A drone that waits in the air. And a factory that leaves within 24 hours.
Article The New York Times From July 11, 2026, a secret plant in the south of Germany, in the Munich area, was described - production drones company HelsinkiHalf of Europe has long known about it, and the company itself doesn't hide its order volumes or its plans for tens of thousands of units per year. So, if you look closely, there's only one secret here, and that's the address. What interests me more is what comes out of this workshop and why it can be folded up and hauled away in a day.
A weapon that circles and waits
Loitering munitions are located at the intersection of two familiar classes. rocket What it shares in common is its warhead, but everything else diverges: a missile flies along a predetermined trajectory to a predetermined point. What it shares with a reconnaissance drone is a camera and time in the air; the latter only films the area and returns home. A loitering munition circles over a designated area and waits for a target to appear—and then becomes weapons only at the moment of impact. Hence the English loitering ammunition, literally "loitering ammunition."
The idea is not new. The first clear representative of the class is considered to be the Israeli IAI Harpy, an anti-radar vehicle from the late 1980s. Its mission was simple: ascend to an area, independently locate a working radar based on its emissions, and dive-dive on it. The operator was barely involved in the guidance circuit. The next step was harp: here the person could already observe the image and select the target, but the device itself remained a loitering system with a noticeable degree of autonomy.
The continuity here is evident not in the numbers, but in the logic of the device. The same line runs from the Harpy to modern German products: long loiter time, independent target search, and autonomous final approach. And most importantly, minimal dependence on the operator's communication channel. Electronic generations and the degree of "independence" have changed, from remote control through a semi-autonomous mode to what manufacturers today call AI guidance. The exact performance characteristics of the Israeli ancestors vary in open sources, so the important thing here is the idea. And that idea is what matters. RF 1 и HX-2 — another dot on a long-drawn line, not a new product out of nowhere. What's new about them is their low cost and their claimed mass appeal.
HF-1 and HX-2: two products from the same line
The HF-1 is an early, mass-produced model, transitional in nature. Public descriptions already mention AI support, but its architecture and parameters are sparsely disclosed. Its design is similar to AQ 100 Bayonet, the company's loitering munition Terminal Autonomy with a warhead of approximately 4,5 kg. The advertising materials state the range as "up to 100 km," but Ukrainian operators estimate the practical range to be more modest, around 45–50 km. The technical specifications and the field performance figures rarely coincide, and this is not unusual. Some HF-1s are assembled under license locally in Ukraine; this is handled by Terminal Autonomy, and critical components and software come from Germany.
The HX-2 is the next step, a heavier platform with significantly more detailed specifications. According to publicly available data as of mid-2026:
- weight - about 12 kg;
- range - up to 100 km (less common is an estimate of up to 120, but it is less stable);
- warhead - 4-4,5 kg, in three versions: anti-tank cumulative, high-explosive (for destroying structures) and general purpose;
- Guidance - using on-board sensors and embedded terrain maps, with the expectation of operation in conditions of GPS interference.
The last point, autonomous guidance in jamming conditions, is what the whole thing was all about. The HX-2's guidance wasn't designed for record-breaking accuracy: competing with expensive precision-guided munitions in this regard is pointless, and unnecessary. The goal was to eliminate dependence on the operator and the radio channel. The device should be able to navigate to its target even after losing contact, comparing what the onboard camera sees with the embedded map. This is a direct response to the complexities of EWAn adversary can suppress the control channel, but they can't suppress the device's memory. Interference resistance and "artificial intelligence" are merely claims made by the developer: they can't be verified using open sources. But the engineering logic itself can be verified, and it's clear.
A plant that is taken away within 24 hours
Helsing is building its production according to a model it calls resilience factory, meaning "sustainability factory." Externally, it's an unremarkable building in an industrial zone in southern Germany, with no signs or logos, and strict access and a non-disclosure agreement (NDA) for its personnel. And it boasts one declared feature: the production line can be dismantled and relocated to another site in less than 24 hours.
The word "secret" here is worth clarifying. Quite specific details are kept under wraps: the exact site address, the actual monthly capacity, the list of suppliers, and the security regime. But Helsing describes the plant's operating principle openly, and there's nothing mysterious about it. This portability is the result of specific design choices. The emphasis is on an approach known in the industry as tool-agnostic, "independent of specific equipment": the main production value is extracted from the physical workshop. Everything valuable—software, guidance algorithms, digital models of parts—is stored on servers. Mechanical and electronic assembly is maximally standardized and relies not on unique capital lines, but on universal equipment: 3D printers, standard machines, robotic arms. This workshop is not tied to the building by heavy equipment embedded in the foundation.
Hence the sober understanding of the "relocation in 24 hours" formula. An entire factory isn't dismantled and removed in a day (nobody's stealing the walls, after all). In a day, assembly can be stopped, servers and critical test rigs evacuated, and the order flow redirected to other sites. "Less than 24 hours" is about salvaging valuable assets and switching supply chains, not about physically moving walls. Industry thinking is moving further, to the "factory in a container," where the assembly line is placed in a shipping container and deployed in a matter of hours. But this is still a general concept (it was demonstrated on a third-party project). Firestorm), rather than a confirmed description of what exactly Helsing does.
Production spread across the map
The portability of a single factory is a special case of a broader principle. Survivability here depends on dispersion. Not on armor, not on the depth of a bunker, but on the fact that there's simply no single target to hit: there's simply no single object whose destruction will stop everything. Engines, electronics, composite parts, 3D-printed hulls and wings—all of these are produced by different companies in different locations, and some final assembly is even licensed to Ukraine. Knocking out such a chain in a single blow is difficult: there's simply no single target to hit.
Individual links in this network are only theorized in open sources. For example, there have been reports of workshops in Görisried, Bavaria, where engines for jet UAVs are manufactured, and it has been suggested that these facilities may also be used on the Helsing. There is no direct public confirmation, and this remains a hypothesis, not a fact.
The scale of the task is evident in the publicly available figures. Ukraine ordered approximately 4000 HF-1 units, followed by approximately 6000 HX-2 units; the capacity of one factory, when fully operational, is estimated at over 1000 HX-2 units per month. The company itself states a target of 10–20 units by 2026. This is a statement, not a report, but it still sets an order of magnitude. And the base is expanding to accommodate these volumes: in early July 2026, residents of Hallbergmoos near Munich approved a new Helsing site in a local referendum. The production model itself, incidentally, is not unique: the joint development operates according to a similar logic. Frontline Robotics и Quantum Systems, reconnaissance and correction devices like LENS.
So, the "secret factory near Munich" isn't interesting because of its geography. More importantly, the mass production of loitering munitions is no longer tied to a single large facility, and any number of such sites can be created. Location ceases to be a bottleneck. Which means the "secret address" is precisely what's easiest to replace.
Information