Anduril’s $1B Factory Is Thunder’s Real Test

By August 2026, Anduril and Archer Aviation had spent two years developing Thunder, a series hybrid-electric autonomous VTOL. The aircraft’s harder test sits in Columbus, Ohio, where Anduril plans a $1 billion factory intended eventually to produce tens of thousands of autonomous systems and weapons each year.

Autonomous military aviation is becoming an industrial-systems business. The strategic unit now links sensors, mission software, command authority, airframes, and replenishable production inside contracts built for continuous updates.

Thunder’s value lies in the system around it

Archer describes Thunder as part of a broader aircraft platform for commercial and military uses. The partnership matters because an autonomy vendor and an aircraft manufacturer had to combine separate technical blocks before either could offer a deployable vehicle.

Thunder follows an older design choice at Anduril. Years before the rotorcraft appeared, the company was building software to connect systems ranging from ground radar to fighter jets so officers could act on information more quickly. Thunder gives that architecture another physical form.

When onboard software detects and identifies an object, commanders must still decide which sensors to trust, how to transmit the result, who may authorize an action, what the aircraft should do when communications disappear, and how operators review the outcome. Better prediction increases the need to assign human authority and accountability, even without a cockpit.

The U.S. Army made that dependency visible by bringing nine defense firms into Operation Jailbreak, a hackathon on AI weapons integration and interoperability lessons from Ukraine. By convening nine vendors, the Army treated interoperability as an operational requirement.

Thunder remains an announcement, with published details on its design and co-development but none on price, production rate, operational deployment, or a procurement award. Commanders, maintainers, factories, and contracting officers still have to keep its technical blocks working together.

The factory is the real flight test

A flight demonstration proves that a system can work under selected conditions. A production line tests whether Anduril can preserve that performance across suppliers, software versions, inspections, repairs, and replacement units, turning autonomous hardware into replenishable military capacity.

Anduril has made that constraint explicit with Arsenal-1, the planned $1 billion facility in Columbus. The company says the factory could eventually produce tens of thousands of autonomous systems and weapons each year.

To hit that target, Anduril must design products that workers can build, test, and revise as software, components, and mission requirements change. Incident logs, version control, and inspections turn code and prototypes into controlled physical output.

Anduril’s projections show how much capital that conversion can consume. The company told prospective investors it expected roughly $4.3 billion in 2026 revenue, while operating loss would rise roughly 50% to $1.2 billion. The projections pair rapid growth with rising losses as Anduril builds factories, develops aircraft, and supports deployed systems.

Thirty-seven current and former employees and contractors described safety concerns and project challenges in Anduril’s manufacturing operations. Anduril called the claims inaccurate. The conflicting accounts leave Arsenal-1’s manufacturing discipline to be demonstrated.

Ukraine reportedly stopped using Anduril’s Altius drones in 2024 because of problems. In the field, operators experience an autonomy stack through the reliability of the complete vehicle.

Rivals are standardizing the test

Anduril’s rivals are assembling similar combinations of vehicles and autonomy. Shield AI raised $2 billion at a $12.7 billion valuation, with autonomous software reported to represent half its business. Helsing says its AI attack drone can fly as far as 100 kilometers and re-identify and engage targets without a data connection. The company also says it can produce the drone by the tens of thousands annually at lower cost than existing autonomous systems.

Anduril has also moved across airframe categories. Its jet-powered unmanned drone completed a first flight semi-autonomously while the U.S. Air Force pursued swarms of unmanned jet aircraft. Rotorcraft, strike drones, and jet-powered vehicles have different operating requirements, but their builders are assembling the same capabilities: onboard autonomy, contested-environment performance, integration with other systems, and a credible manufacturing path.

Each vendor still owes buyers proof. Helsing’s annual output remains a company claim, and reports of years of V-BAT technical hitches and safety concerns leave Shield AI with its own reliability test. Thunder likewise lacks operational evidence. As autonomy becomes a baseline, military buyers can discriminate on reliability, integration, and replenishment.

The contract now sits inside the machine

The U.S. Army is buying a broader unit of capability from Anduril. Its 10-year agreement worth up to $20 billion covers software, hardware, and services, including a five-year optional ordering period. The agreement sets a $20 billion ceiling rather than a spending commitment. By covering all three layers, it keeps the components needed to update and support capability within one procurement vehicle.

That structure gives procurement rules technical consequences. A buyer that acquires an airframe under one process, autonomy software under another, and integration work under a third can create contractual seams where the operational system requires continuous coordination. Ordering periods, update rights, permissible uses, and service responsibilities determine who can change the system after deployment and under what authority.

Because prediction can feed physical action, buyers and vendors must assign who approves missions, who can modify models, who accepts the risk of degraded communications, and who investigates failure. The contract assigns much of that deployment accountability before an aircraft enters service.

Capital multiplies Anduril’s tests

In 2019, Anduril had raised $120 million and was shipping drone interceptors. In May 2026, it raised $5 billion at a $61 billion valuation. That seven-year capital leap lets Anduril develop multiple airframes, construct Arsenal-1, and buy adjacent capabilities. Physical learning still sets the pace of safe manufacturing and combat reliability.

Anduril agreed to acquire missile-defense modeling startup ExoAnalytic Solutions, a deal that would expand its space-unit staff from about 120 to about 250. The company has also reportedly discussed another funding round that could value it at roughly $100 billion. With more capital, Anduril can combine software, sensors, and hardware across more programs, multiplying the integration and production tests it must pass.

China barred government purchases from 46 U.S. companies including Anduril and separately placed 10 U.S. companies under export controls. Those restrictions can limit customers, suppliers, and cross-border partnerships even as security spending increases. Defense manufacturers must scale within politically fragmented markets rather than a single global demand curve.

Thunder makes the industrial loop visible without proving it complete. Behind it, Anduril is wagering $1 billion in Columbus that software changes can become trusted replacement inventory. The aircraft is where that industrial loop touches the sky.