6 minute read

TL;DR: The interceptor gets the headlines, but the data chain decides the drone fight. Look at where the Pentagon actually put its money and authority this year: a counter-UAS task force built around speed and command and control, a $20 billion contract vehicle whose first order was pure software, and a fire control recompete on a two year cycle. Detection, classification, and track identity are the fight. The effector is the easy part.

I have watched counter-UAS systems succeed and fail on open air ranges for years, and the pattern is remarkably consistent. When a system fails, it is almost never because the interceptor missed. It is because the system did not know what it was looking at in time to take the shot. That is a data problem, and the evidence from the last eighteen months says the Department is finally treating it like one.

Follow the Money, Then Follow the Data

Start with the institutional moves. In August 2025 the Secretary disestablished the Joint Counter-small UAS Office and stood up Joint Interagency Task Force 401, reporting directly to the Deputy Secretary, with acquisition authority up to $50 million per effort and a 36 month sunset review. The fix was not a new missile. It was authority, speed, and plumbing.

Then follow the contracts. The Army’s marquee counter-UAS award this March was an enterprise agreement with Anduril worth up to $20 billion over ten years, consolidating more than 120 separate contracts around the Lattice software platform. The first task order under that vehicle was $87 million of pure command and control. Not one interceptor. And when the Army replaced its legacy FAAD fire control system last October, the program office promised recompetes on roughly a two year cycle. The C2 layer is being treated as the load bearing capability, continuously competed like software, while effectors become swappable endpoints.

Even the vendor selling both agrees. Anduril’s own managing director called detection and identification the fundamental challenge of counter-UAS. When the company with the biggest effector ambitions in the business says the hard part is the data layer, believe it.

What the Red Sea Bill Really Says

Over fifteen months of Red Sea operations, the Navy fired roughly 120 SM-2s, 80 SM-6s, about 20 ESSMs and SM-3s, and 160 five inch rounds against some 380 Houthi threats. Some of those two million dollar missiles were engaging drones that cost about as much as a used pickup truck. The cost exchange ran 400 to 1 or worse.

The usual takeaway is magazine depth, and that is real. But look closer and it is a classification problem wearing a magazine depth costume. When the combat system cannot confidently tell a cheap drone from a cruise missile in the seconds available, doctrine forces the expensive answer every single time. Nobody gambles a destroyer on the hope that the inbound track is the five thousand dollar kind. Better discrimination, earlier, is what unlocks the cheaper shot.

The Lesson from Ukraine: Cheap Sensors, Smart Fusion

Ukraine’s Sky Fortress network is the cleanest proof I have seen. Roughly 14,000 acoustic sensors, each costing hundreds of dollars rather than millions, fused with radar and pushed to mobile fire teams on tablets. That network now carries a meaningful share of the national air picture and helped intercept 80 of 84 inbound one way attack drones in a single documented raid. Operators train in about six hours. When Russia altered the sound signature of its Shaheds, classifier accuracy dropped only about three percent. Lithuania is fielding the system this year, the first combat proven Ukrainian counter-UAS transfer to NATO.

The counterexample from the same war is just as instructive. Researchers at RUSI found that both sides have shot down large proportions of their own drones because detection, classification, and friend or foe identification fail at scale. Fratricide is a pure data chain failure. There is no effector deficiency anywhere in that picture.

The Homeland Problem Is a Data Problem Too

The commander of NORTHCOM told the Senate there were 350 drone detections across roughly 100 US installations in a single year, and that the primary concern is detection and surveillance, not attack. Incursions rose more than 80 percent the following year. Only about half of US bases even hold the legal authority to defeat a drone over their own airspace, and every engagement decision at home starts with the same question the Red Sea combat systems face: what exactly is that thing? Copenhagen’s airport closed for nearly four hours last September over two or three drones that police could never even locate the operators of. The gap on display was situational awareness, start to finish.

What the Range Keeps Teaching Us

At the joint counter-drone swarm demonstration at Yuma Proving Ground, with more than 40 drones converging per session against nine competing systems, the acquisition lead’s verdict was blunt: no single characteristic or capability, kinetic or otherwise, could defeat that profile by itself. The physics explain why. A quadcopter’s radar cross section looks like a bird. RF sensors are blind to a drone that is not emitting. Cameras cannot search a volume of sky without something cueing them first. Single sensor false alarm rates in cluttered environments can exceed a third of all tracks. Fusion is not a nice to have. It is the product.

The same lesson shows up when systems leave the lab. The Army’s 50 kilowatt Stryker mounted laser struggled in its Middle East deployment, and the program’s own leadership acknowledged that results from the lab and test ranges were very different from the tactical environment. Effectors that demo well can still fail the environment. A test community that only grades the shot, and not the picture that cued it, is grading the wrong thing.

The Autonomy Gate

Here is where this stops being an efficiency argument and becomes a safety one. The Pentagon validated an automated sense and shoot counter-drone capability this June, while the task force insists a human retains lethal force authority. Both of those things can only coexist on top of a trusted, fused track picture. You cannot delegate any part of an engagement decision to a machine that is fed by an unfused, false alarm prone sensor picture. As autonomy creeps into the kill chain, the quality of the fused data is no longer just the targeting input. It is the safety case.

So What

• If you buy: fund the data layer first and put it on a software recompete cadence, the way the Army is doing with fire control. Treat effectors as swappable endpoints and interrogate any contract ceiling that headlines look like capability. The $20 billion Lattice vehicle had no money attached at award. The task orders are where reality lives.
• If you test: instrument the discrimination layer, not just the intercept. Track probability of correct classification, time to identity, and false track rates as first class test measures, in clutter, against signature changes, at swarm scale. Define what fielded means before you report it.
• If you operate: train for the picture. Ukraine gets six hour operators onto a fused air defense network. Our fusion cells should be staffed and drilled like fire control teams, because that is what they are.

Interceptors matter. Magazines matter. But the side that wins the drone fight is the side that knows what it is looking at first. Buy the picture, test the picture, and the shot takes care of itself.

Tags: AIcUASUAS