Scientists have relied on HPC facilities to help them process large volumes of data, execute sophisticated simulations, and power large-scale research. However, as AI becomes more deeply integrated into scientific research, the role of HPC facilities may also be starting to evolve.
According to Thomas Uram, Group Leader of the Data Services and Workflows Team at Argonne National Laboratory (ANL), the next generation of HPC facilities will do more than provide scientists with access to compute. 
In his presentation at the ISC 2026, Uram outlined how AI inference services and access to computing resources are transforming supercomputing centers into platforms that actively support scientific discovery. He said that the transformation begins by building AI capabilities directly into the facility rather than treating them as standalone tools. 
“In addition to systems like Aurora, we’re also actively developing and deploying systems for AI inference because we recognize that this is an enormous opportunity for the future… Having dedicated systems for inference is one of the key factors that we’re after here.
“Everyone talks about the frontier models and believes that they need the frontier models for their science. In many cases that’s increasingly not true. There has been a lot of discussion this week about smaller models that are sufficient for various problems, smaller models that are faster for the tasks that you’re after. Maybe they can run even on the same machine that you’re using for your simulations. So there’s a distribution here that we can exploit.”
(Source: ISC 2026 presentation by Thomas Uram)
To support that vision, Argonne is building a dedicated AI infrastructure alongside its traditional HPC systems. Uram mentioned several inference-focused platforms, including Sofia (A100), Minerva (B200) and Tara (GH200). 
While Aurora is the main compute platform, the others are being added specifically to provide AI inference as a facility service. Researchers can access dozens of open-weight large language models and domain-specific science models through a centralized inference service. 
Uram argued that many scientific workloads don’t need expensive frontier models anymore. Instead, researchers can use open models through a shared inference service. They don’t have to build or manage the infrastructure themselves. But inference alone isn’t enough.
For AI to become part of scientific workflows, however, it also needs a way to interact with HPC systems. According to Uram, that requires more than inference. AI agents must also be able to access computing resources, submit jobs, and coordinate work across multiple systems. 
“So I mentioned the hardware, and I mentioned the software, which is the inference service, and I’ll mention one other thing, which is, how do agents get access to the systems? We have a variety of different systems at ALCF (Argonne Leadership Computing Facility), and some of them are better for some tasks than they are for other tasks…
“One of them is having a job submission API to the various systems. You have an agent that lives on your laptop or elsewhere, so that it can orchestrate tasks running across different systems… Those things all together, I would say, the hardware, the inference service, and the ability to submit jobs is basically laying the foundation for AI workflows.”
Uram demonstrated how those capabilities are already being applied across a range of scientific disciplines. 
At the Advanced Photon Source, Argonne’s synchrotron X-ray facility, researchers automatically transfer experimental data to ALCF as it is generated, run analyses on the facility’s computing resources, and return the results during the experiment itself. The same infrastructure is now being used to apply AI-based image segmentation. This allows scientists to analyze tomography data in near real-time rather than waiting until experiments have concluded and results have been gathered.
(Source: ISC 2026 presentation by Thomas Uram)
He also highlighted similar work supporting fusion research, where researchers have just 20 minutes between experiment cycles to analyze results before the next run begins. AI inference services and automated workflows allow data to be processed immediately. Researchers are now looking to combine those capabilities with digital twins that can operate alongside live experiments.
Looking ahead, Uram described AI agents that can go beyond analyzing results, using drug discovery as one example.
“The reasoning agent plans out how to approach this problem, interacts with the system and runs simulations. And in this closed loop, can explore the results of the simulation and how they factor into the objective, and continue to produce simulations that should be run until the result is achieved.”
“They were able to, one, test a lot of targets, but also by doing this and having it be AI driven, they were able to arrive at targets that outperformed what the targets that humans were able to predict.”
Uram’s presentation highlighted the changing role of HPC facilities. They are no longer just providers of compute, but are now platforms that integrate AI inference, workflow orchestration and programmatic access to support scientific discovery. As AI becomes more deeply embedded in research, that evolution could fundamentally change how scientists interact with supercomputers and how future HPC facilities are designed. 
Modern AI can identify forests all the way from space. It can estimate tree heights…
AI is becoming an increasingly important tool for drug discovery. However, for many pharmaceutical and…
Last week’s TPC26 event brought together many of the world’s preeminent experts in using AI…
The AI industry has spent the past few years trying to give models access to…
AI is creating a new set of demands for HPC centers. Researchers are no longer…
Have we reached a point where AI agents can reliably function as scientific collaborators? Can…
Elon Musk’s SpaceX has filed for what could become the largest IPO in history. It…
Progress in artificial intelligence continues to accelerate across a range of expert disciplines, according to…
Forbes recently released its 2026 AI 50 list that showcases the most influential and fast-growing…
For years, hyperscalers built infrastructure largely for their own cloud ecosystems. However, AI is starting…
AWS expanded its enterprise AI portfolio this week with a collection of new tools aimed…
Datadog’s latest State of AI Engineering report points to a measurable failure problem in enterprise…
More than four out of five (83%) organizations say they need to upgrade their infrastructure…
The next wave of enterprise standardization is here – and it’s about control, not just…
Meta has announced that it will start manufacturing a new custom AI chip from September…
Modern AI can identify forests all the way from space. It can estimate tree heights…
Artificial intelligence has redrawn the data center design landscape, and in doing so has propelled…
Microsoft has launched Microsoft Frontier Company, a new consulting organization built to help enterprises plan…
Alex Woodie Editorial Director +
HPCwire Managing Editor
Ali Azhar BigDATAwire Managing Editor
Jaime Hampton AIwire Managing Editor
Drew Jolly QCwire Managing Editor
Position: High Performance Computing Engineer
Location: Switzerland
ISC 2026 connects scientists, engineers, and technology leaders to explore breakthroughs in AI, HPC, and quantum technologies at the EU’s longest-running HPC conference.

The National Software Testing Conference is a globally recognised event for professionals in software testing, quality assurance, and quality engineering. Bringing together industry leaders, innovators, and practitioners, the conference provides a dynamic platform to explore the latest trends, tools, and strategies shaping the future of quality engineering.
With artificial intelligence transforming the technology landscape, now is the time to stay ahead of the curve. Join us for a two-day conference in July designed to help you navigate this rapidly evolving industry. Engage with influential experts and thought leaders who are redefining the role of testing and quality in modern software development.
Take part in interactive workshops that explore the practical applications of AI in testing and discover new approaches to improve efficiency, innovation, and impact. Gain valuable insights you can apply immediately to your work and career.
Be part of the conversations driving the future of software quality and innovation.

For decades, Google Cloud has supported scientific and technical computing for researchers, developers, and engineers across industry, academia, and government labs. With recent major investments in HPC infrastructure and AI, Google Cloud aims to be the most comprehensive, capable, and accessible platform for scientific discovery and technical computing.
Join us at in person at Google Chicago to learn more about our commitment to this mission, explore our newest infrastructure, and see how Google Cloud can accelerate your organization’s innovation and productivity. Through a mix of presentations, hands-on labs, panels, and networking, you will gain valuable insights and forge new collaborations. This is your opportunity to connect with Google experts, HPC and AI practitioners, and peers dedicated to elevating their expertise and driving organizational success.

© 2025 BigDATAwire. A TCI Media Publication (Formerly Tabor Communications). All Rights Reserved.
Use of this site is governed by our Terms of Use and Privacy Policy.