Air Force’s $20M Flyer Supercomputer Targets Hypersonic Research

The Air Force Research Laboratory commissioned Flyer, a $20 million supercomputer, at Wright-Patterson Air Force Base in Ohio on June 18. The system is built to handle hypersonic weapons research, AI model training, and next-generation aircraft design for the Department of Defense. AFRL officials said Flyer can complete in 24 hours what a standard laptop would need about 500 years to solve. The new machine is the tenth supercomputer AFRL has deployed at the base.

It is the unclassified half of AFRL’s TI-23 supercomputing initiative, originally announced in 2023. Its classified counterpart, Raven, has been operational for some time, an AFRL spokeswoman told the Springfield News-Sun. Together, the two systems are designed to deliver around 14 petaflops of capacity. Flyer alone clocks about 8.7 quadrillion calculations per second at peak, Dayton 24/7 Now reported. Officials said the machine will run nonstop for five years, saving the Department of War more than $800 million over its lifetime.

What the Air Force Just Turned On

AFRL cut the ribbon on Flyer at a June 18 ceremony at Wright-Patterson Air Force Base, the Air Force’s primary scientific research base. The machine is named after the Wright brothers, the Dayton bicycle mechanics whose 1903 Flyer airplane gave the city its aviation legacy. The naming was the centerpiece of the ceremony’s framing, with speakers tying the new system to Dayton’s role in the earliest days of flight.

Flyer is the tenth supercomputer AFRL has deployed at Wright-Patterson, joining earlier systems that include the 2014 Lightning Cray XC30 and its sibling Spirit. The new machine is the public-facing half of AFRL’s TI-23 supercomputing initiative, originally announced in 2023. Its classified counterpart, Raven, has been operational for some time, an AFRL spokeswoman told the Springfield News-Sun. Together, the two systems are designed to deliver roughly 14 petaflops of capacity, placing TI-23 among the most powerful military research computing initiatives in the United States.

The table below summarizes the specifications AFRL disclosed at the June 18 unveiling. Peak performance is from the lab’s published hardware specs. The supporting figures are from AFRL statements and from reporting at the ceremony.

Inside the Machine

The hardware inside Flyer follows a hybrid CPU-plus-GPU architecture, the same broad pattern used by AI training clusters across industry. The compute layer is built on 884 nodes holding 169,782 AMD Epyc “Genoa” processor cores. The accelerator layer adds 64 Nvidia H100 GPUs and 16 Nvidia L40 GPUs, per Flyer’s 884 compute nodes and Nvidia accelerators. That mix is designed to split general-purpose simulation off from AI and visualization work.

• 884 compute nodes carrying 169,782 AMD Epyc “Genoa” cores for general simulation and physics modeling
• 64 Nvidia H100 GPUs for AI training and inference at scale
• 16 Nvidia L40 GPUs for visualization and smaller AI workloads
• 800 TB of system RAM and 18 PB of storage backing the whole stack

Bryon Foster, director of AFRL’s DoD Supercomputing Resource Center, said Flyer would be dedicated to solving difficult problems for the Department of War. The system’s specifications translate to roughly 8.7 quadrillion calculations per second at peak, Dayton 24/7 Now reported. Marc DeNofio, an AFRL spokesperson, said Flyer can “crack problems in 24 hours that would take a standard computer half a millennium to solve.” DeNofio added that the scale of data modeling gives “strategic planners and warfighters unmatched predictive clarity for decisive battle scenarios.” AFRL officials have previously said matching Flyer’s memory capacity alone would require roughly two million laptops.

Flyer’s 800 terabytes of RAM, by AFRL’s count, equal the working memory of about two million conventional laptops. The 18 petabytes of storage can hold an estimated 3.6 billion photos or 46 years of high-definition video. That capacity is what the modeling and simulation workloads Flyer is meant to run require. Without it, those workflows do not fit on smaller systems.

The compute is sized for problems too hazardous, too expensive, or too slow to tackle through physical testing alone. That is the operational logic Brig. Gen. Douglas Wickert, the AFRL commander, and Kelly Dalton, the HPCMP director, kept returning to at the ceremony. The savings claim is built on the same logic: cutting simulation timelines from months to weeks compounds across the Department of Defense’s research portfolio.

Built in the Wright Brothers’ Spirit

AFRL officials used the June 18 ribbon cutting to frame Flyer as a continuation of Dayton’s aviation tradition. Kelly Dalton, director of the Department of Defense’s High Performance Computing Modernization Program, opened her remarks with a comparison to the Wright brothers’ early work. Dalton recalled that the brothers built a six-foot wind tunnel in their west Dayton bicycle shop to refine their first airplane designs. They then ran systematic laboratory experiments on hundreds of air foils, Dalton said, translating raw empirical data into precise mathematical equations.

The Wright brothers did not just build an airplane. They calculated their way into the sky. It is in this spirit that we dedicate the ‘Flyer’ supercomputer today.

The line came from Kelly Dalton, director of the Department of Defense’s High Performance Computing Modernization Program, speaking at the June 18 ribbon cutting. Wickert placed Flyer in a longer arc: “The 21st century is the century of data.” The capabilities from the Nvidia chips and GPUs that Flyer brings, Wickert said, will allow the lab to “invent the future so that we can win the future and preserve the peace and prosperity of the global world order.” “That is Dayton’s legacy,” Wickert added. U.S. Rep. Mike Turner, R-Dayton, a member of the House Armed Services Committee, called Flyer “an investment that is going to be paying dividends for the future.”

Officials said Flyer cost about $20 million and will run nonstop for the next five years. The expected lifetime savings to the Department of War top $800 million. That figure comes from cutting simulation timelines from months to weeks across multiple research programs that previously relied on physical testing.

Hypersonics, AI, and the Workload Flyer Will Run

The Department of Defense listed three primary workloads for Flyer: hypersonic weapons research, AI model development, and next-generation aircraft design. All three share one constraint: they cannot be fully replicated in physical testing. Hypersonic vehicles traveling several times the speed of sound are difficult and prohibitively expensive to test in real-world conditions. AFRL has previously highlighted hypersonic vehicle development as one area where supercomputers provide the clearest advantage over wind-tunnel or flight-test programs.

Next-generation aircraft and weapons systems generate enormous volumes of data from sensors, flight tests, and satellite feeds. Processing that information requires compute capable of trillions to quadrillions of calculations per second. Flyer is sized to handle that data load at simulation speeds.

AI development has its own compute hunger, one that has reshaped supercomputer design across the wider industry. AFRL has emphasized the role of high-performance computing in training more sophisticated AI models and running larger-scale experiments. Flyer joins a wave of hybrid CPU-plus-GPU systems built specifically with that workload in mind. Its 64 Nvidia H100 GPUs are the same accelerator family used in commercial AI training clusters worldwide. Flyer’s job is to apply that same scale to military AI programs with no commercial counterpart.

Flyer is meant to let researchers simulate scenarios that are too hazardous, too expensive, or too slow to test in the real world. That includes high-fidelity aerodynamic models, virtual flight testing, and full-weapons-system simulations. Previous AFRL systems have reduced simulation projects from months to weeks while enabling more detailed studies of complex engineering problems, the lab has said. The pitch is fewer physical prototypes and faster iteration on digital ones.

Hypersonic and AI workloads are not new to AFRL. What Flyer adds is the scale to run them in parallel and at higher fidelity than the lab’s older systems allowed. That extra headroom is the system’s real selling point at the ceremony.

The Tenth Computer on the Base

Flyer joins nine other AFRL supercomputers at Wright-Patterson, where the AFRL DoD Supercomputing Resource Center is the largest high-performance computing facility in the Air Force. Flyer is dedicated to unclassified workloads, the lab said. Classified work runs on Raven, which has been operational for some time. An AFRL spokeswoman told the Springfield News-Sun that Raven is already at work on its assigned programs. Together, the two systems form a paired public-and-classified research stack under the TI-23 initiative.

The split between Flyer and Raven mirrors how the broader Department of Defense organizes its high-performance computing fleet. The DoD operates multiple supercomputing resource centers across the country, each tied to a service branch. Wright-Patterson’s DSRC is one of the Air Force’s two main centers; the other is AFRL’s Maui facility, which runs the Riptide system. Flyer is the most recent arrival in that distributed fleet.

AFRL’s earlier systems at the base include the 2014 Lightning Cray XC30, which AFRL unveiled as part of a broader $150 million HPCMP acquisition cycle that year, in a piece covering AFRL’s earlier Lightning supercomputer at the same base. Lightning and Spirit together doubled the AFRL DSRC’s compute power when they came online. The pattern at Wright-Patterson has been a roughly four-to-five-year refresh cycle, with each new machine replacing the generation before it.

Flyer fits the same pattern: it is sized for a five-year nonstop run before the next generation comes in. The $20 million price tag is in line with the cost of earlier AFRL systems, including Lightning at $20.8 million in 2014 dollars. The savings claim, $800 million over Flyer’s lifetime, is a projection built on the assumption that the lab’s simulations keep replacing physical testing at the rate officials described. None of those figures have been audited externally.

The Simulation-First Reality Behind the Headlines

Flyer is the largest expression of a broader shift in how AFRL approaches military research: simulation-driven workflows for hypersonic, AI, and aircraft design run through digital models before any physical prototype. The TI-23 initiative under the Department of Defense’s High Performance Computing Modernization Program is the institutional expression of that move. Flyer is the largest unclassified surface of that program so far. Its arrival closes one compute gap and signals the next: every new AFRL machine is now expected to carry a simulation-first workload from day one. That is what the next-generation AFRL procurement cycle will be sized for.

Hypersonic, AI, and next-generation aircraft work has migrated steadily toward digital engineering workflows since the early 2010s. Flyer, like Lightning and Spirit before it, is built to push that migration further. Its job is to carry that workload through the end of the decade.