53 Silicon Valley investors passed on autonomous combat aircraft. The founder who kept going just moved his own deadline back by up to a decade.

A former Navy SEAL who bet Silicon Valley on autonomous combat aircraft in 2015 is publicly pushing his own prediction back by five to ten years. He blames slow Pentagon buying and a thin pool of patient defense capital, not the autonomy stack his company sells.

Brandon Tseng, co-founder and president of Shield AI, told War on the Rocks that he was turned down by thirty consecutive Silicon Valley investors in August 2015. The next year he met twenty-five more, and two said yes. His original thesis, that by 2035 every military asset would be powered or piloted by physical AI, has slipped to 2040–2045. The reason, he says, is the buying cycle, not the lab.

Shield AI builds aircraft only when it can deliver a minimum 10x price-performance improvement over what the military already flies, Tseng said. On missions where that bar is unreachable, the company sells Hivemind, its autonomy software, instead. Board member Peter Levine of Andreessen Horowitz has compared the bet to Uber and Airbnb in their early days, when both were called contrarian.

Tseng's clearest evidence for software as the durable asset is the portability data. According to him, the same Hivemind stack that took three years to put on an F-16 now ports to a Kratos Firejet in 120 days, and two engineers can build and fly an AI pilot for a one-way attack drone in under two weeks. Roughly 90% of the code carries over between platforms. The hardware that pays the bills is the V-BAT vertical-takeoff ISR aircraft, which Tseng says can do a Predator drone's mission set at about one-tenth the total cost, a claim he ties to deployed use in Ukraine rather than to a published benchmark.

Tseng estimates that across any 24-hour window, warzones are covered for less than 10% of significant terrain, with no infrastructure connecting strategic surveillance to tactical drones. He says about 50 Predator-class aircraft have been lost since 2023, many to cheap mobile systems. Satellites handle wide-area, persistent coverage of fixed or predictable targets, while crewed aircraft and runway-bound drones cannot loiter over a single patch of contested ground for hours. V-BAT, in his telling, exists to fill that gap, including in GPS- and comms-denied environments where operators in Ukraine say range and endurance are the deciding factors.

Tseng's procurement critique names the actual cause of the slip. He argues the Pentagon still treats software as nearly free, a posture he has carried into congressional testimony where he has argued autonomy now matters more than stealth or GPS satellites did at introduction. The Collaborative Combat Aircraft program and the Low-Cost Unmanned Combat Attack System, known as LUCAS, are the first real attempts to fund autonomy and mass at the same time. Without that combined buy, the technology races ahead and the fielded fleet does not.

Enrique Dans makes the broader hardware-is-hard case the same way: software iterates on a laptop, hardware iterates on a factory floor, and the gap between the two is what kills most defense startups. Tseng's answer is to build the airframe only when the mission forces it and to let the autonomy stack amortize across the rest. The 30-of-30 investor rejections in 2015 are still the cleanest measure of how much the capital side has had to relearn that lesson.

Two things have not changed. Tseng was wrong on the 2035 date, and he knows it. The bet that the AI pilot matters more than the airframe is still his, not a settled industry view, and the public record does not support claims about direct Pentagon adoption, exposure, or customer relationships beyond what he says. The binding constraint on autonomous combat aircraft is no longer the technology. It is the procurement cycle and the patient capital behind it, and a founder with skin in the game is the one saying it.