Aurora is up and running — and open to the global scientific community. The exascale supercomputer at Argonne National Laboratory, funded by the U.S. Department of Energy, recorded 1.012 exaFLOPS on the May 2024 Top500 list — one quintillion floating-point operations per second — and has been officially available for large-scale research projects since January 2025. Nuclear fusion sits at the very top of its agenda.
Built in partnership with Intel and Hewlett Packard Enterprise, Aurora is equipped with 63,744 GPUs and 84,992 network endpoints. The machine weighs 600 metric tons, occupies roughly 10,000 square feet of floor space, and its nodes are interconnected by 300 miles of networking cable. Each node carries two Intel Xeon Max processors and six Intel Max Series GPUs. On the AI performance front, Aurora claimed the top spot on the HPL-MxP benchmark ranking in November 2024 — the international standard for mixed-precision workloads. No other system in the world has matched that result.
The connection to nuclear fusion is anything but peripheral. Researchers are using Aurora to simulate plasma behavior inside tokamaks at a level of detail that was simply out of reach just a few years ago. The plasma stability problem — confining an artificial star without it flickering out or damaging reactor walls — demands extremely high-fidelity models. Aurora can run simulations in a fraction of the time that would previously have taken years. For next-generation fission reactors, the system’s simulation capabilities allow researchers to model extreme conditions, predict material behavior, and refine designs before a single physical prototype is built. The same applies to Monte Carlo codes that track the motion of subatomic particles inside fission systems.
Before its official launch, dozens of research teams had already been working on Aurora through the Aurora Early Science Program. The goals were twofold: scientific teams optimized their codes for the machine’s architecture, while ALCF staff collected data to fix hardware and software issues. The result was a suite of applications ready to go at production launch — the product of years of joint work between ALCF, Intel, HPE, and DOE researchers through the Exascale Computing Project, which concluded in 2023.
Aurora is not the only exascale system in existence. It shares that tier with Frontier at Oak Ridge National Laboratory and El Capitan at Lawrence Livermore National Laboratory — the three fastest systems on the planet, all American, all DOE-funded. But Aurora has a distinguishing trait: its AI-oriented architecture makes it particularly well-suited for workloads that blend physical simulation with machine learning, precisely the kind of approach that fusion research now requires to predict plasma instabilities and train predictive models in real time during experiments.
In the years ahead, as fusion programs such as ITER and private-sector projects move toward operational phases, access to computing power at this scale will be decisive. Simulating a reactor before switching it on is not a luxury — it is the only way to reduce uncertainty margins in systems operating at tens of millions of degrees. Aurora, open to researchers worldwide without commercial restrictions, becomes in this sense a shared infrastructure for the future of fusion energy.
