DOE launches $1B supercomputer and AI partnership with AMD - what public sector leaders need to know
The U.S. Department of Energy (DOE) has entered a $1 billion partnership with AMD to build two new AI-enabled supercomputers aimed at nuclear energy, national security, and medical research. Energy Secretary Chris Wright said the systems will speed progress on fusion energy, strengthen oversight of the nuclear stockpile, and accelerate drug discovery.
DOE will host both systems. The companies will provide the machines and capital investment, with both sides sharing compute capacity. Additional public-private projects of this type are planned across DOE labs.
What's being built
System 1 - Lux - is scheduled to come online in about six months. It centers on AMD's MI355X AI accelerators and includes AMD CPUs and networking, co-developed with Hewlett Packard Enterprise, Oracle Cloud Infrastructure, and Oak Ridge National Laboratory (ORNL). ORNL Director Stephen Streiffer expects Lux to deliver roughly three times the AI capacity of current supercomputers.
System 2 - Discovery - will use AMD's MI430 series tuned for high-performance computing and AI. Designed by ORNL, HPE, and AMD, Discovery is slated for delivery in 2028 and operations in 2029. AMD CEO Lisa Su said Lux's rollout reflects the speed and agility the U.S. is aiming for in AI infrastructure.
Why it matters for government missions
- Fusion and clean energy: Researchers are modeling unstable plasmas at extreme conditions to move closer to practical fusion. Wright said AI-driven compute could lead to practical pathways to fusion energy within two or three years.
- National security: Advanced simulations support stewardship of the U.S. nuclear arsenal without live testing, improving confidence, safety, and lifecycle planning.
- Health and biotech: Molecular-level simulations can shorten the time to identify potential cancer treatments. Wright said the aim is to turn most cancers into manageable conditions within five to eight years.
How the partnership is structured
DOE will host the systems at national labs. AMD and partners supply hardware and capital, with compute time shared between the government and industry contributors. According to DOE, this is a template for additional collaborations across the national lab network.
Access and coordination
Agencies, lab programs, and qualified consortia are expected to compete for allocations through DOE processes. Plan early for data pathways, workload readiness, and security approvals, especially for controlled or health-related data.
- Workload mapping: Identify simulations, AI training, and analysis jobs that exceed current capacity (e.g., multiphysics models, large-scale graph analytics, multimodal AI).
- Data governance: Define classification, residency, and sharing rules. Establish audit trails, model cards, and provenance for scientific and policy review.
- Security: Prepare ATO packages, zero-trust access, SBOM requirements, and supply-chain review for software stacks.
- Interagency alignment: Coordinate with DOE, NNSA, NIH, and DoD where missions overlap to avoid duplicate spend and to share datasets and models responsibly.
- People and skills: Budget for HPC/AI engineers, domain scientists, and program managers who can run at supercomputing scale.
Timeline signals for planning
- Lux: target availability in ~6 months. Suitable for pilots, rapid experiments, and high-priority missions that need near-term scale.
- Discovery: delivery in 2028, operations in 2029. Plan long-horizon programs, multi-year datasets, and method development that benefits from a step-change in compute.
Technical notes for decision makers
- Architecture: AMD MI355X and MI430 accelerators with AMD CPUs and high-speed interconnects. Designed for both traditional HPC and AI workloads.
- Throughput: ORNL projects Lux at ~3x the AI capacity of current systems; Discovery is expected to exceed that mark substantially.
- Ecosystem: Co-development with HPE and Oracle Cloud Infrastructure points to hybrid workflows, cloud bursting, and shared tooling.
- Facilities: Expect advanced cooling and power profiles. Coordinate with facilities for energy planning and sustainability reporting.
Risks and guardrails
- Dual-use oversight: Put governance in place for sensitive models (e.g., nuclear, bio) and enforce export-control compliance.
- Reproducibility: Require documented configurations, datasets, and evaluation protocols for mission credibility and oversight.
- Data protection: Enforce encryption, role-based access, and monitoring for workload isolation across shared resources.
- Cost control: Track utilization against mission outcomes; use allocation dashboards and chargeback models to prevent waste.
Where to learn more
For program context on fusion research, see the DOE's Fusion Energy Sciences program at science.osti.gov/fes. For lab access and user programs, visit Oak Ridge National Laboratory.
If your team is planning workforce upskilling for AI operations and program management, you can review role-based options here: Complete AI Training - Courses by Job.
Bottom line
Lux and Discovery signal a step up in national compute capacity for science and security. Agencies that scope the right workloads, secure the data pathways, and staff for scale will be first in line to convert this infrastructure into measurable mission outcomes.
Your membership also unlocks:
