Defence AI is consolidating around fusion
Rather than isolated pilot models, defence organisations are buying large data‑fusion platforms that ingest imagery, signals and text and output operational insight, a shift visible in recent NATO and contractor activity. Reporting says NATO acquired Palantir’s Maven system and that similar fusion platforms are being rolled across alliances, raising concerns about speed, oversight and the compression of human decision loops. The engineering implication is clear: defence workflows now prioritise schema discipline, low‑latency joins and governance that survives procurement and regulatory audits. (rnz.co.nz, ibtimes.com.au)
NATO did not buy a single chatbot. On March 25, 2025, the NATO Communications and Information Agency finalized a deal for Palantir’s Maven Smart System NATO, and the alliance said the software would be used inside Allied Command Operations, the command that runs NATO military operations from Supreme Headquarters Allied Powers Europe in Belgium. (ncia.nato.int, nato.int) What NATO bought is a fusion system, which is software that takes different kinds of military data and lines them up into one picture. NATO said Maven Smart System NATO would combine large language models, generative artificial intelligence, and machine learning to improve intelligence fusion, targeting, battlespace awareness, planning, and decision-making. (ncia.nato.int) That is the real shift in defence artificial intelligence right now. The market is moving away from isolated pilot models that label one image or summarize one report, and toward platforms that ingest imagery, radio and sensor signals, and text so commanders can act from one shared screen instead of five separate systems. (ai.mil, ncia.nato.int) The United States built the template years ago. A Pentagon memo from April 26, 2017 created Project Maven to use machine learning on the flood of surveillance data, and the current Chief Digital and Artificial Intelligence Office now describes Maven Smart System as a tactical platform that fuses sensor data for real-time detection, tracking, and decision support. (dodcio.defense.gov, ai.mil) By 2024, Maven had stopped being a narrow experiment. Palantir announced a September 20, 2024 Army Research Laboratory contract worth up to $99.8 million to expand Maven Smart System access across the Army, Air Force, Space Force, Navy, and Marine Corps, which is what a platform rollout looks like after a pilot phase ends. (investors.palantir.com) The Pentagon was building the plumbing around it at the same time. The Chief Digital and Artificial Intelligence Office says Combined Joint All-Domain Command and Control is meant to give commanders “decision advantage” by sharing data across air, land, sea, space, cyberspace, and partner networks, and the Government Accountability Office said in 2025 that the concept was expanded in 2023 to include international allies. (ai.mil, gao.gov) NATO’s deal showed how fast this is moving from theory to procurement. Defense reporting said the requirement-to-contract cycle took about six months, which is unusually quick for alliance buying, and NATO’s Joint Warfare Centre in Norway was already training staff on the system by August 2025 for use in exercises. (defense-update.com, jwc.nato.int) Other governments are now in the same conversation even when they say no. Radio New Zealand reported on April 9, 2026 that New Zealand’s government said its Defence Force had “no existing plans” to use Palantir in emerging technologies, while also confirming that the Defence Force already uses Palantir as an analytics platform for planning and that the wider government partnership is led by the Government Communications Security Bureau. (rnz.co.nz) The pressure behind all this is speed. Bloomberg reported on March 24, 2026 that Palantir’s chief technology officer, Shyam Sankar, called the Iran war the first major conflict where artificial intelligence played a central role, and International Business Times reported on April 8, 2026 that Palantir tools were being described as compressing the “kill chain,” the military term for the steps between finding a target and striking it. (bloomberg.com, ibtimes.com) A fusion platform changes the engineering problem too. When one system has to join satellite images, drone video, logistics records, chat messages, and sensor alerts in seconds, the hard part is no longer just the model; it is whether every data field uses the same labels, whether the joins happen with low latency, and whether every recommendation leaves an audit trail that survives military reviews and procurement scrutiny. (ncia.nato.int, gao.gov, ai.mil) That is why defence artificial intelligence is consolidating around fusion. The winners are not the companies with the flashiest demo model, but the ones that can make dozens of data sources speak one language, keep humans formally in the loop, and still deliver an answer before the battlefield changes again. (ncia.nato.int, ai.mil, rnz.co.nz)
