America's first AI factory: Lux supercomputer hits 220 petaflops for science

The US Department of Energy just spun up Lux, a 220-petaflop AI supercomputer built by AMD and HPE to run the nation's Genesis Mission.

Research ยท Source: AMD Newsroom

What happened

The US Department of Energy just powered up Lux at the Oak Ridge National Laboratory. It stands as the first dedicated AI factory for the Genesis Mission. HPE and AMD co-designed the system to push boundaries in AI-driven science, energy research, and national security. The government wants to transform American innovation through raw AI power.

Lux hit the global TOP500 list debuting at number fourteen. It delivers a massive 220.79 petaflops of measured performance. Under the hood, the system runs on AMD EPYC processors and AMD Instinct MI355X GPUs. It also utilizes AMD Pensando networking and ROCm open-source software to tie the massive compute together.

This is not a traditional legacy supercomputer. HPE designed Lux to offer cloud-like, multi-tenant access for scientific users. Researchers across the country can now scale their AI training and inferencing workloads seamlessly. It marks the first time HPE Networking solutions have been deployed at a Department of Energy national laboratory.

Key facts

Why it matters

The government is finally building infrastructure purpose-built for AI rather than just traditional modeling and simulation. Lux shifts the national focus directly toward AI-driven discovery. Builders in the scientific and energy sectors now have a massive environment to train models that require government-scale compute. This public-private partnership proves the US is serious about maintaining technological leadership on the global stage.

AMD is quietly taking over the global supercomputing market. They now power over forty percent of the TOP500 list and half of all new entries. Their chips run four of the ten fastest supercomputers in the world, including El Capitan and Frontier. By pairing their latest MI355X GPUs with HPE networking, they are proving they can deliver end-to-end AI factories. This deployment validates their hardware and software stack at the highest possible level of enterprise demand.

For builders

Cloud-like access for national researchers

HPE built Lux with flexible multi-tenant capabilities. If you are building AI models for US science or energy projects, you get scalable access to train and infer without managing bare metal infrastructure.

AMD hardware proves enterprise readiness

Lux runs entirely on the AMD stack from EPYC CPUs to MI355X GPUs and ROCm software. Founders building compute-heavy startups should watch how ROCm handles these massive government workloads as a viable alternative to Nvidia.

New networking standards for AI data centers

This is the first DOE lab deployment of HPE Networking solutions. Engineers designing AI clusters can learn from how HPE and AMD structured this network to handle extreme AI training loads.

My take

The government is finally treating AI compute like critical national infrastructure. AMD is proving they can power the biggest AI factories on earth. If their ROCm software can run the Genesis Mission smoothly, the moat around Nvidia is going to shrink faster than people think.

Original reporting: AMD Newsroom. This is my rewrite and opinion.

More AI news for builders