Computing moves closer to the fight
The strategic value of Scaleout’s technology is not simply that a drone can recognize an object. Militaries have demonstrated computer vision on autonomous systems for years. The more consequential development is the attempt to create a distributed learning network that can adapt models near the battlefield, where conditions change faster than centralized systems can respond.
Scaleout, founded by researchers from Uppsala University in 2018, originally developed machine learning for commercial vehicles and other edge hardware. Russia’s full-scale invasion of Ukraine in 2022 pushed the company toward defense applications. In 2025, it was selected for NATO’s Defence Innovator Accelerator for the North Atlantic, known as DIANA, and began adapting its platform for drones, pilot tablets and field command posts.
The company’s approach relies on decentralized, or federated, learning. A drone can process camera and sensor data locally using a small AI model, rather than sending every image to a remote data center. When connectivity is available, selected updates can move to a nearby node at a platoon or company headquarters. That node can combine information from multiple devices, retrain a model and distribute an updated version.
This architecture addresses two military constraints at once: communications may be unreliable, and battlefield conditions are not static. A model trained on one landscape may struggle in a city, under different weather or after adversaries alter camouflage and tactics. A system that can learn from recent local data may improve performance without waiting for a large centralized retraining operation.
From reconnaissance to lethal decisions
Ars Technica reported that Scaleout is participating in the BAE Systems Bofors led Affordable Loitering Modular Ammunition, or ALMA, project. During a public Swedish demonstration in January 2026, the concept used onboard computing to detect, identify and geolocate potential threats, prioritize a mission-defined target and carry out an attack without direct human commands.
A human operator could still control and direct the drone. That distinction matters commercially and operationally, but it does not eliminate the central governance question. If software decides which objects deserve priority and then receives a locally updated model, the human role may shift from making an individual targeting decision to approving a system whose behavior is harder to inspect in real time.
Scaleout also tested its platform at a Swedish Air Force base in Uppsala in June. The company demonstrated that a local node could continue inference and active-learning processes after losing its connection to a central node. That capability is valuable in an environment where jamming or physical damage can isolate units from national command networks.
The same feature creates risk. Battlefield data can be incomplete, misleading or deliberately manipulated. A model trained on rapidly collected examples could become more responsive, but it could also reinforce a classification error and distribute that error across allied devices. Faster adaptation is therefore not automatically better adaptation.
A different defense technology competition
Scaleout’s opportunity lies in a market where cheap, replaceable hardware may matter more than expensive platforms. Large defense contractors can supply missiles, sensors and command systems, while specialist software companies provide the intelligence layer that connects them. A lightweight model that runs on affordable drones could offer a more scalable product than a system dependent on powerful centralized infrastructure.
Its NATO connection may also become a commercial moat. Defense customers need interoperability, cybersecurity and evidence that a system can function inside allied networks. Participation in DIANA and the ALMA project gives Scaleout exposure to those requirements and to potential procurement channels that a commercial AI startup would struggle to reach.
The competitive challenge is validation. Military buyers will need to know which data changed a model, who approved the update and whether the resulting behavior remains within mission limits. They will also need reliable rollback procedures when an update performs poorly.
That makes the decisive product less likely to be autonomy alone. The stronger business proposition is controlled, auditable adaptation at the edge. If Scaleout can demonstrate that commanders retain meaningful authority while units gain resilience against disrupted communications, its platform could become part of a broader allied architecture.
The market is moving toward smaller models, distributed hardware and software that continues working when cloud access fails. In defense, however, the companies that win will be judged not only by how quickly their systems learn, but by whether they can prove when those systems should stop.
Rebeca Smith is not a person. No notebook, no deadlines, no face behind the name — just a byline this newsroom publishes under. Here is the production line underneath it, because a name beside a portrait reads like a journalist, and this one is not one.
The models. Writing: gpt-5.6-luna and qwen3-max. Out on the live web: gpt-5.6-luna and gpt-5.6-terra. Pictures: gpt-image-1 and gpt-image-1-mini. Swap one in the newsroom and this line swaps with it — it is read off the machines, not typed here.
How a story is made
- Research. The searching model reads around the story, pointed at primary sources — the filing, the post, the repository — rather than at somebody else's write-up of them.
- Writing. The writing model drafts it against what was found, at Rebeca Smith's usual length and in Rebeca Smith's usual register.
- The loop. A reviewer reads the draft and sends it back with notes. Then reads it again. A piece can go round several times before it leaves the building.
- Enrichment. A quotation has to appear word for word on the page it is taken from. A chart may only use figures that appear in the source it cites. Whatever fails is dropped, and the reason is kept.
- Fact check. A last pass hunts for claims the article makes and its sources do not.
- A human stop. Sensitive subjects are held for a person to read before publication, and a person can kill any of it at any point.
If that sounds less like a newsroom and more like a factory: quite. It is called Press Factory.
This article was generated using AI and published automatically without human pre-publication review.
How this article was made
The article was produced by the Grandmonts Media News Engine using automated research, drafting and verification workflows. No human editor reviewed the article before publication. Grandmonts Media remains responsible for the published content. Errors can be reported at office@grandmonts.cz.