Anduril Gains Ground in Pentagon's Drone Autonomy Race
Anduril Industries had a good week, and it had almost nothing to do with anything it built. In the span of a few days, Hermeus, the Atlanta-based hypersonic startup, announced it had selected Anduril's autonomy software to fly its Quarterhorse drone, while General Atomics, the longtime kingpin of American drone-making, watched its Dark Merlin combat aircraft go down in the California desert. Neither event took place on a battlefield. Both took place in the market for what the Pentagon calls "collaborative combat aircraft" — the AI-piloted wingmen meant to fly alongside human pilots in the next generation of American air power.
The Hermeus deal is a straightforward win: another company, this one building a jet meant to fly at hypersonic speeds, has decided it would rather buy Anduril's autonomy stack than build its own. The Dark Merlin crash is a win of a different sort — the kind that arrives by not being yours. General Atomics has spent decades as the default supplier of American drones, the company whose Predators and Reapers became shorthand for a certain era of remote warfare. Dark Merlin was its bid to stay relevant in the autonomy era, a program built to prove the company could make an uncrewed jet think for itself, not just fly in a straight line over a desert. Instead it ended up in the desert in pieces.
None of this happens in a vacuum. The Air Force is trying to field a fleet of autonomous fighters that can fly alongside manned jets, absorb risk, and multiply the number of aircraft in the sky without multiplying the number of pilots. Whoever supplies the software that lets those drones sense, decide, and act without a human hand on the stick stands to become as central to future air power as the engine-makers were to the last one. It is a contract worth fighting for, and increasingly a fight defined not by dogfights but by demonstrations — whose drone flies, whose crashes, whose software gets picked by someone else's airframe.
Anduril, a company barely eight years old, has built its reputation on exactly this kind of software-first pitch, arguing that autonomy is the product and the airframe is almost incidental. Getting chosen by Hermeus suggests that argument is landing with more than one customer. A crash at General Atomics, meanwhile, is a reminder of how fragile incumbency can be when the technology underneath a decades-old business model changes. Nobody outside the test range yet knows what caused Dark Merlin to fail, and one mishap does not settle a competition this large. But in a race where the Pentagon is watching closely for who can prove reliability first, optics compound. For the taxpayers eventually footing this bill, and for the pilots who may one day fly with — or without — these machines beside them, the difference between a software win and a wreckage photo is not trivial. It is, increasingly, how these contracts get decided.