Somewhere inside the Pentagon, Air Force strategists want to compress a full day of simulated warfare into roughly nine seconds. In a formal Request for Information posted to the federal contracting portal SAM.gov in early 2025, the Air Force’s strategy division, known as HAF/A5, asked industry to help build what it calls the WarMatrix Ecosystem: a suite of AI-powered tools capable of running military wargame simulations up to 10,000 times faster than real time. The RFI, filed under Notice ID FA70142611232025, marks the Pentagon’s most explicit move yet toward replacing slow, labor-intensive tabletop exercises with machine-speed scenario planning.
As of May 2026, no contract has been awarded and no prototype has been publicly demonstrated. But the document itself reveals how seriously the Air Force is rethinking the way it tests force structures, logistics chains, and battle plans against adversaries whose own military modernization is accelerating.
What the Air Force is asking for
The RFI lays out an ambitious set of requirements. At 10,000 times real-time speed, planners could theoretically run thousands of variations of a single conflict scenario in the time it currently takes to complete one or two. Different troop configurations, supply routes, and adversary responses could be tested in rapid succession, giving decision-makers a far wider view of possible outcomes before committing to a force design.
Speed is only part of the ask. The document specifies that any tools developed under WarMatrix must be interoperable with existing Pentagon modeling systems and must scale across classification levels, from unclassified research environments to the highly classified planning cells where real operational decisions are made. That requirement alone represents a significant engineering challenge; defense simulation systems are notoriously siloed, and bridging classification boundaries adds layers of security and compliance overhead.
Federal procurement infrastructure documents, including security guidance published on FSD.gov, outline the encryption, access control, and data-handling rules that would govern any classified simulation data flowing through the ecosystem. Vendors tracking the opportunity can follow related acquisition updates through the GSA Interact portal.
The speed-versus-fidelity problem
The 10,000x benchmark sounds transformative, but it raises a question that military simulation researchers have studied for decades: what do you sacrifice to go that fast?
Traditional wargame simulations are slow for a reason. They attempt to model complex physics, human decision-making under stress, logistical friction, and the fog of war with high resolution. Compressing all of that processing by four orders of magnitude requires one of three things: vastly more computing power, significant simplification of the underlying models, or AI techniques like surrogate modeling, where machine-learning algorithms approximate complex calculations rather than solving them from first principles. According to the Department of Defense’s own modeling and simulation coordination office, balancing computational speed against model fidelity is one of the persistent challenges in defense simulation development.
Each approach carries tradeoffs. Brute-force computing power is expensive and energy-intensive. Simplified models risk stripping out the very friction and unpredictability that make wargames useful. Surrogate models can be fast and surprisingly accurate within their training data, but they can also fail in unexpected ways when confronted with novel scenarios, precisely the kind of scenarios planners most need to explore.
The RFI does not specify which approach the Air Force prefers, and no publicly available technical assessment compares these methods head-to-head for defense wargaming applications. That gap matters. A simulation that runs in seconds but produces unreliable outputs would not just be unhelpful; it could actively mislead the planners who depend on it to shape billion-dollar force structure decisions.
Why the push is happening now
The WarMatrix RFI did not emerge in a vacuum. The Defense Department has been investing in AI-augmented decision-making tools for years, from Project Maven’s early computer-vision work to the Joint All-Domain Command and Control (JADC2) initiative that aims to connect sensors and shooters across every branch of the military. Wargaming sits upstream of all of those efforts: it is where the Pentagon stress-tests the force designs and operational concepts that those systems are built to execute.
The traditional approach to that stress-testing has not kept pace with the threat environment. Tabletop wargames, while valuable for surfacing strategic insights and forcing senior leaders to confront hard choices, are slow, expensive, and limited in the number of scenarios they can explore. Computer-assisted simulations have improved over the past two decades, but most still run at or near real time, meaning a week-long simulated campaign takes roughly a week to execute. Against adversaries whose military research institutions are investing heavily in AI-driven simulation, that pace is increasingly seen as a strategic liability.
The 10,000x target, if achievable, would represent a qualitative shift. Instead of asking “what happens if we fight this way,” planners could ask “what happens across 5,000 variations of this fight” and get answers before the next morning’s briefing.
What has not been answered
For all its ambition, the WarMatrix RFI leaves critical questions unresolved. No budget line item tied to the program has appeared in publicly available defense spending documents. No timeline for moving from market research to a formal solicitation has been disclosed. No vendor responses have been made public, so the feasibility of meeting the stated targets at the classification levels the Pentagon requires remains unconfirmed.
Governance is another open question. The RFI states that WarMatrix should support integrated force design, but it does not spell out who would own the underlying models, who would be authorized to modify the assumptions baked into each scenario, or how disagreements over inputs would be resolved. In any simulation powerful enough to shape force structure decisions, the choices about what to model and how to represent adversary behavior are themselves strategic judgments. Without a transparent framework for making those choices, there is a risk that the tool’s speed and sophistication could obscure the subjective assumptions driving its outputs.
There are also no independent technical evaluations of the 10,000x speed claim in the public record, no congressional testimony addressing the program, and no Government Accountability Office reviews assessing its viability or risks. The evidence base, as of spring 2026, consists of the RFI and its associated procurement documents. The gap between a procurement document’s stated ambitions and a fielded system’s actual performance is often wide in defense acquisition, and WarMatrix has not yet begun to close it.
Where this leaves the Pentagon’s simulation ambitions
The WarMatrix RFI is best understood as a declaration of intent. The Air Force has told industry, in formal procurement language, that it wants to move military planning into a fundamentally faster computational environment, one where thousands of alternative futures can be explored in the time it once took to examine a handful. That alone is significant. It puts AI-driven wargaming on the official acquisition radar and invites the defense technology sector to show what is possible.
But intent is not capability. The hardest work, building simulations that are both fast enough to be transformative and accurate enough to be trusted, lies ahead. Whether WarMatrix becomes a genuine tool for force design or remains an aspirational line in a procurement document will depend on answers that no RFI can provide: how well AI-generated models hold up under adversarial scrutiny, how willing senior leaders are to trust machine-speed analysis, and whether the Pentagon can field the system before the strategic window it is designed to address has already closed.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
