Scaleout Systems is a NATO-backed startup that wants to put autonomous target detection on small drones, and the company is betting that smaller AI models are the key to making it work. Founded by researchers from Uppsala University in Sweden in 2018, the company initially trained machine learning models on commercial truck hardware. That changed in 2022, after Russia launched its full-scale invasion of Ukraine, and the company pivoted toward defense applications.
The shift was fast. Scaleout now builds models that fit on the hardware of drones and forward-deployed computers, rather than relying on the heavy models from OpenAI or Anthropic. The CEO, Andreas Hellander, framed the move in strategic terms. “With the war in Ukraine and a shifting world, we realized that this technology can be very important to operationalize edge data and sensor data for machine learning to make sure that NATO allies have found that strategic advantage,” he said.
Hardware Limits
The challenge is the hardware itself. Drones and forward bases use a wide range of processors, from small embedded devices to powerful edge workstations. Hellander noted that the company’s models need to adapt to all of them. “They need to fit on forward-deployed hardware and edge hardware, which can vary quite a bit from small embedded devices to quite powerful edge workstations,” he said.
The company’s approach is to train models that are smaller and leaner, which means they require less computational power and can run on the actual hardware deployed in the field. The models have to work without a constant connection to a cloud server.
NATO’s DIANA Program
Scaleout was selected for NATO’s Defence Innovator Accelerator for the North Atlantic (DIANA) Challenge Program in 2025. Through DIANA, the company has worked on the Federated Aerial Intelligence for Recon project, adapting machine learning models for the edge computing hardware found in drones, drone pilot tablets, and field command posts.
The project is about making AI useful at the edge, not just in the cloud. For a military operation, that means drones can identify targets and make decisions on their own, even in areas where communication is down.
The Targeting Question
The technology raises a question that has no easy answer. Should machines be allowed to decide who to attack? Scaleout’s models are designed for surveillance and attack missions, and the company is building systems that could make those decisions autonomously. The company’s founder cites a strategic advantage for NATO allies, but the question of whether machines should decide who to attack remains unresolved.
The question is not a technical one. It is a moral one, and it has no easy answer.
“With the war in Ukraine and a shifting world, we realized that this technology can be very important to operationalize edge data and sensor data for machine learning to make sure that NATO allies have found that strategic advantage.”
The Bottom Line
Scaleout Systems is doing real work for NATO allies, and the technology it is building could change how small drones operate in combat. But the question of whether machines should decide who to attack is not a technical one. It is a moral one, and it has no easy answer.
The company’s founder believes the technology can give NATO allies a strategic advantage. Whether that advantage is worth the risk is a judgment each reader will have to make for themselves.
Where the paper stands
The paper backs Scaleout Systems’ push for smaller AI models on drones, which avoids locking in today’s leaders, and is against any regime that hands the market to the incumbents. Scaleout’s approach keeps its models small enough to run on the actual hardware deployed in the field, without constant cloud connection. That path lets newer firms compete against the giants, unlike licensing regimes that only giants can afford.
The company’s pivot came after Russia’s invasion of Ukraine, and its work with NATO’s DIANA program shows the practical stakes. Smaller models running on drones mean decisions can be made at the edge, even when communications fail. That is a genuine advance for military operations.
The paper is also against letting big companies hide their failures. Disclosure requirements for safety failures that were concealed are part of the narrow rules it supports, alongside clear limits on who gets to build these systems.
Source material: “Small AI models let drones autonomously identify and attack battlefield targets,” Ars Technica.
Get the Notebook.
The day's best stories and every fresh verdict, in plain English, in your inbox by seven. One email a day, no more.