Autonomous weapons startups attract record $14.6 billion amid regulatory race

Autonomous weapons startups have pulled in more than $14.6 billion in venture capital in the first five months of 2026, according to Crunchbase data cited in the lead article, already overtaking the defence sector’s previous full-year record before summer has begun. The surge has been led by a small number of platform companies, with Anduril Industries at the centre of the market’s latest re-rating after closing a $5 billion Series H round at a $61 billion valuation, more than doubling its worth from June 2025.

That flood of money is arriving just as Washington is revisiting the rules that determine what autonomous systems may legally do. Ten days after President Donald Trump signed National Security Presidential Memorandum 11, Senator Ruben Gallego wrote to Defence Secretary Pete Hegseth asking whether the Pentagon’s Defence Autonomous Working Group has enough dedicated staff focused on preventing civilian harm as it develops and deploys autonomous and semi-autonomous weapons. The memorandum gives the Pentagon 90 days to revise Directive 3000.09, the policy framework that has governed military autonomy since 2012. In other words, the capital race is moving far faster than the legal framework.

The scale of the funding boom marks a sharp break from the recent past. Defence tech raised $1.6 billion in 2020, about $3.9 billion in 2021, then remained in a narrow band through 2024 before jumping to $9.6 billion in 2025. In the first five months of this year alone, startups have already attracted $14.6 billion across 107 venture rounds. The money is also concentrating more heavily in a handful of firms able to win large contracts, build manufacturing capacity and, in some cases, acquire younger players. Andreessen Horowitz, Thrive Capital, Advent International, JPMorgan Chase, Kleiner Perkins and Founders Fund have all backed major rounds this year.

Anduril is the clearest example of how the category has changed. The Costa Mesa, California-based company said its revenue reached $2.2 billion in 2025, a figure that helps explain why investors are comfortable assigning it a valuation that now stands at roughly 28 times trailing sales. The company is also expanding aggressively, with plans to build Arsenal-1 in Ohio and a new campus in California. Yet it is still expected to post an operating loss of about $1.2 billion in 2026, with profitability not anticipated before 2030.

At the centre of Anduril’s pitch is Lattice, the software platform that links sensors, vehicles and weapons into a shared operational picture. Its Lattice Mesh layer is designed to move data across tactical nodes without relying on a central hub, while Lattice C2 turns that stream into command-and-control outputs for human operators or, where permitted, machine-speed decision-making. The U.S. Army’s $20 billion, ten-year enterprise contract awarded in March gives Anduril a major role in counter-drone operations and consolidates more than 120 separate procurement routes into one framework. That contract does not just make Anduril a supplier; it places its software inside the debate over how much autonomy the military is willing to allow.

Shield AI has become the other major emblem of the sector. The San Diego company raised $1.5 billion in March at a $12.7 billion valuation after the US Air Force chose its software to run on Anduril’s Fury autonomous fighter jet. Its Hivemind system is built to operate onboard aircraft rather than through a remote control station, allowing it to function when GPS is jammed, communications are degraded or no human pilot is available. The system has already been used in Ukraine and Israel in environments where electronic warfare or close-quarters operations make remote control impractical.

Other firms are drawing capital with similar arguments. Allen Control Systems, based in Texas, raised $200 million in a Series B round to expand production of its Bullfrog autonomous weapon station, designed to counter aerial threats with artificial intelligence, computer vision and precision robotics. Overland AI, a Seattle company, raised $100 million to meet military demand for autonomous ground vehicles that can handle reconnaissance and resupply in difficult terrain. Saronic, which is developing autonomous warships, is also said to be raising as much as $1.5 billion at a valuation of about $7.5 billion, with Kleiner Perkins leading the round.

The investment thesis is straightforward: the Pentagon is buying more like a commercial customer, while wars and geopolitical tensions are pushing governments to rearm faster and from newer suppliers. But the exit maths remain difficult. Anduril’s eventual path may be an initial public offering or an acquisition, yet both routes are complicated by its scale, its valuation and the political sensitivity of its customer base. Shield AI’s structure suggests a public listing could come no earlier than 2028, if growth continues and regulators do not intervene.

The regulatory question may prove more consequential than the financing itself. Directive 3000.09 requires that autonomous weapons be designed so commanders can exercise “appropriate levels of human judgment over the use of force”, but that phrase has been contested for more than a decade. A June analysis by the Centre for Strategic and International Studies argued that the key issue is how the Pentagon defines an autonomous weapons system in the first place. Senator Adam Schiff’s Human Authority in Lethal Operations Act, introduced this month, would require a designated commander to authorise every use of force involving autonomous or semi-autonomous systems and document target selection.

For investors, that legal ambiguity is not a side issue; it is the core of the bet. Anduril’s Lattice and Shield AI’s Hivemind are built to compress decision time to machine speed. Whether the Pentagon’s updated rules allow that speed to be used in practice will help determine how far this boom can run, and whether the biggest venture wager in defence history turns into a durable industry or an expensive proof of concept.