Most conversations about autonomous warfare start with the aircraft. It’s the part that photographs well and the part closest to becoming a commodity. Building a cheap flying machine stopped being difficult some time ago, and the war in Ukraine has established that these things are consumable. Airframes will get cheaper and more interchangeable, the way most manufactured hardware eventually does.
What won’t commoditize is the software that turns a hundred separate machines into something that behaves like one. That’s where the defensible businesses are likely to sit, and it rests on three capabilities.
You don’t need to be able to build any of them. You do need to be able to tell the difference between a company that has one and a company with a persuasive demo.
1. Distributed Autonomy
Each unit works out its own next move rather than waiting for instructions from an operator or a central computer. It reads its own situation, notices what its neighbors are doing, and acts.
Two things follow, and both are commercially significant. The first is that one person can supervise a hundred platforms instead of flying one, which is the entire economic argument for swarms; the cost per effect collapses only if you don’t need a pilot per aircraft. The second is that there’s no headquarters to destroy. Centralized control hands the adversary a single high value target; distributed control doesn’t offer one.
The clearest evidence that this has become a real product category came in early 2026, when Anduril’s YFQ-44A flew a single sortie running Shield AI’s Hivemind, then switched mid-flight to Anduril’s own Lattice and repeated the same test points. Two competing autonomy stacks, one airframe, swapped in the air. That is what it looks like when software stops being a feature of the aircraft and becomes something with its own price tag.
2. Resilient Mesh Networking
Every unit relays traffic for every other unit. When one drops off, the network reroutes around the gap by itself. Nobody reconfigures anything.
This has gone from research topic to procurement priority faster than anything else on this list, for the obvious reason: the electromagnetic environment in Ukraine has been hostile enough that a system which loses its link is simply a system that doesn’t work. Resilience stopped being a spec-sheet line and became the selection criterion.
a system which loses its link is simply a system that doesn’t work.
The capital markets noticed before most generalist investors did. Motorola Solutions paid $4.4 billion to acquire Silvus, a mesh networking company. Persistent Systems booked an $87.5 million Army order for networking hardware feeding the service’s next-generation command-and-control prototype. Neither company builds an aircraft. Both are being valued as though the connective layer is the durable asset, which is this article’s thesis, restated in cash.
3. Collaborative Perception
Any single platform catches a fragment. One picks up a radar emission. Another has thermal. A third is running vision. Individually these are weak signals a competent operator would ignore. Combined, they’re a detection.
This is the least discussed of the three and possibly the most valuable, because of how the economics work. Every platform you add improves every platform already in the field; more angles, more sensing types, better coverage. That’s closer to a network effect than to anything normally found in defense hardware, where the hundredth unit is usually worth exactly what the first one was. It also raises the cost of deception considerably: fooling four different kinds of sensor simultaneously is a much harder problem than fooling one.
Beyond Drones
None of this is specific to aircraft. The same three capabilities run ground vehicles, surface and underwater vessels, logistics robots, and satellite constellations. The physics change; the coordination problem doesn’t. A company that solves it for one domain has a credible path into the others, which is a materially different growth story than a firm that manufactures one platform well.
Which leads to the uncomfortable question worth putting to any company in this space: which layer are you actually selling, and does it survive being separated from your hardware? A number of firms currently valued as autonomy companies are, on inspection, manufacturers with software attached. The distinction won’t matter much this year. It will matter enormously by the time procurement cycles catch up.
