Military AI Is an Industrial Contest

Military AI Is an Industrial Contest

Ukraine and Russia each produce roughly four million drones a year, and China probably more than double that, US Army Secretary Daniel Driscoll told Reuters in November. His own service buys about 50,000 a year. It’s an industrial rather than a technological divide. The contest over artificial intelligence is unfolding not in a laboratory, but in a factory and in a battle zone and the states that make the most of the technology and the experience by rapidly building inexpensive prototypes, mass producing them and incorporating feedback into designs within a few weeks are the ones that will determine how the next wars are fought and how crises will be handled.

The loop is the advantage

The secret of Ukraine is its cycle. Software and hardware are changed 3 to 4 times each year, and the lessons learned on the field are passed on to the next production run in days or weeks. The pattern is demonstrated in jamming. In 2024, a Ukrainian official told Reuters that while most FPV strike rates were down to 30 to 50 per cent, they could be boosted to 80 per cent with AI guidance, and even lower than 30 per cent for new pilots. The Ukrainian Defence Ministry now says it has over 70 systems installed and that the number of strikes carried out using AI and computer vision technology has increased by tenfold this year.

The Atlantic Council estimates that roughly three-quarters of last year’s Russian loss to Ukraine was inflicted by Ukrainian drones, while CSIS estimates the ratio for the first half of 2026 is closer to 8:1, albeit admitting that such statistics are unverifiable. This edge might not persist. Ukraine retains a production edge, while Russia has caught up fast. It’s not about independence but the loop around the system.

With the war in Iran, the cost principle was reversed. Firing $3.4 million Patriot or $12.7 million THAAD interceptors at $35,000 drones proved cost-ineffective, and the Army’s entire stockpile of PrSM missiles was expended. The response is Gulf states have signed agreements with Kyiv for counter-drone expertise, including interceptor drones costing as little as $1,000. Washington requests nearly $55 billion for the Defense Autonomous Warfare Group in fiscal 2027, along with separate $74 billion for drone and counter-drone technology. Driscoll says drones are disposable weapons, and that making motors, sensors, batteries and circuit boards at home is his top priority. A loop requires a supply base as well as funding, which determines capacity. The cheap hardware determines the tempo and whoever imports the cheap hardware takes on another risk.

Faster wars, thinner warning

There are three reasons why mass changes crisis behaviour. Cheap, expendable: cost of using force is reduced. More signals are read in a shorter amount of time because of faster iteration and automated sensing. Drones are difficult to trace and an incident can be seen as an accident, a probe or an opening move. All three were displayed in the incursion into Poland in September 2025. Russian drones made at least 19 incursions into Polish airspace that night, and Polish and allied aircraft have been reported as shooting down several of them, using “missiles priced at just $400,000 against drones costing only $10,000”. Warsaw accused it of being an intentional provocation and Moscow stated that the drones were targeting Ukraine and that any intrusion was an accident. Secretary of State Rubio said the drones were intentionally launched but that the question was whether they were aimed at Poland. NATO’s secretary general refused to state whether the incursion was deliberate or not, while NATO launched the Eastern Sentry with intent yet undecided.

Both Washington and Beijing seem to understand the danger: Following Xi’s visit, they agreed to establish a channel for AI-related incidents and to finalize a military crisis-communication memorandum. A channel is only effective when someone responds to it, and there are doubts about the functioning of the current US-China hotline after the Chinese government refused to take calls at the time of the spy balloon incident in 2023.

The one thing that could help is a rule that would be an ownership and mode statement. Each state would, within a brief agreed period, declare after any event involving an uncrewed system near another state’s forces or territory whether or not the system was theirs, whether or not it was autonomous and whether they had lost control of it. In Poland, that would have been the answer to the question that was most important. The United States-Soviet Incidents at Sea Agreement of 1972 was a precedent that called for advance notice of dangerous actions, passed incident information through naval attachés and provided for annual reviews, and a later secretary of the U.S. Navy attributed to it the sharp reduction in incidents. Obviously, it couldn’t have worked prior to the change of Soviet policy following the Cuban Missile Crisis, so it needs political will before it needs drafting. A state with deniability will not be answering voluntarily, so the statement will have to be routinely tested and it will have to be independently verifiable. It would also be alongside November 2024 principle statements by US President Biden and Chinese President Xi Jinping that people should remain in charge of nuclear-use decisions.

What mid-sized militaries should take from it

The lesson for the other states, those of the smaller powers, is one of practice. Few are able to build the full stack, and even fewer want to. What’s valuable is the loop: the ability to understand frontline feedback, the capacity to adapt and rapidly manufacture simple systems, and the control over training and design data. Relying on one foreign part, model or data source is a weakness in peacetime and an even greater weakness in a crisis. The tempo of this competition will not be set by these small states, but they will have to live with its consequences, which is a reason for them to press to agree on tested channels and on rules of incidents, while the major powers have still some room for such negotiations. The competition will be the one who builds, fields and learns faster and more often. It is unclear if those states can maintain the pace without outstripping their capacity to handle a crisis.

Table of Contents

Amna Saqib

Amna Saqib is research officer at Center for International Strategic Studies (CISS), Islamabad.

Share this post :

Facebook
Twitter
LinkedIn

Leave a Reply

Your email address will not be published. Required fields are marked *