Exascale computing has moved from engineering ambition to operational reality. As of 2024, there are three exascale systems currently operational at U.S. Department of Energy (DOE) National Laboratories: El Capitan, Frontier, and Aurora, each capable of at least one quintillion calculations per second. The fourth exascale computer, JUPITER, was declared Europe’s first exascale computer in November 2025 and will be located at the Jülich Supercomputing Centre in Germany, ranking No. 4 on the TOP500 list. These systems are already producing scientific results across climate science, national security, and biomedical research.
This growth highlights how exascale computing is reaching tremendous heights in the computing industry. Kings Research notes that the global exascale computing market is expected to reach $20,289.3 million by 2031.
This blog covers what exascale computing is, how it compares to previous generations, the systems leading the field, and the developments shaping 2026.
What is Exascale Computing?
DOE defines exascale computing as the ability to deliver at least one exaflop, or 10^18 floating-point operations per second, on standardized benchmarks. In other words, this means one quintillion floating-point operations per second, sustained over millions of processing cores. It is capable of converged workloads, meaning it can deliver modeling, simulation, AI inference, and data analytics on the same platform simultaneously. This is what differentiates exascale from previous generations of supercomputing and makes it immediately relevant to the most computationally intensive problems facing science today.
Exascale vs. Petascale: How Big is the Leap?
One exaflop equals 1,000 petaflops, which is not an incremental upgrade. The performance gap between petascale and exascale is comparable to the gap between a pocket calculator and a modern laptop, except the underlying workloads are nuclear physics simulations and planetary climate models.
The World's Operational Exascale Systems
Frontier, Oak Ridge National Laboratory
Frontier gained worldwide recognition as the world’s first exascale computer in 2022, realizing a benchmark HPL performance of 1.102 Exaflop/s in its first verification test by TOP500. Built on the HPE Cray EX235a architecture with AMD EPYC processors and 8,699,904 cores, it achieved this performance while maintaining an energy efficiency of 52.59 gigaflops/watt, simultaneously ranking on the GREEN500 list. Frontier enables active research in nuclear energy modeling, materials science, and cancer research simulations at Oak Ridge National Laboratory.
Aurora, Argonne National Laboratory
Aurora takes its place as the second known Exascale system, reaching speeds of 1.012 Exaflop/s using the HPL benchmark. Built using an Intel-based architecture, Aurora was designed as a converged high-performance computing (HPC) and artificial intelligence (AI) workload machine. Currently, at the Argonne National Laboratory, Aurora hosts climate modeling, cosmology, and large-scale genomics research projects.
El Capitan, Lawrence Livermore National Laboratory
El Capitan currently leads the list with a score of 1.809 exaflop/s, as confirmed at the SC25 conference in November 2025. It uses the HPE Cray EX255a architecture, featuring AMD 4th Generation EPYC processors and AMD Instinct MI300A accelerators. It also ranked 23rd on the GREEN500 list, with a performance of 60.94 gigaflops/watt. El Capitan was mainly built for the National Nuclear Security Administration’s nuclear stockpile stewardship program, which is a simulation of the safety and reliability of nuclear weapons instead of physical testing.
Jupiter, Forschungszentrum Jülich, Europe
JUPITER reached the operational stage at Forschungszentrum Jülich in Germany in the year 2024 and reached the milestone of confirmed exascale performance at 1.01 ExaFLOP/s in November 2025, thus making it the first exascale computer in Europe to be ranked in the TOP500 list. This marks the beginning of exascale supercomputing outside the DOE infrastructure.
Key Applications of Exascale Computing
There are a multitude of key applications, including:
Climate Modeling and Environmental Science
Exascale systems enable kilometer-scale resolution climate models that petascale systems could not sustain.
Healthcare and Drug Discovery
Frontier is already running cancer research simulations at Oak Ridge, applying predictive molecular modeling to identify drug interaction pathways at a speed and resolution impossible on earlier systems.
National Security and Nuclear Stewardship
El Capitan simulates the physical behavior of aging nuclear assets without requiring live testing, a mission-critical application that defines the system's procurement and operational priorities.
Artificial Intelligence and Machine Learning
The exascale infrastructure is becoming more and more foundational for the training and execution of large-scale AI models. Aurora was specifically designed to support converged HPC and AI workloads, allowing for the execution of classical simulations and AI inference on a single system. With the continued increase in the size of AI models, the exascale architecture is becoming more foundational for scientific AI development.
Energy and Materials Science
The Exascale Computing Project (ECP) in the DOE has shortlisted clean energy research, nuclear reactor simulation, and materials science as the major application areas. These applications require simulation complexity that can only be addressed by exascale computing within a reasonable time frame.
Technical Challenges of Exascale Computing
Despite their challenges, exascale supercomputing faces numerous challenges, including:
Power Consumption
Achieving exascale performance while staying within a practical power budget was a key engineering challenge of the decade. Frontier tackled this with architectural innovation.
Data Movement and Memory Bandwidth
Exascale supercomputing addresses data and memory bandwidth issues by utilizing a high-bandwidth memory that is stacked directly on the GPU dies, thus increasing the data movement speed by one order of magnitude compared to traditional memory architectures.
Software Scalability
The programming systems that consist of millions of processing cores require new software paradigms. The development of applications that can effectively use this degree of parallelism remains one of the most active areas of research in the field.
Fault Tolerance and Reliability
Exascale systems need to be able to detect, isolate, and recover from component failures without stopping active scientific workloads, a reliability need that requires a software layer for fault tolerance.
Industry Example: The DOE Exascale Computing Project
The DOE Exascale Computing Project, launched in 2016 and completed in April 2024, took seven years and was a $1.8 billion undertaking. It involved about 3,000 multidisciplinary researchers. The ECP provided a full software ecosystem including applications, programming tools, and middleware tailored to exascale architectures. Most importantly, it provided production-level scientific applications at the time of the Frontier supercomputer’s activation, thus obviating the usual software development period that follows the installation of a new supercomputer.
What Comes After Exascale?
The next theoretical performance milestone that is being pursued is zettascale computing, which is 10^21 operations per second, and is still in the early stages of research planning. Neuromorphic computing, which is based on the structural layout of the human brain, is also being explored for efficient computing at and beyond exascale scales.
Quantum computing is being explored as a complement to exascale classical computing for specific problem types, cryptography, optimization, and quantum chemistry simulations. AI system management, where quantum computing outperforms classical computing, is already being used to optimize exascale computing, predict hardware failures before they happen, and improve energy efficiency in active systems.
Conclusion
Exascale computing has evolved from a benchmarking goal to a reality. The next era will focus on access and software maturity, and as the cost of development decreases and application communities grow, exascale supercomputing will move beyond the national laboratories into the academic and commercial research communities.
