On the 18th of June 2026, Air Force Research Laboratory officials gathered in a secure building on Wright-Patterson Air Force Base's Area B for a ribbon-cutting ceremony that, in their telling, echoed a ceremony performed 122 years earlier a few miles away. That earlier moment involved Orville and Wilbur Wright and a canvas-and-spruce glider. This one involved rack upon rack of NVIDIA-accelerated processing nodes, a machine they named in deliberate homage to the Wrights' Flyer of 1903—and which they say can accomplish in 24 hours what would require approximately 500 years of continuous computation on a modern consumer laptop.

The new system, officially designated TI-23 and christened "Flyer," represents AFRL's latest technology insertion under the Department of Defense High Performance Computing Modernization Program (DoD HPCMP), a congressionally directed program in continuous operation since 1992. The $20 million machine joins nine other supercomputers already resident at Wright-Patterson, which hosts the largest Air Force DoD Supercomputing Resource Center (DSRC) of the program's five centers nationwide.

System Architecture and Throughput

AFRL officials have disclosed the top-level configuration of Flyer: 186,000 processing cores, 800 terabytes of random-access memory, and 18 petabytes of storage capacity, all accelerated by NVIDIA graphics processing units whose specific model designations were not publicly disclosed at the ribbon-cutting. Benchmarked peak throughput is reported at 8.7 petaflops—that is, 8.7 quadrillion floating-point operations per second. To contextualize that number: if every human being on Earth performed one arithmetic operation every second without pause, it would take the planet's entire population roughly four years to equal what Flyer computes in a single second.

Flyer's 800-terabyte RAM pool is equivalent, by AFRL's own metric, to the working memory required to outfit two million conventional laptops simultaneously. Storage capacity is expressed by the laboratory in more cinematic terms: 3.6 billion photographs, or 46 years of high-definition video. The system will operate continuously—24 hours a day, 365 days a year—for at least five years. Bryon Foster, director of AFRL's DoD Supercomputing Resource Center, has projected that the $20 million capital cost will yield more than $800 million in savings to the Department over Flyer's operational lifetime, chiefly by displacing expensive physical testing with validated high-fidelity simulation.

"All of this will be dedicated to solving problems—difficult problems—for the Department of War." Bryon Foster, Director, AFRL DoD Supercomputing Resource Center, 18 June 2026

Flyer operates as an unclassified resource. Its companion system, "Raven"—also designated TI-23—handles classified workloads and was commissioned ahead of Flyer's public rollout; an AFRL spokeswoman confirmed Raven has been operational for some time. Together, the two TI-23 machines are projected to deliver a combined 14 petaflops of computational capability, supporting both open and controlled research environments within AFRL's mission space.

The DoD High Performance Computing Modernization Program Context

Flyer is not an isolated procurement but the latest increment in a deliberately rolling modernization strategy that AFRL has operated for more than three decades. The DoD HPCMP, managed through the Office of the Assistant Secretary of the Army for Acquisition, Logistics, and Technology, maintains five DSRCs—at Wright-Patterson (Air Force), Aberdeen Proving Ground (Army Research Laboratory), Stennis Space Center (Naval Meteorology and Oceanography Command), Vicksburg, Miss. (Army Engineer Research and Development Center), and Maui, Hawaii (Maui High Performance Computing Center). The centers share computational resources across the Air Force, Army, Navy, and affiliated DoD organizations, including sponsored contractors, civilian researchers, and academic partners.

Kelly Dalton, director of the DoD High Performance Computing Program, described the procurement philosophy at the ribbon-cutting: the center operates on a continuous insertion cycle in which each system's procurement often overlaps with the commissioning of its predecessor. "Sometimes even before the system is running, we're ordering the next one," Foster had noted in an earlier interview, characterizing a planning cadence driven by the scale and lead time of large HPC procurements. Dalton drew a direct line from the Wright Brothers to the current investment, noting that Orville and Wilbur built a six-foot wind tunnel in their west Dayton bicycle shop to systematize aerodynamic experimentation—an early form of computational substitution for costly trial-and-error. "The brothers then conducted systematic, rigorous laboratory experiments on hundreds of airfoils, translating raw empirical data into precise mathematical equations," Dalton said. "They computed lift, drag, and thrust."

Predecessor Systems and the AFRL HPC Lineage

AFRL's supercomputing lineage at Wright-Patterson follows a naming tradition drawn from aviation history. Flyer's immediate predecessor in the unclassified tier was Raider—designated TI-21 in honor of the Doolittle Raiders of World War II—commissioned in September 2023. Raider, built by Penguin Systems and powered by AMD CPUs paired with NVIDIA GPUs, delivered approximately 12 petaflops and briefly ranked in the vicinity of the fiftieth most powerful system on the TOP500 list. Its own predecessor, Thunder, installed in 2015, produced 3.1 petaflops—less than one thirty-sixth of the performance projected from the combined Flyer-Raven TI-23 pairing.

  • 2015 Thunder commissioned — 3.1 petaflops; AMD/NVIDIA hardware.
  •  2019 AFRL unveils first-ever shared classified DoD HPC capability at WPAFB, comprising Mustang (unclassified), Voodoo, Shadow, and Spectre (classified tiers).
  •  2023 Sep Raider (TI-21) commissioned — ~12 petaflops; Penguin Systems, AMD CPUs + NVIDIA GPUs; accelerated a Navy simulation project from six months to three weeks.
  •  2023 Q4 AFRL orders TI-23 Flyer and TI-23 Raven, targeting combined 14 petaflops; Raven (classified) begins installation and testing.
  •  2026 Jun Flyer (TI-23, unclassified) commissioned at WPAFB Area B ribbon-cutting ceremony — 8.7 petaflops, 186,000 processors, 800 TB RAM, 18 PB storage, NVIDIA GPUs; Raven already operational.

The acceleration trajectory is instructive from a systems-engineering standpoint: AFRL has achieved roughly a 4.5-fold improvement in peak throughput from Raider to Flyer across a three-year cycle, consistent with the broader industry trend of GPU-accelerated supercomputing scaling faster than Moore's Law predictions for CPU-only clusters. The shift from CPU-dominant to GPU-dominant architectures, already evident in Raider's design and now fully realized in Flyer's NVIDIA-centric configuration, reflects the same architectural pivot that has driven commercial AI infrastructure investment at hyperscaler scale.

Mission Applications: Hypersonics, AI, and Digital Engineering

The primary mission drivers for Flyer span three intersecting domains of current Air Force priority: hypersonic weapons development, artificial intelligence and machine learning infrastructure, and next-generation aircraft digital engineering. Each represents a category of computational workload that exceeds the practical capacity of earlier AFRL systems and, in some cases, cannot be meaningfully addressed through physical testing at all.

Hypersonic vehicle modeling is particularly demanding. Aircraft traveling at speeds exceeding Mach 5 generate aerothermal environments—shock-induced heating, plasma sheath formation, turbulent boundary layer transitions—whose physics are nonlinear and tightly coupled across fluid dynamics, thermochemistry, and material response timescales. Ground-based testing in hypersonic wind tunnels, such as AFRL's own Mach 6 Ludwieg tube facility at Dayton, provides short-duration burst data on the order of 100 milliseconds per run, severely limiting the scope of design parameter exploration. Computational fluid dynamics codes running on HPC resources extend that exploration space by orders of magnitude. AFRL has previously noted that hypersonic vehicles "travel too fast to test on a range"—a constraint that elevates computational simulation from a supplementary tool to the primary design environment for critical performance envelope regions. The HPCMP's Frontier program has separately funded computational ceramics research aimed at identifying thermal protection materials capable of surviving hypersonic aerothermal loads, predicting approximately 900 candidate compounds in a single computational campaign that would have been impractical through laboratory synthesis alone.

AI and machine learning workloads represent the second major demand driver. AFRL's Digital Capabilities Directorate has articulated compute capacity as the foundational substrate of the laboratory's digital transformation strategy. Brig. Gen. Douglas "Beaker" Wickert, AFRL's commanding general, spoke in those terms at the commissioning ceremony: "The 21st century is the century of data. It's the century of information, and the capabilities—the NVIDIA chips and the GPUs that Flyer is going to bring—are going to allow us to invent the future." Large-scale AI training runs and inference workloads for autonomous systems, sensor fusion, electronic warfare signal processing, and logistics optimization all benefit directly from GPU-dense HPC configurations, and AFRL researchers have access to Flyer's resources through the same allocation mechanisms used by the broader DoD community.

"The Wright brothers did not just build an airplane. They calculated their way into the sky." Kelly Dalton, Director, DoD High Performance Computing Program, 18 June 2026

Digital engineering for next-generation aircraft—the third pillar of Flyer's stated mission—encompasses high-fidelity aerodynamic simulation, structural analysis, propulsion integration modeling, and electromagnetic compatibility prediction across platform concepts that may never be physically prototyped at intermediate design stages. The DoD HPCMP's CREATE software suite, developed specifically for defense acquisition CFD and computational electromagnetics, runs on AFRL's DSRCs. Historically, the same modeling and simulation paradigm accelerated a Navy project from six months to three weeks on the Raider platform; AFRL officials project proportional or greater gains on Flyer's 8.7-petaflop architecture.

Cost-Benefit Architecture: Simulation as Test-Cost Offset

The $800 million lifetime savings figure cited by program director Dalton warrants engineering-level scrutiny. The HPCMP's value argument rests on a well-established cost-offset model: replacing physical hardware-in-the-loop testing, range instrumentation, and experimental hardware fabrication with validated simulation reduces total acquisition cost and accelerates program schedules. The model has documented precedents. At the system level, computational design exploration compresses the design space before hardware commitment, reducing expensive late-cycle engineering changes. At the material level, as illustrated by the HPCMP Frontier ceramics program, HPC allows prediction and down-selection among thousands of candidate compounds, reserving physical synthesis for a much smaller validated set.

The 40-to-1 savings ratio implied by $20 million in capital cost against $800 million in projected savings is aggressive but not unprecedented in DoD HPC program documentation. It reflects the cumulative value of simulation hours provided across multiple programs and multiple services over a five-year service life, rather than a single-program test-replacement calculation. Independent verification of that figure is not available in public sources; it is an AFRL/HPCMP program estimate presented at the commissioning event.

Infrastructure, Access, and Security Architecture

Flyer occupies a secure facility on Wright-Patterson's Area B and operates within the Defense Research and Engineering Network (DREN), the DoD's dedicated high-speed wide-area network connecting the five DSRCs and more than 210 DoD sites including research laboratories, test centers, universities, and industrial facilities. DREN's fourth-generation infrastructure, provisioned under a commercial services contract awarded to Verizon in 2021, delivers data transfer rates from 1 Gbit/s to 100 Gbit/s and operates as an IPv6 network with legacy IPv4 support.

Access to Flyer is not restricted to uniformed Air Force personnel. As with the other AFRL DSRC systems, the resource is available to DoD-sponsored contractors, civilian researchers, military users from all services, and academic investigators operating under a valid DoD sponsor relationship. The HPCMP manages allocation through a competitive proposal review process, with dedicated allocation tracks including the Frontier Project program for large-scale computationally intensive research. The complementary Raven system handles classified workloads under the same HPCMP umbrella but with access governed by cleared-facility and need-to-know restrictions appropriate to its classification tier.

In May 2025, the HPCMP further expanded its compute ecosystem by authorizing InfiniteTactics' Analytics Gateway platform—known as AWS-Gate—to operate as a hybrid HPC-as-a-service layer bridging DSRC on-premises resources and AWS GovCloud capacity. That authority-to-operate approval formally extended HPCMP-managed resources into commercial cloud infrastructure for the first time, allowing burst capacity for AI/ML workloads that periodically exceed on-premises queue bandwidth. Flyer operates within this hybrid ecosystem as the on-premises anchor of unclassified compute, with AWS-Gate providing elastic overflow.

Congressional Context and the Department of War Rebranding

Representative Mike Turner (R-Dayton), a senior member of the House Armed Services Committee, attended the ribbon-cutting and framed the investment in explicitly competitive terms: "This is the next level of what's going to occur right here for the men and women who serve in the Air Force and what they'll be doing here for the Air Force Research Laboratories." Turner's district has a direct institutional stake in AFRL's continued presence and investment: Wright-Patterson AFB is one of the largest employers in the Dayton metropolitan area, and AFRL's DoD DSRC anchors a regional defense technology ecosystem spanning Tier 1 and Tier 2 defense contractors, academic research partners, and small-business technology developers.

Several officials at the ceremony made use of the newly redesignated "Department of War" styling—a rebranding of the Department of Defense that has appeared in budget and organizational documentation in the current administration. Foster stated directly that Flyer's resources "will be dedicated to solving problems, difficult problems, for the Department of War." Dalton invoked the same designation in projecting Flyer's operational schedule. The nomenclature shift carries no programmatic consequence for HPCMP operations, which remain governed by the same statutory framework under the Assistant Secretary of the Army for Acquisition, Logistics, and Technology.

Significance in the Global HPC Landscape

At 8.7 petaflops, Flyer is a capable but not record-setting system by the standards of the global TOP500 supercomputer list; its immediate AFRL predecessor, Raider, ranked approximately fiftieth on the TOP500 at commissioning in 2023. The world's most powerful system, Frontier at Oak Ridge National Laboratory, operates at roughly 1,200 petaflops (1.2 exaflops). Flyer's strategic value lies not in raw top-list ranking but in its security-enveloped availability for defense-specific workloads that cannot be run on commercially accessible systems, its connection to DREN's low-latency classified-adjacent network infrastructure, and its integration with DoD-specific computational software frameworks—including the HPCMP CREATE suite for aerodynamic and electromagnetic simulation—that are maintained and optimized specifically for defense acquisition engineering.

The HPCMP's broader pivot toward AI and machine learning—formalized in a 2024 User Group Meeting and dedicated AI/ML workshop at the Doolittle Institute in Niceville, Florida—positions Flyer as part of a deliberate infrastructure build intended to support autonomous systems research, advanced signal processing, and decision-support AI across the services. That strategic alignment with the National Defense Strategy's AI/ML priority areas gives the Flyer commissioning significance beyond the raw petaflop count, as GPU-dense HPC infrastructure is the enabling substrate for large-scale model training and high-fidelity simulation-based reinforcement learning for autonomous vehicle and weapon system development.

Looking Forward

AFRL's documented procurement cycle—in which a successor system's acquisition often begins before its predecessor achieves full operational capability—suggests that planning for TI-25 or equivalent next-generation systems is likely already underway, consistent with Foster's earlier observation that the lead time for large HPC acquisitions demands overlapping procurement horizons. The trajectory from Thunder's 3.1 petaflops in 2015 to the combined 14-petaflop TI-23 pair in 2026 represents an approximately 4.5-fold improvement per generation over three generations—a cadence that, if sustained, would place AFRL's next DSRC system in the exaflop-class range by the early 2030s, coincident with the projected initial operational capability windows for several next-generation Air Force platform programs including the Next Generation Air Dominance family of systems.

Whether that projection holds will depend on GPU architecture roadmaps, DoD budget trajectories, and the degree to which hybrid cloud integration via HPCMP's AWS-Gate architecture alleviates pressure on discrete hardware insertions. For now, Flyer represents the latest proof point in a continuous computational arms race that the Air Force Research Laboratory has been running—largely out of public view—since the Wright Brothers first made the Miami Valley a proving ground for powered flight.