If there is a golden age in defence manufacturing, it is not simply because governments are spending more. It is because money is being redirected into the things recent wars have exposed as decisive: ammunition depth, drone volumes, layered air defence, software-defined systems, and the industrial capacity to replace losses quickly. EU defence spending hit €343 billion in 2024, is projected by the European Defence Agency to reach €381 billion in 2025, and defence investment alone is forecast to approach €130 billion. At the same time, the EU’s SAFE instrument sits inside a Readiness 2030 plan aimed at unlocking more than €800 billion, with the first SAFE wave for eight member states covering about €38 billion and first payments expected in March 2026; NATO has also committed allies to a path toward 5 percent of GDP by 2035, including a dedicated share for resilience, innovation and the defence industrial base.
That is why the most revealing launches of the past year are not only prestige missiles or next-generation aircraft. They are cheaper one-way effectors, anti-drone missiles, collaborative combat aircraft, autonomous launchers, solid-rocket-motor factories, and procurement tools meant to move at software speed. Taken together, these programmes suggest a lesson shaped in large part by Ukraine and the return of attrition: precision still matters, but mass, update cycles and industrial tempo now matter just as much.
In Europe, Rheinmetall has become a symbol of this turn from boutique capability to industrial throughput. The company handed over the first Boxer Skyranger 30 verification model to the Bundeswehr in early 2025, a mobile air-defence system that combines cannon, missiles and sensors on one platform, while separately opening what it says will be Europe’s largest ammunition factory at Unterlüß after investing nearly €500 million. The new plant is set to produce up to 350,000 artillery shells annually at full capacity, with rocket-engine production planned from 2026.
Other European firms are moving in the same direction, but with a stronger software accent. MBDA unveiled its ONE WAY EFFECTOR at the 2025 Paris Air Show and then secured its first contract with France’s DGA in January 2026, presenting the system as a long-range, lower-cost weapon designed to saturate enemy air defences and built through an industrial model that reaches beyond traditional arms makers. MBDA also added SPEAR GLIDE, a lower-cost strike variant built for rapid production, using AI-driven image-based navigation for GNSS-denied environments.
Helsing has gone even further toward what might be called software munitions. The company says its HX-2 strike drone, unveiled in late 2024, is swarm-capable, mass-producible and able to re-identify targets under electronic warfare, and it has linked that vision to factory scale by opening its first Resilience Factory in southern Germany, with an initial capacity of more than 1,000 HX-2s per month, while also announcing a 6,000-drone batch for Ukraine. Saab’s Nimbrix, aimed at cheap counter-UAS defence with first deliveries targeted for 2026, and Airbus’s partnership with Kratos to build a German mission-system variant of the XQ-58A collaborative combat aircraft, show that Europe’s rearmament is no longer confined to legacy primes reproducing old catalogues.
Across the Atlantic, the United States is pursuing the same logic through a mix of government financing, private capital and commercial-style manufacturing. The Pentagon’s Office of Strategic Capital now has authority to issue loans and loan guarantees across 31 covered technology categories with a pilot lending limit of $984 million, while firms such as Anduril and L3Harris are building the physical base to absorb that money. Anduril says its Arsenal-1 campus in Ohio will bring more than 5 million square feet of manufacturing space and over $900 million in capital investment, while L3Harris has broken ground on five new solid-rocket-motor facilities in Virginia and says it is investing nearly half a billion dollars across major sites, including a new large-motor campus in Arkansas.
The products coming out of that ecosystem are also different from the post-Cold War norm. Anduril’s Barracuda family is being marketed as a low-cost, software-defined autonomous air vehicle built for hyper-scale production; L3Harris has debuted Red Wolf and Green Wolf launched-effects vehicles for kinetic strike and electronic warfare, with in-flight collaboration and re-targeting; Raytheon has demonstrated the self-driving DeepStrike launcher and ASTARTE, AI-enabled software that automates airspace deconfliction and course-of-action generation. This is not simply a faster production cycle for old weapons, it is a merger of munitions, autonomy and command software into one industrial stack.
The government AI layer matters just as much as the hardware. The Pentagon’s Thunderforge project, awarded by DIU in 2025, is meant to insert commercial AI into operational and theatre-level planning so staffs can synthesize more data, generate alternative courses of action and run AI-assisted wargames faster; the Replicator initiative was designed to field autonomous systems in the multiple thousands, and Replicator 2 shifted that model toward countering small drones at critical installations and force concentrations. NATO has followed a parallel path, acquiring Maven Smart System NATO for Allied Command Operations and training more than 100 personnel on what it describes as the alliance’s first AI-enabled command-and-control software.
European governments are building their own state-facing AI infrastructure as well. Britain’s Defence AI Centre is piloting the AI Model Arena, a secure vendor-neutral framework that can assess up to 100 supplier models at once, and the UK’s 2025 partnership with Palantir tied that push to up to £1.5 billion in investment and AI-enabled tools for decision-making, military planning and targeting. France’s AMIAD was created to turn defence AI experiments into deployed solutions, while the Bundeswehr now openly describes AI as increasingly foundational to planning and command processes in digitalised land operations.
This is also where the phrase “AI-guided bombs” needs care. Traditional smart bombs such as JDAM and Laser JDAM are still mainly about GPS, inertial and laser guidance, not autonomous battlefield reasoning; the frontier now lies in weapons that can detect and classify targets in poor visibility, navigate without reliable satellite signals, or re-identify targets after jamming. Raytheon’s StormBreaker autonomously detects and classifies moving targets, MBDA’s AKERON LP uses AI-based image processing and data fusion, SPEAR GLIDE uses AI for image-based navigation in GNSS-denied conditions, and Helsing’s HX-2 is designed to search and re-identify targets even without a continuous data link.
For all the rhetoric of autonomy, Western policy still insists that human judgment remain central. The Pentagon’s Directive 3000.09 says autonomous and semi-autonomous weapon systems must allow appropriate levels of human judgment over the use of force, and Helsing says a human operator stays in or on the loop for critical decisions; British defence doctrine likewise frames adoption around ambitious, safe and responsible AI. That means the golden age now emerging is not a clean handover from soldiers to machines. It is the arrival of an arsenal in which factories, code, sensors and munitions are being fused into a faster, denser and more adaptive war-making system.