AMD Fuels U.S. Sovereign AI Supercomputers for Open American AI

A major leap forward for American artificial intelligence capabilities is taking shape at Oak Ridge National Laboratory, where two new supercomputers, Lux and Discovery, are set to establish a powerful sovereign AI infrastructure. Powered by AMD technology, these systems represent a strategic investment exceeding one billion dollars, directly supporting the U.S. AI Action Plan. Their mission is to accelerate scientific discovery, bolster national competitiveness, and ensure that critical AI research and development remains securely on American soil.

The collaboration between AMD, the U.S. Department of Energy (DOE), and Oak Ridge National Laboratory (ORNL) is central to this initiative. Both Lux and Discovery are designed as flagship supercomputers to drive transformative progress in vital areas including energy, medicine, and national security. This public-private partnership aims to provide national researchers and agencies with the advanced computing power needed to tackle the country’s most pressing challenges.

Dr. Lisa Su, chair and CEO of AMD, emphasized the significance of this collaboration. She stated that the company is proud to partner with the DOE and ORNL to strengthen America’s foundation for science and innovation. She noted that Discovery and Lux will utilize AMD’s high-performance and AI computing technologies to advance the nation’s most critical research priorities, showcasing the powerful results achievable through public-private cooperation.

Scheduled for deployment in early 2026, the Lux AI supercomputer is a joint development by ORNL, AMD, Oracle Cloud Infrastructure, and HPE. It will be the nation’s first dedicated AI factory for science. The system will be equipped with AMD Instinct MI355X GPUs, AMD EPYC CPUs, and AMD Pensando networking technologies. Lux is specifically designed to expand the DOE’s leadership in AI and accelerate innovation in fields like materials science, medicine, and advanced manufacturing.

The Discovery supercomputer further deepens the collaboration between the DOE, ORNL, HPE, and AMD. At its core are next-generation AMD EPYC CPUs, codenamed “Venice,” and new AMD Instinct MI430X GPUs from the MI400 Series. These accelerators are engineered specifically for sovereign AI and scientific computing, enabling the U.S. to train and deploy sophisticated AI models on domestically built systems that protect national data and maintain scientific competitiveness.

U.S. Secretary of Energy Chris Wright highlighted the importance of such innovative partnerships for winning the global AI race. He explained that the administration’s approach, exemplified by the Lux project and the competitively procured Discovery system, is about bringing new capacity online rapidly. By working with industry leaders like AMD and HPE, shared innovation is being transformed into tangible national strength.

ORNL Director Stephen Streiffer pointed to the transformative potential of the Discovery system. He stated that ORNL’s supercomputing leadership has already dramatically shortened the time from a research problem to its solution. With Discovery, the deep integration of high-performance computing and AI promises to deliver breakthroughs at the accelerated pace and massive scale required for the United States to maintain its leadership in an intensely competitive global environment.

Antonio Neri, president and CEO of HPE, expressed pride that the partnership with AMD is at the forefront of advancing supercomputing for ORNL. He explained that HPE’s newest solutions harness converged AI and HPC architectures, enabling the lab to achieve unprecedented productivity and scale. This effort, he noted, reinforces American leadership in applying AI to critical domains like science, energy, and national security.

Mahesh Thiagarajan, executive vice president of Oracle Cloud Infrastructure, also commented, stating that Oracle is honored to work with the Department of Energy. The company will deliver the sovereign, high-performance AI infrastructure required to support the co-development of the Lux AI cluster.

The Lux AI supercomputer is purpose-built as the nation’s first AI factory for science, energy, and national security. Its advanced architecture is optimized for data-intensive workloads, allowing it to train, fine-tune, and deploy foundational AI models that will accelerate discovery and engineering innovation. As a critical resource, Lux is designed to give the United States an early and decisive advantage in deploying frontier AI capabilities, ensuring that the breakthroughs of the AI era are discovered and developed within America.

Lux is expected to unleash new scientific capabilities aligned with the DOE’s mission, from revealing nature’s hidden structures and discovering breakthrough materials to accelerating energy innovation and advancing national security resilience.

The Discovery supercomputer, purpose-built for AI and scientific computing, features a “Bandwidth Everywhere” design. This architecture allows complex science and AI applications to run with high efficiency, delivering productive results. Building upon the foundation laid by the Frontier supercomputer, Discovery will use HPE’s next-generation Cray Supercomputing GX5000 platform. It maintains a consistent programming environment to ensure a seamless transition for researchers and users.

Arriving in 2028, Discovery will serve as the DOE’s next flagship supercomputer at ORNL, extending U.S. leadership in high-performance computing and AI. It reflects a strong public-private commitment to keeping the United States at the forefront of global innovation. Building on the exascale breakthroughs of Frontier, Discovery is engineered to deliver dramatically higher performance, improved energy efficiency, and enhanced AI capabilities.

Discovery will drive advancements in numerous fields, including the design of next-generation reactors, batteries, catalysts, semiconductors, and critical materials. Its key features include impressive memory capacity and network bandwidth that outpace first-generation exascale machines, a seamless user transition from the Frontier system, the ability to deliver more computational output at similar power costs, and a foundation built on open-source software and open standards to ensure AI sovereignty.

User operations for Discovery are expected to begin in 2029. Once fully operational, it will stand as a cornerstone of the American AI Stack, fundamentally advancing the nation’s scientific, security, and innovation objectives for the next decade.

(Source: ITWire Australia)