The US Department of Energy and AMD are building two supercomputers at Oak Ridge National Laboratory, creating a staged infrastructure plan rather than waiting for one perfect future machine. Lux AI is scheduled for early 2026; Discovery follows in 2028 as the laboratory’s next flagship system.
Together, the projects represent approximately $1 billion in public and private investment. They will join a secure national computing network supporting research in science, energy, medicine, and national security through standards-based infrastructure built in the United States.
The partnership also offers enterprises a useful model: combine specialised vendors around a defined mission, protect data between collaborators, and plan the next computing generation before the current one arrives. US Energy Secretary Chris Wright describes that coordination as a way to bring government, science, and industry into one national capability.
Lux Is Designed as a Scientific AI Factory
Lux will bring together Oak Ridge, AMD, Oracle Cloud Infrastructure, and Hewlett Packard Enterprise. AMD Instinct MI355X GPUs, EPYC processors, and Pensando networking will support data-intensive models used in areas such as materials, energy systems, and medicine.
The system is designed so multiple groups can work together while keeping their information secure and separated. That balance mirrors a challenge faced by large companies whose research partners or divisions need shared computing without unrestricted access to one another’s data. Security is therefore part of the collaboration model, not a perimeter added after construction.
Calling Lux an “AI Factory” emphasises a complete research environment rather than a collection of accelerators. The objective is to train and deploy advanced models, turning enormous scientific datasets into tools that shorten the route from a question to an experiment or practical result. Oak Ridge director Stephen Streiffer expects the combination of high-performance computing and AI to accelerate that route.
Discovery Targets More Work Without More Power
Discovery will use AMD’s forthcoming Venice EPYC processors and MI430X GPUs. Its “Bandwidth Everywhere” design is intended to increase memory and network performance without raising power consumption, enabling larger datasets and more complex models while containing energy cost.
That constraint is strategically important. As AI systems scale, electricity, cooling, and data movement can limit useful capacity before raw processor counts do. An architecture that feeds computing units efficiently may deliver more research for the same energy envelope.
Discovery builds on Frontier, the world’s first exascale computer, with an emphasis on moving existing applications to the new platform. Continuity protects the software and expertise accumulated around earlier investment rather than requiring every scientific team to begin again. Compatibility turns the new machine into an extension of a research programme rather than an isolated replacement.
Partnership Turns Infrastructure Into Capability
AMD supplies processors and accelerators, HPE contributes supercomputing expertise, Oracle supports sovereign cloud infrastructure, and Oak Ridge brings the research environment. No participant alone represents the complete system.
Once operational, Lux and Discovery will support models aimed at batteries, reactors, semiconductors, critical materials, biology, energy, and defence. Their value will be measured by the discoveries and development cycles they accelerate, not only benchmark position.
Private organisations can apply the principle at smaller scale. Infrastructure strategy should connect capacity, governance, energy, software portability, and partner responsibilities to a specific research or business objective. Shared systems need rules that make collaboration possible without dissolving data boundaries. They also need a roadmap for applications, because idle theoretical performance creates no scientific or commercial advantage.
Lux and Discovery show that long-term AI advantage is constructed across generations. One system delivers near-term capability while the next incorporates lessons about bandwidth, power, and existing applications. The hardware matters, but the strategic asset is the coordinated environment around it—designed to keep scientific work moving as computing technology changes.
