Faster Simulations, Better Defense Decisions

Every system the U.S. military fields carries a line item that never appears in a budget justification book. No program office tracks it, and no auditor asks about it. It is the cost of the analysis that was never run.

In mission-critical engineering, teams face a choice they rarely say aloud: fast or accurate. A single high-fidelity simulation of a complex physical system can take days or weeks. One aircraft component can require on the order of 10,000 simulations to understand across its operating conditions, and an aircraft has millions of components that all have to work together.

Faced with such computational loads, engineers do what engineers have always done. They analyze parts instead of the whole system. They substitute estimates for full analysis. And they bury the remaining uncertainty inside an oversized safety margin.
This is simply a rational response to tools that cannot keep up. But the bill still comes due, and the Pentagon pays it in three ways:

Questions that never get asked. Slow analysis also narrows ambition. When each scenario takes hours, planners evaluate only a handful instead of thousands. Designers explore two or three variables instead of a dozen, and decisionmakers get a best guess instead of the full range of outcomes.
Space domain awareness shows what is at stake. U.S. Space Force Guardians can’t just ask where one satellite will be tomorrow; they have to ask which of thousands of objects in orbit is behaving like an adversary. If projecting a single object's path takes several minutes, that question cannot be answered at operational speed. It turns into triage over which few objects the team can afford to examine.
The instinctive answer is to buy more computing, and the DoD is doing that. But focusing on hardware alone attacks the problem from the wrong end. Much of the simulation software in defense engineering still runs on numerical methods designed around decades-old computing architectures. As a result, running these legacy methods on modern processors captures only a fraction of the performance those processors can deliver. Rewriting that math for the latest hardware they own is where the larger gains are.
Those gains are no longer hypothetical. Under a SpaceWERX contract, my team at BQP propagated a catalog of more than 3,000 space objects in under one second on a single GPU, about 250 times faster than legacy tools. The underlying model shrank from 14 million parameters to 2,000, small enough to run on edge hardware in orbit. In commercial aerospace, drone maker Modovolo used the same approach to design its propeller and motor together as one system rather than two separate parts, expanding its design space exploration from two or three design variables to more than 12 and enabling significant improvements in range and payload.
The lesson here is that the tradeoff between speed and accuracy is no longer static. Once system-level analysis is affordable, a program can justify tighter margins, catch interactions before testing, and ask the questions it used to skip. Three steps would help the DoD get there.
First, make the invisible cost visible. Milestone reviews should ask not only what was analyzed but what was not: which subsystem interactions were estimated instead of simulated, and how much margin was added to cover them. A margin no one can explain is a risk no one is managing.

Second, treat analytical capacity as capability rather than overhead. Simulation often competes with everything else for the same infrastructure dollars and loses. The ability to run system-level analysis on demand belongs in the same conversation as readiness, because it decides how quickly defense programs can design, adapt, and field.
Finally, close the gap between a successful demonstration and a program of record. Pathways like SBIR and SpaceWERX let small firms prove new approaches on real defense problems quickly. Too often, a result validated in one program never reaches the next. The department needs a clear route for proven analytical tools to move across programs and services.

Ultimately, the issue is not simply the speed of simulation, but the quality of the decisions that follow from it. When engineers can analyze more of the system, explore more scenarios, and resolve uncertainty earlier, they can design with greater confidence, identify problems before they become costly, and make better-informed decisions. Safety margins will always have a place in engineering. But they should be made with full information, not a blanket thrown over what we could not afford to calculate. The side that can see the full picture first, across every scenario rather than a single best guess, will field better systems and make faster decisions. That advantage is within reach, on hardware the DoD already owns.

Rut Lineswala is the Founder and CTO of BQP, where he has led the development of the company's core technology platform and continues to drive technology strategy and product innovation. Recognized as one of the "51 CTOs Transforming the World of Quantum Tech," he is advancing the convergence of high-performance and quantum computing, Physics-AI, and optimization for mission-critical industries. Rut also leads the Quantum Focus Group within the Digital Twin Consortium and is a frequent keynote speaker at industry, academic, and government events.