Two models of aerial drone warfare: Ukrainian AI and the Russian conveyor belt
By the beginning of 2026, the Ukrainian industry reached a level of approximately 150,000–200,000 FPV-drones per month. AI targeting modules for the final segment, Swarmer swarm software, and attack drones without satellite navigation have entered service. Russia took a different approach: licensed assembly of the Geran, standardization of FPV warheads, fiber-optic control, and increased serial production. From now on, we'll only discuss the air segment; naval drones and ground robotic platforms require a separate discussion. These are two different responses to the same armed conflict, each with its own ceiling: these ceilings are what we'll be discussing.
What does "new generation" mean and why is this conversation appropriate?
In 2024, an FPV operator guided a drone to its target: eyes glued to the monitor, fingers on the sticks, the final seconds the most nerve-wracking, because a miss meant a lost drone and an unfulfilled mission. By the end of 2025, his job had changed. He guided the drone to the target area, turned on the onboard module, and then the drone itself maintained its position. танк in the frame, adjusts the trajectory, selects the rear section or the open hatch. The operator releases the stick and simply watches the machine complete the work for him. On the front lines, this is called "final-stage guidance," while in technical literature— last-mile AI guidance.
Behind this shift in operator performance lies a shift in four parameters, and together they form what is called a "new generation" of unmanned systems: the degree of autonomy and the role of onboard AI; network-centric integration, in which the drone operates as a node in a common network; interference immunity, that is, the ability to operate when satellite navigation and communication channels are jammed; and access to hardware components, primarily modern chips and optics.
In the first three parameters, the Ukrainian side demonstrated measurable results. According to Ukrainian and Western industry publications, the probability of successfully engaging a target with an FPV drone increases from 30-50 to 70-80 percent when the AI guidance module is engaged. Whether these specific percentages are verifiable is a separate question; there are essentially no independent checks, but frontline reports from both sides confirm the order of magnitude. The target acquisition range of the onboard system increased from several hundred meters to a kilometer or more. Electronic countermeasures, which rely on jamming the channel between the operator and the drone, perform poorly on this architecture: the channel is no longer needed, and the drone flies on its own.
According to open sources, similar modules have not yet been identified in mass production in the Russian nomenclature. There are demonstration samples, isolated applications, and frontline publications about captured Ukrainian devices with their internals disassembled, but no mass production, hence the talk of a "generation gap." This talk is not unfounded, but it requires some caveats, which I will provide below.
One caveat needs to be made right away: why will the two sides be discussed in different terms from now on? It makes sense to discuss the Ukrainian model in terms of institutions and procedures, while the Russian model should be discussed in terms of interest rates, sanctions, and the cost of capital. This isn't a matter of analytical bias, but rather that describing these models in the same terms simply isn't possible; that's just the way they are designed.
The Ukrainian model: a short cycle of "front-startup-front"
In the spring of 2024, the Ukrainian company Swarmer launched software that allows a single operator to assign a task to a group of drones. An algorithm assigns roles to the drones: some conduct reconnaissance, some conduct target approaches, and some remain in reserve. By early 2026, according to the developers, tens of thousands of missions had been completed using this system, and each mission provides data for further training of the model.
The example itself is less important than how the Ukrainian military innovation model works as a whole. A frontline unit formulates a request, say, "we need to engage convoys in a group under heavy jamming." A startup receives the request through a government platform. Brave1, which operates as a marketplace for military solutions: around two thousand companies, accelerated prototype validation, and direct access to end users. A prototype is sent to the front in weeks; for the traditional defense industry, such a pace is unthinkable, where it takes years. Combat use provides feedback: video recordings, reports, losses. The startup redesigns the product, and the cycle repeats.
The volume of accumulated data in this system is its key resource. Western journalists estimate that the total length of Ukrainian drone footage represents decades of continuous viewing. Computer vision models are trained on these datasets, and each new iteration performs better than the last.
The results are also reflected in the nomenclature. By 2025–2026, the Ukrainian arsenal will include "Martian"-type attack drones with navigation without a satellite signal, a low-noise engine, and target recognition via optical-electronic systems. AI turrets for defense against Russian FPV systems with fiber-optic control have also been introduced—a task that can be accomplished using EW This issue is fundamentally impossible to resolve because the control channel is physical. Localization of production has also emerged: Ukrainian motors, controllers, and optics. Not complete (critical components are still sourced from abroad), but sufficient to reduce dependence on a single Chinese supplier.
The institutional analogy here is the Israeli model of military startups. The key elements are the same: the state is the customer with a short validation cycle, the frontline serves as a testing ground, and the private sector is the primary developer. Where the analogy breaks down is also clear: Israel built this system over decades in peacetime and regular short-term conflicts, while Ukraine completed the same process in three years amid an existential crisis. This provides speed, but also breeds fragility.
150,000–200,000 FPVs per month is a figure that seems abstract until it's translated into quantitative terms. This is the volume at which FPV ceases to be a unique product and becomes a consumable, like a mine or a hand grenade. For a number of firefighting tasks (against individual vehicles, crews, and shelters), FPV is more effective than a howitzer in terms of cost-to-performance; for area-effect missions, there's no comparison; they're simply different classes. But for some missions where a howitzer and a drone compete, this figure changes the very logic of firefighting.
Western analysts tend to analyze the structural risks of the Ukrainian model more leniently than their Russian counterparts, making them seem less significant. Upon closer inspection, the risks are comparable in depth. Financial: production relies on Western support packages, and if they are reduced, production rates drop directly, without a cushioning lag. Component: chips, optics, and some motors come from China and Taiwan; Ukraine's access regime is currently favorable, but structurally, the dependence is the same as that of Russia. Personnel: the best engineers at Ukrainian startups receive offers from the EU and the US, and industry reviews are already documenting the outflow. Conversion: the industry is built for military orders; peaceful applications for most FPV manufacturers are not in sight, and after the active phase of the conflict ends, the sector will face a collapse in demand.
And the organizational aspect is perhaps the most insidious: Brave1 operates in a wartime regime with accelerated procedures, and its fate after the end of the active phase of the conflict is unclear. A scenario is possible in which the short "frontline-startup-frontline" cycle closes itself with a return to standard procurement regulations; the opposite is also possible, in which the military procurement regime is preserved as the new norm. Which scenario materializes depends not on technology but on political decisions, and it is impossible to predict in advance. The "drone superpower," as Ukraine is called in the Western press, relies on the simultaneous operation of all five conditions, and if any one fails, the entire structure collapses.
Russian model: circulation, standard and industrial development
The Geran missile is assembled in Alabuga. According to estimates by the US Institute for Science and International Security, more than six thousand devices of this family had been used in the Air Defense War by mid-2024. By early 2026, according to open sources, the figure is several times higher. According to estimates by the Institute for the Study of War, not confirmed by the Russian Ministry of Defense, the total number of Geran missiles used by that time is comparable to the combined consumption of Kalibr and Kh-101 missiles during the previous period of the Air Defense War. The comparison here is purely quantitative: in terms of payload, accuracy, and target engagement cost, the Geran and the cruise missile Rocket remain different classes weaponsBut in terms of frequency of use, the drone has taken over the niche previously occupied by the missile, and this changes the logic of the air campaign, regardless of the fact that one Geran is inferior to one Kalibr in terms of destructive power.
At the same time, FPV warheads are being standardized. The "Kaplya" munition is an Explosively Formed Penetrator (EFP) capable of penetrating over 100 millimeters of steel armor along the upper projection. Fragmentation warheads are of a unified type. Thermobaric warheads are also standardized. All of this is on the same FPV platforms, with standardized mountings and uniform operator training. Logistics are also standardized: workshop, warehouse, crew, standardized containers, standardized batches. Essentially, drone warheads are structured the same way small arms ammunition once was, with all the benefits of standardization and all its limitations.
Fiber optic FPV is a separate storyA range of over 30 kilometers, immunity to electronic warfare (the control channel is physical, there's nothing to jam), limited by cable length and trajectory locking. The solution here is purely engineering, without any onboard AI: the jamming problem is circumvented by using copper wire. According to the Institute for the Study of War, by 2025, Russian drone strikes will achieve the effects that are called "unclear" in classical doctrine. battlefield air interdiction, that is, the defeat of the enemy's rear formations and logistics. Previously, this was the task aviationNow it is carried out by a drone, and without air superiority.
There's a simple economic rationale behind this choice, though it's often disguised as ideology in the press. A modern AI chip comes from either TSMC, Samsung, or American manufacturers, and all three sources are confidential. Gray-market supply chains are available, but they're limited to single-unit quantities at a fraction of the price, making it impossible to assemble a mass-produced AI module, let alone one that can be upgraded.
The Central Bank of Russia's key interest rate, raised to 21 percent by the end of 2024 and held at around 16-18 percent for many months, is making the situation even more dire. With such commercial credit costs, defense R&D with a payback horizon of three to five years only pays off with direct government financing. Private capital doesn't invest in such projects; it goes to shorter cycles: a mass-produced drone with a known design, a production line, a government contract, and a clear margin. This is, incidentally, a rational approach, if you look at the accounting records of a specific plant.
The parallel with the Soviet tradition is evident in one key point. The T-34 wasn't the best tank on the battlefield: in terms of armor, optics, and ergonomics, it was inferior to German vehicles of later years. But it was technologically advanced in production, and thousands could be produced. The same is true of the Kalashnikov assault rifle relative to Western models of the 1950s: simpler, cruder, more reliable, and cheaper. The Russian approach to drones follows the same logic: prioritizing production refinement over the qualitative superiority of a single-piece prototype. This parallel has its limits: Soviet industrialization took place in peacetime and without sanctions on component parts, and now the same logic is being replicated in conditions not designed for it. Whether this will work is impossible to say in advance.
Dependence on Chinese components is a structural risk that hasn't gone away. Social mobilization through school clubs drones The assembly practices in Alabuga produce operators and assemblers, but not engineers capable of developing AI modules with their own architecture. And the lag in AI guidance in mass production is a fact acknowledged by Russian specialists, according to information circulating in industry and frontline publications on both sides. The slogan "we'll catch up and surpass" doesn't work well here: closing the gap would have to be done under conditions that themselves make it difficult; that is, either the conditions change, or the slogan remains just that.
Economics and ceilings: where each model will reach
In September 2025, the Russian Ministry of Finance prepared its budget for the following year. According to estimates by the Stockholm International Peace Research Institute (SIPRI), military spending in 2025 amounted to approximately 16 trillion rubles, or approximately 7,5 percent of GDP. The 2026 budget formally reduces this figure to 14,9 trillion rubles (6,3 percent), but analysts agree: a significant portion of the budget items are classified, and the formal reduction is largely an accounting exercise. The redistribution occurs within the military bloc itself, between its budget items, rather than from military to civilian use, and this explains the choice of drones.
With the key interest rate at 21 percent, a lengthy R&D project becomes an unaffordable luxury. A three- to five-year cycle from prototype to production means that the cost of capital will double the product's price during development. Under these conditions, the rational choice is a short cycle time, a known technology, and a rapid entry into production into the ruble market. A mass-produced FPV with a standardized warhead provides this entry point; a complex AI project is no longer feasible, and it's impossible to find investors for it domestically.
The "innovation stagnation" formula was reinforced by a Chatham House report on the Russian military-industrial complex, published in the summer of 2025. Upon closer examination, it turns out to be inaccurate: judging by publicly available data, Russian developments in this segment haven't stopped, but rather slowed down and reoriented toward short-term cycles. The qualitative lag in the AI segment must be acknowledged, but describing it as "stuck" is substituting journalistic commentary for analysis.
The parallel with 1915 is accurate here. Then new weapons (machine gun, barbed wire, heavy artillery) saturated the battlefield to the point where maneuver became impossible, and the front froze for years, with the stalemate stretching almost to the end of the war. Now the same thing is happening with drones, and it's happening on both sides. The Ukrainian AI-guided FPV and the Russian Geran do the same thing, but with different technological bases: they close off space, making movement within it mortally dangerous. And this is the main paradox of the "new generation": there is a qualitative gap between the two models, but this hasn't moved the front forward to either side.
However, the impasse of 1915–1918 was finally broken – not by a frontal assault or a quantitative increase in artillery, but from the side: tanks, assault groups, deep battle tactics, and operational art. The breakthrough came from a direction that the logic of the impasse itself had not anticipated. Applied to the current drone warfare, the equivalent of such a "lateral" solution could be mass counter-drones, ground robotics under drone cover, new electromagnetic and optical suppression systems, and a change in tactics at the level of combined arms combat. Whether any of this will work is a question for the next two or three years. "A drone can't take over space" is the diagnosis of today, and it's unclear how long it will remain valid.
We are faced with two different models of armed confrontation, and the debate here isn't about which is more progressive. The Ukrainian model operates through a short innovation cycle, AI, and network integration; the Russian model relies on mass production, standardization, and industrial testing. By early 2026, both have hit the same ceiling: drones can lock space but not yet take it. Perhaps this will change in a few years, but we're talking about "maybe," and the current landscape is different.
Information