Argonne National Laboratory launches AI inference service from spare supercomputing capacity

Argonne National Laboratory Transforms Idle Computing Power Into Research Tool

The Department of Energy's Argonne National Laboratory near Chicago has launched an AI inference service built from spare supercomputing capacity, marking a significant shift in how scientific institutions leverage their existing infrastructure. Announced on Tuesday, the service provides researchers across the United States—including those at DOE labs and teams working on the Genesis Mission—with secure access to advanced AI capabilities for scientific discovery1

1

.

Argonne houses some of the world's largest computing clusters, including the No. 3-ranked Aurora supercomputer. The new service taps into smaller, AI-optimized systems that weren't being fully utilized. Currently, the platform runs on two clusters: the Sophia system with 192 Nvidia A100 GPUs (most with 40 GB of memory), and Metis, featuring 32 of SambaNova's SN40L AI accelerators. Plans are underway to extend the service to the Nvidia GH200-based Tara and B200-based Minerva systems.

Cloud-Like Access to Large Language Models for Open Science

The AI inference service offers cloud-like access to large language models through a chatbot-like portal, with Argonne appearing to use Open WebUI, a popular self-hosted chatbot service, for at least some of its operations. Researchers can access OpenAI's GPT-OSS, Google's Gemma family, Meta's Llama herd, and domain-specific models like AuroraGPT1

1

. The service also includes computer vision models and in-house models developed at Argonne2

2

.

"By making AI inference available as a shared resource, we are enabling researchers to apply AI at scale to their data, their simulations and their experiments without having to build and maintain their own infrastructure," said ALCF director Michael Papka1

1

. The Argonne Leadership Computing Facility (ALCF) is a DOE Office of Science user facility.

High-Performance Computing Systems Enable Multi-Lab Collaboration

The service has gained traction across the DOE national laboratory ecosystem, actively supporting researchers from Brookhaven National Laboratory, Fermi National Accelerator Laboratory, Los Alamos National Laboratory, Lawrence Berkeley National Laboratory, Lawrence Livermore National Laboratory, Oak Ridge National Laboratory, Sandia National Laboratories, and Thomas Jefferson National Accelerator Facility. Users can access the platform using their home institution credentials, enabling seamless integration into existing research workflows2

2

.

This cross-lab adoption positions the service as a key enabler for DOE's Genesis Mission, a national AI initiative to build the world's most powerful scientific platform. The service will also support the American Science Cloud (AmSC), the Genesis Mission's integrated environment connecting DOE supercomputers, experimental facilities, and data resources2

2

.

AI Models on High-Performance Computing Accelerate Data Analysis

Researchers are already deploying these AI models on high-performance computing infrastructure for practical applications. In fusion energy research, scientists analyze experimental data in real time to predict plasma disruptions before they occur, enabling safer reactor control. Teams working with particle accelerators and telescopes use the service to sift through massive data volumes, narrowing the search radius for rare events and new phenomena rather than wasting computational cycles on brute-force approaches1

1

.

In materials science and chemistry, researchers are using ChemGraph, an AI framework that leverages the ALCF Inference Service for LLM-driven tasks. This allows scientists to simplify molecular simulation workflows, explore more candidate molecules, and manage large-scale calculations as an integrated process2

2

.

Secure Alternative to Public Generative AI Platforms

A critical advantage of the service is enabling DOE researchers to experiment with generative AI and LLMs in a secure environment that doesn't expose sensitive data to public services like ChatGPT. "Inference services allow researchers to spend less time managing models and more time testing hypotheses," said Venkat Vishwanath, AI and machine learning lead at the ALCF. "Instead of taking days or weeks to analyze data, scientists can rapidly interpret results, refine experiments and explore complex systems in ways that weren't practical before"2

2

.

The service grew from a 2025 paper outlining a framework for providing secure and scalable AI inference on high-performance computing systems. The goal was to enable researchers to run multiple AI tasks in parallel on different models without relying on commercial cloud services2

2

. While concerns about hallucinations and erroneous behavior in generative AI persist, a growing body of research suggests these technologies can automate research tasks and supplement traditional climate or physics models. Lawrence Livermore National Laboratory, for instance, has used El Capitan—the world's most powerful publicly known supercomputer—to develop tsunami forecasting models, while Nvidia has demonstrated that AI climate models can identify storm cells faster and more accurately than existing approaches1

1

.