Join War on the Rocks and gain access to content trusted by policymakers, military leaders, and strategic thinkers worldwide.
Despite the hype, drone swarming doesn’t exist yet. That’s because the U.S. Department of Defense has been focused on platform capability inputs like hardware, manufacturing, and GPS, while so far neglecting the architectural question of how drones are supposed to work together. The “swarming” on offer today consists of robotic maneuver en masse. However, a transformative strategic leap forward is within reach: resilient, collaborative, autonomous problem solving at machine speed, without any single point of failure. Achieving that vision of swarming will require shifting acquisition priorities toward distributed systems infrastructure, not just quantity or quality of platforms.
What Swarming Is and Isn’t
We are experts in distributed systems, collaborative autonomy, and military modernization strategy. We’ve both had security clearances. We’ve interviewed at least a hundred subject matter experts across the public and private sectors. We have seen no evidence of true drone swarming, anywhere. Those mind-blowing Chinese drone light shows? Not swarming. Leader-follower autonomous teaming? Not swarming. One hundred drones operated by a single person? Not swarming. A genuine swarm is singular, not plural. It is overwhelming not just in its scale, but in its unity and resilience. It adapts intelligently to changing circumstances at machine speed. It achieves more than the sum of its parts. It’s a distributed system.
The U.S. defense industry has so far failed to deliver distributed systems for useful, resilient, collaborative swarming behavior among autonomous platforms. By calling groups of their product “swarms” or characterizing their behavior as “swarming,” defense contractors have robbed that concept of meaning, confused their customers, and blunted the demand signal that should be fostering a breakthrough capability. If no competitor ever developed a true distributed swarm, this wouldn’t matter. But we know that’s not the case.
Swarming is the next step in a historic evolution of winning approaches to warfare. Military operations have advanced in sophistication over time, from unstructured melee through concentrated mass and mechanized maneuver. Starting in the late 20th century, decisive weaponry and infrastructure froze conflict and forced it to the unfortunate margins where maneuver still reigned. As the 21st century unfolds, surprisingly crude tools and tactics have grown relatively more effective against expensive infrastructure, and in 2025 arguably a new winning approach is to field numerous small, autonomous, and inexpensive platforms.
Every step in warfare development improves on three core factors that determine which side in a conflict will succumb to defeat. The first is moving violence further away from the warfighter, so that fewer die, and the people maintain their will to fight. The second is wise investment, so that resources last longer and the people maintain their ability to fight. The third is destructive effectiveness, so that an adversary experiences high costs and their people lose the will or ability to fight. Smaller improvements in weaponry, force protection, and tactics are decisive only when opposing forces are operating at the same developmental stage.
Swarming represents the next developmental stage, right around the corner technologically but still out of reach in practice. The deployment of numerous small, autonomous, and inexpensive platforms has accomplished great things over the past few years, but that has amounted to the combination and automation of earlier developmental stages — mass and maneuver. The platforms are individuals, plural in the sense that each requires remote control by a pilot or individual onboard intelligence.
Of course, robotic maneuver en masse is important. The attrition of autonomous devices instead of people is more acceptable because it does not result in human deaths. The devices can hold a surprising amount of the enemy’s blood and treasure at risk. And, of course, it’s disruptive to a U.S. military culture that has emphasized presence and technical sophistication. But robotic maneuver en masse is nothing more than a machine version of what humans would do, on foot or in manned platforms. It’s only a gesture in the direction of a step change.
Each true step change in warfare development doesn’t just improve on what came before it — it defeats it. Mass beats melee through attrition and organization. It destroys effectively with greater numbers to expend before exhaustion. Maneuver beats mass by attacking weak points and creating surprise and confusion. It creates a high destructive return on investment and often circumvents massed forces. But an uncoordinated fleet of small autonomous devices can only beat maneuver through greater mass and faster maneuver.
Short of nuclear weapons, what would defeat a large group of autonomous devices that can maneuver exceptionally well? A swarm. It would have to be a distributed, intelligent system that dynamically adapts to changing circumstances in real time, without reliance on a central controller, a designated leader drone, or even an internet connection. The military force that achieves true swarming first will not just win battles — it could overwhelm entire systems and strategies that have not anticipated coordinated and adaptive attack, self-healing networks, or swarm intelligence. And that is possible now.
Imagine a battle between two drone swarms, Red and Blue. On the Blue side, 500 drones maneuver in perfect sync under the control of a single human operator, executing pre-scripted formations. Each drone makes decisions autonomously to target a nearby enemy and avoid collisions. The group resembles a machine-speed marching band, almost blotting out the sun with their numbers and tight coordination.
By contrast, the Red side is different. Its own 500 drones collaboratively assess the situation and assign targets on a one-to-one ratio, conserving ammunition and minimizing unintended damage. The primary actor is not any individual drone, but the group as a whole, because the group leverages swarm intelligence to act as one. It organizes itself like a colony of ants, a hive of bees, or even a slime mold, achieving the “hive mind” associated with science fiction characters like the Borg of Star Trek or the Buggers in Ender’s Game. Its common operating picture stays in sync over ad hoc short-range connections, even as communications back to human operators and headquarters are jammed.
While the Red side is demonstrating swarming capability, the Blue side falls short. Instead, the Blue side is maneuvering robotics en masse, allowing one human to operate multiple drones simultaneously in a one-to-many model or imposing hierarchy in a leader-follower model. When the situation changes unexpectedly, the Blue drones have no way to adapt at machine speed, all at once, in real time, the way the Red swarm can. And, by maintaining centralized control, the Blue side suffers from a vulnerable single point of failure.
Swarming is such an important strategic advancement that to conflate it with intermediate improvements in massing and maneuver is misleading and dangerous. This is not just a linguistic or hypothetical concern — now that the term has been so diluted, the demand signal for true swarming is too unclear for industry to act on. Based on three years of interviews with subject matter experts across the public, private, and academic sector, we assess that the U.S. defense technology industry is uniformly driving toward the Blue approach: massing and maneuvering robots as mere surrogates for humans, not true swarming.
By contrast, when policymakers hear the term “swarming,” they tend to envision the Borg, a hive of ants, or something like starling murmuration. They’re anticipating a future swarming capability like the Red side. Policymakers don’t realize that they need to communicate all the Blue approaches and vulnerabilities they don’t want. If that distinction remains unclear, then policymakers lose the opportunity to send the demand signal they intended, and the U.S. military loses the opportunity to field the best fighting force.
How is Swarming Possible?
Achieving true swarming requires incorporating distributed systems infrastructure that can provide a common operating picture and spread out the locus of control across the swarm. This is the characteristic that makes a swarm strategically game-changing: distribution of its situational awareness and control across many synchronized leader nodes, instead of one. The swarm creates a consistent, self-healing common operating picture by periodically synchronizing the state of the system and jointly deciding what to do next. When done well, this approach maintains coordination despite the expectation that individual nodes will fail and network communications will falter. Distributed systems are well understood in hyperscale data center engineering. Emerging research (especially in China) is beginning to apply distributed architectures to autonomous systems in the physical world over wireless connections.
The computer science and mathematics required to accomplish that can be very counterintuitive for defense integration teams rooted in AI, robotics, or automation. Those fields of expertise are organized around centralized systems that assume a single central processor absorbs all sensor inputs and directs the remaining parts of the system. This is the foundation for one-to-many and leader-follower approaches to collaborative autonomy. It is more easily scalable than a distributed system, but it’s also more easily defeated.
Why AI Isn’t Enough
AI cannot create swarming because intelligence and architecture are not the same thing. A group of highly intelligent drones, each powered by advanced AI, will not swarm unless they can agree — continuously and reliably — on a shared picture of the world. That agreement is the domain of distributed systems. AI can enhance swarming, but it cannot enable swarming on its own. In short, AI can make many platforms smarter, but only distributed systems can make a swarm coherent.
Recommendations for Acquisition Policymakers
Acquisitions and research, development, testing, and evaluation leaders should include the language of distributed systems in any initiatives that they intend to involve true swarming. For example: “The solution should leverage a consensus-based state management and data safety infrastructure layer to enable secure, distributed operations at scale.”
This is necessary because solicitations and requests for proposals almost always conflate problems that distributed systems solve with different problems that have network, application, or AI solutions. Almost every public call for collaborative autonomy “swarming” capabilities from the Department of Defense or the services describes goals in alignment with a distributed system, but only requests network or application solutions and leaves system characteristics vague. Because of this, distributed systems providers don’t receive a clear demand signal and don’t engage. Those that do engage struggle to communicate the difference their solution would make, and they eventually give up.
As an example of this confusion, all 25 U.S. Army officers and civilian modernization professionals we interviewed for this research cited “bandwidth constraints” that prevent a resilient and consistent common operating picture at the edge. However, bandwidth is a network constraint. Increased bandwidth would not enable a local common operating picture. It would simply enable local devices to leverage a remote common operating picture via hyperscale cloud providers. Resilient and scalable distributed systems can and should be implemented locally, without that dependence on a cloud connection. End users on the ground don’t need to understand this distinction, but it’s important that modernization and acquisition professionals do.
More broadly, to avoid fielding vulnerable and ineffective distributed military systems, the Department of Defense should clearly articulate initiatives to research, test, and procure distributed systems infrastructure independent of cloud provider dependencies. This is necessary because relying on distributed systems infrastructure in the cloud requires more bandwidth than deployed operators can count on. It also constitutes a single point of failure. Program Executive Offices and requirements developers should stipulate that distributed software infrastructure be cloud-independent and locally self-contained.
The field of research into cloud-independent, local distributed systems is shockingly underdeveloped, and the demand signal for it is vague and subject to confused messaging. But that capability is a foundational technological prerequisite to the U.S. military’s modern warfighting approaches. Indeed, each of the services’ official operating concepts are based on the assumption that some distributed infrastructure layer will enable dispersed groups of platforms to coordinate at machine speed.
That’s because swarming — collaborative, autonomous, machine-speed adaptation to changing circumstances — represents a step change in military operational development, and the kind of capability that will win future wars. Anyone who agrees with the premise underlying these concepts should be asking their congressmen, acquisition executives, and political leadership for a plan to build “cloud-independent, resilient distributed systems” into collaborative autonomy and every mission critical piece of defense infrastructure.
Nevertheless, without a clear understanding of the current and desired future capabilities, the United States risks expending immense resources designing systems and strategies that are misaligned, vulnerable, and unprepared for future challenges. Swarming is the next step in a momentous progression of operational capabilities, and it is within reach for whichever side gets serious about it.
Emma Bates is the Founder and Chief Executive Officer of Cachai, a pre-seed software startup licensing self-contained distributed systems infrastructure for national security systems. She previously worked at the Defense Innovation Unit, U.S. Army Futures Command, and the Center for Strategic & International Studies.
S. Ryan Quick is the founder of professional services firm Providentia Worldwide. He has spent decades solving the hardest problems at the intersection of distributed systems, hyperscale, Web3, distributed cybersecurity, and data safety for clients like Oak Ridge National Lab, Samsung, L3Harris, Paypal, and Ebay.
Image: Midjourney
