AMD & The US Energy Department: Building AI Supercomputers

The push to secure AI computing infrastructure is impacting how governments acquire technology, with public-private partnerships increasingly replacing traditional procurement models. 

Now the US Department of Energy is putting US$1bn behind that shift, teaming up with Advanced Micro Devices (AMD) to build two supercomputers that will tackle everything from fusion energy to cancer treatment.

Energy Secretary Chris Wright and AMD CEO Lisa Su tell Reuters what they are aiming to achieve and how they are going to do it.

The aim of the partnership 

Instead of the government writing a cheque and owning the kit outright, the private sector provides the capital and equipment while both sides share access to the computing power. 

AMD, which competes with Intel and Nvidia in the processor market, will supply its latest AI chips for both systems.

Chris wants the machines to “supercharge” advances in nuclear power, fusion energy, defence technologies and drug development, he says. 

On fusion – the process that powers the sun by forcing light atoms together in superheated plasma – he acknowledges the challenge: “We’ve made great progress, but plasmas are unstable – and we need to recreate the centre of the sun on Earth,” Chris says.

“We’re going to get just massively faster progress using the computation from these AI systems that I believe will have practical pathways to harness fusion energy in the next two or three years.”

That is considerably more optimistic than many in the research community, where scientists often speak in terms of decades rather than years.

Chris also sees potential in drug discovery. 

The supercomputers will model cancer treatments at the molecular level, he says. 

“My hope is in the next five or eight years, we will turn most cancers, many of which today are ultimate death sentences, into manageable conditions.”

Lux and Discovery: What makes the two supercomputers unique

The first machine, Lux, will be running within six months. That is remarkably quick for a supercomputer of this scale.

It will use AMD’s MI355X chips, which handle the mathematical operations required for training and running AI models, along with the company’s central processing units and networking chips.

Lux is being built by AMD, Hewlett Packard Enterprise, Oracle Cloud Infrastructure and Oak Ridge National Laboratory in Tennessee. 

Lisa Su is impressed by the pace, since the deployment is the fastest rollout of a supercomputer this large that she has seen. 

“This is the speed and agility that we wanted to (do) this for the US AI efforts,” she says.

Stephen Streiffer, Director of Oak Ridge National Laboratory, says Lux will deliver roughly three times the AI processing capacity of current supercomputers at the facility. 

The second machine, Discovery, will be more powerful still. 

It will use AMD’s MI430 series chips, which are optimised for high-performance computing – the kind that combines thousands of processors to solve complex calculations.

Discovery arrives in 2028 and begins operations in 2029. 

Stephen expects substantial improvements in computational capability.

Lisa explains that the MI430 merges characteristics of traditional supercomputing processors with features needed to run AI applications. 

That matters because the two types of chips have historically been designed for different jobs.

The structure of the partnership is straightforward. 

The Department of Energy provides the buildings and infrastructure, while the companies supply the machines and cover capital costs. 

Both sides then share the computing time. 

A Department of Energy official says the two AMD-based supercomputers are the first in what will be a series of similar deals between private companies and government laboratories.