The AI Pilot Doesn't Need Your Signal

A $3,000 jammer used to be a death sentence for a $500 FPV drone. In 2027, the drone just stops listening — and keeps flying on what it sees.

The Link Is No Longer the Kill Chain

Electronic warfare built its reputation on a simple bet: sever the radio link and the drone falls. For most of the Ukraine war that bet paid. Front-line units reported FPV loss rates to jamming well above 50 percent in contested sectors through 2024, and both sides responded the same way — more jammers, more frequencies, more fiber-optic cable.

Fiber was the tell. When your most reliable counter-jamming technology is a 20-kilometer spool of glass, the RF link fight is already lost. The next move was obvious and is now arriving: remove the link from the terminal phase entirely.

An FPV that locks onto a target visually at 2-3 kilometers out and guides itself in does not care what your jammer broadcasts. It is not receiving anything. Ukraine's use of machine-vision lock-on modules — fielded in volume through 2025 and 2026 under programs tied to the Brave1 accelerator — turned last-mile autonomy from a lab demo into a standard feature. Russia followed with its own lock-on FPVs within months, as it has with every Ukrainian adaptation.

The Compute Got Cheap Enough

The enabling fact of 2027 is a bill of materials, not a breakthrough paper.

- Edge AI modules in the class of the NVIDIA Jetson Orin Nano now deliver enough inference for real-time object tracking at prices that dropped below $250 per unit in volume, and commodity alternatives from Chinese vendors undercut that further. - Complete terminal-guidance autopilots — camera, tracker, flight controller — are quoted by Ukrainian manufacturers at $400-800 on top of a base airframe. - A fully autonomous FPV with lock-on and target recognition costs roughly $1,200-1,500, still an order of magnitude below the $60,000-plus unit cost of a Switchblade 600.

The economics have inverted against the defender in a specific way. A jammer that cost $3,000-30,000 per position protected against drones that needed a pilot. Against autonomous terminal guidance, the same expenditure protects against nothing. The money has to move somewhere else.

Map-Matching and the GPS-Free Bomber

FPVs are the loud case. The quiet one is strategic.

Long-range one-way attack drones — the Shahed-136/Geran-2 family and its derivatives — always flew most of their route on inertial navigation with satellite correction. Jam GPS and accuracy degraded, sometimes badly. The answer fielded through 2026 and maturing now is terrain and scene matching: downward cameras comparing the ground below against stored imagery, updating the inertial solution without any external signal. Russia's Kometa adaptive antenna arrays already hardened the receivers; vision-based navigation removes the receiver from the critical path.

No RF emission. No RF reception. The drone's entire sensor suite is photons it collects itself. This is the design pattern spreading across the class in 2027 — Ukraine's own long-range programs, funded under the stated ambition of millions of drones per year, are building to the same specification, and Western programs like the Replicator tranches and Anduril's Barracuda family assume GPS-denied operation from the first drawing.

Sensing the Sensor

Counter-autonomy is now a funded research race, and it attacks a different surface.

You cannot jam a camera the way you jam a radio link, but you can lie to it. Three lines of effort matter:

- Visual spoofing and decoys. Inflatable and thermal decoys are old; the new work is decoys tuned to fool machine classifiers, not human pilots. Classifiers trained on limited target sets misidentify at rates that disciplined deception can exploit. Ukraine and Russia both ramped decoy production through 2026 for exactly this reason. - Adversarial attack on the model. DARPA's Guaranteeing AI Robustness against Deception (GARD) program and its successors spent years on adversarial perturbations — patterns that cause misclassification. The fieldable version in 2027 is crude but real: paint schemes, netting, and infrared treatments that shift a vehicle out of its training distribution. - Dazzling. Directed infrared and laser energy against electro-optical seekers is mature against missiles and is being adapted against cheap drone cameras. A saturated sensor sees nothing, which is functionally a jam.

The honest assessment from public research: every one of these works sometimes, none works reliably, and the adaptation loop between classifier and countermeasure is measured in weeks.

From Links to Pixels

The EW fight is migrating off the spectrum and onto the image. That changes what an EW unit is.

- The classic EW platoon carried jammers. The 2027 version increasingly carries decoys, smoke and multispectral obscurants, dazzlers, and hardened camouflage — signature management as electronic warfare. - Detection shifts too. A silent drone radiates nothing, so passive RF cueing fails. Radars like those feeding FAAD C2, acoustic arrays, and optical detection networks carry more of the load, and they are more expensive per square kilometer covered than a spectrum analyzer. - Kinetic layers gain relative value. Coyote interceptors, APKWS rockets, and gun systems do not care whether the inbound drone is piloted or autonomous. Expect procurement to tilt further toward hard-kill as soft-kill loses the link to attack.

The implication for budgets is uncomfortable: defending against autonomous drones costs more per engagement, not less, until counter-autonomy sensing matures.

What to Watch

Six to twenty-four months out, four markers will show how fast this transition runs:

- Interceptor economics. Whether programs like Anduril's Roadrunner and next-generation Coyote variants get unit costs below $25,000 while keeping autonomous-drone kill rates high. That number decides whether defense scales. - Adversarial camouflage at scale. Any army that fields classifier-defeating paint or netting as standard issue — watch Ukraine's brigade-level adaptations first — confirms visual spoofing works outside the lab. - Swarm logic without coordination. The next step is multi-drone terminal attacks with no inter-drone radio, deconflicted onboard. A confirmed fielding, not a demo, is the signal. - Procurement language. When NATO EW requirements documents start specifying counter-EO and decoy integration rather than jamming power, the doctrine shift is official.

The last thing a jammer hears is nothing — the drone stopped listening a kilometer ago.