AMD and U.S. Department of Energy team up on Lux and Discovery AI supercomputers
AMD has signed a major agreement with the U.S. Department of Energy to deliver two new AI-optimized supercomputers-Lux and Discovery-for Oak Ridge National Laboratory (ORNL). The combined program represents roughly $1 billion in public and private funding and is focused on sovereign, U.S.-based AI and scientific research. Discovery is planned to come online in early 2026 at ORNL.
For federal leaders, this is about capacity, speed, and security: more compute for mission-driven research, faster time-to-insight, and tighter control over sensitive data.
What's being built
Lux: Co-developed by AMD, ORNL, Oracle Cloud Infrastructure (OCI), and HPE. It uses AMD Instinct MI355X AI GPUs, AMD EPYC CPUs, and AMD Pensando networking to support a secure, standards-based, federated AI stack across on-prem and cloud environments.
Discovery: A deeper collaboration between DOE, ORNL, HPE, and AMD. It introduces next-gen AMD "Venice" GPUs and AMD Instinct MI430X accelerators (MI400 series) built for sovereign AI and scientific computing. DOE's goal is for Discovery to set a new performance bar, with targets several times beyond Frontier, ORNL's current flagship system.
Why this matters for government
- Sovereign AI: U.S.-based infrastructure for training and inference, with data locality and standards-based controls.
- National security: Better footing for AI model development in areas like energy resilience, biosecurity, space, and defense-adjacent research.
- Research throughput: Larger, more complex workloads-climate models, materials discovery, fusion, healthcare-move from proposal to results faster.
- Interagency access: A federated approach simplifies shared capacity and data-sharing frameworks across civilian and defense agencies.
- Domestic supply chain: Coordination across AMD, HPE, and OCI strengthens U.S.-centric compute stacks and procurement options.
Timeline and procurement signals
Discovery is slated for early 2026 availability at ORNL. Lux is part of the same program and will support the broader initiative as deployments ramp.
Expect a mix of on-prem supercomputing and cloud-adjacent capacity through OCI. Agencies should engage early with DOE/ORNL for allocations, data governance agreements, and workload onboarding. Budget windows spanning FY2025-FY2027 will be the critical period for planning, staffing, and interagency coordination.
Technical snapshot
- AMD Instinct MI355X AI GPUs with AMD EPYC CPUs
- AMD Pensando networking for data-intensive, low-latency fabrics
- Next-gen AMD "Venice" GPUs and AMD Instinct MI430X (MI400 series) for Discovery
- HPE systems that blend AI and HPC architectures for scale
- OCI integration to extend capacity with sovereign cloud adjacency
- Secure, federated, standards-based infrastructure across sites
Security and governance
The program prioritizes secure, U.S.-hosted AI infrastructure. Agencies should map data classification to facility controls, plan for export control compliance, define model governance (training data lineage, evaluation, and monitoring), and align with zero-trust principles for multi-tenant research environments.
What leaders are saying
AMD leadership emphasized the mission: advancing U.S. priorities in science, energy, and medicine through a strong public-private collaboration. DOE underscored that new partnerships are helping bring additional capacity online faster, turning shared innovation into national strength. ORNL highlighted the expected gains in moving researchers from problem to solution. HPE pointed to converged AI/HPC architectures that increase productivity at scale. Oracle noted its role in delivering sovereign, high-performance AI infrastructure for the Lux cluster.
Action items for agencies
- Nominate workloads suited for large-scale AI/HPC (e.g., modeling, simulation, classification, generative modeling for science).
- Engage ORNL/DOE program offices on access models, data residency, compliance, and interagency sharing.
- Prepare staffing and training plans for AI engineering, MLOps, and HPC scheduling.
- Coordinate with CIO/CISO on accreditation, data boundaries, and incident response.
- Review budget posture for compute, storage, and personnel across FY2025-FY2027.
The bottom line
Lux and Discovery give the federal community a clearer path to train larger models, accelerate research, and keep sensitive work on U.S. soil. Start the coordination now-workloads, data policies, and people-so your programs are ready to use the new capacity as it comes online.
Your membership also unlocks:
