Palmer Luckey thinks humanoid robots have a powerful future, and he says that belief is why his company won't build them. Luckey founded Oculus VR and the defense technology company Anduril Industries. He gave his reasoning at Moonshots Live 2026 in a Moonshots with Peter Diamandis conversation recorded on September 25, 2026. His answer has two parts. The first is about performance: military machines, he argues, cannot afford to be generalists. The second is a company rule: build only what would not exist without Anduril.
A question about one robot for everything
Computer scientist Alexander Wissner-Gross, founder of Reified, raised the question. He said it is now possible to drop frontier AI models, the most advanced general-purpose systems, into robot bodies and get "pretty stunning results." Anduril already makes machines in many shapes and sizes. Would that variety eventually give way to one general-purpose robot, he asked, just as generalist AI models can now take on almost any task?
"No, I don't think so on the defense side," Luckey answered.
He said civilian work is different. In manufacturing, hospitality and medicine, he said, general-purpose robots might win because of economies of scale: the savings that come from building one design in very large numbers. A humanoid, meaning a robot built in roughly human shape, may not be the best body for a burger griller, a firefighter, a nurse or an assembly technician. But because it shares the human form, Luckey said, one robot can do all of those jobs. He said he would probably accept a robot that is 5% slower, or less reliable, than the best specialized burger-flipping machine if it shared commonality with the robot working as a nurse.
Why a 5% edge matters in war
Luckey said that trade-off does not hold in defense. Militaries are not trying to win narrowly, he argued. They try to overpower their adversaries by a wide margin, to "fight unfair" and to have an overwhelming threat of force behind any diplomacy or negotiation. Because everyone keeps pushing the limits, a jet with a reliable 5% edge over the enemy every time is "worth paying a lot of money for" in his view. He said the performance lost by giving up specialization would probably make a general-purpose military robot unworkable.
He pointed to Anduril's own product range. The company builds robot submarines that, he said, need to dive to 6,000 meters and sit on the ocean floor for months at a time. It also has surface systems, ground robots that roll up a hill and then wait in position, and aircraft that do very specialized tasks. "I just can't imagine a convergence that would make sense," he said, unless someone produced humanoids with onboard fusion reactors that could also dive to the depths.
One Anduril document shows how specialized an underwater design is. In an April 2024 report on subsea cable inspection, the company described its Dive-LD vehicle as having a free-flooding hull, which lets seawater fill the body, along with buoyancy foam and sealed electronics and batteries rated to 6,000 meters. The report covers twelve months of developing and testing a cable-inspection system, with trials along the Point Judith–Block Island cable, rather than months-long stays on the seabed. It also records practical limits: sea conditions, periodic surfacing and line-of-sight communication constrained operations, and cloudy water made camera images unusable, so the team inspected the cable again with sonar.
Peter Diamandis, the host and founder of XPRIZE, asked why Luckey wasn't building such machines. Luckey said specialization would quickly beat them. If a machine could dive to the bottom of the ocean and fly to the moon, he said, "it could get to the moon a lot faster if it didn't have to also dive to the bottom of the ocean." He compared the choice to cars. Carmakers share platforms, standards and interfaces across their models. Motorsport, by contrast, is "all completely different" and keeps pushing forward. Luckey expects defense to remain "this weird branch of exotic, highly specialized robotic system."
Opting out because he believes in them
Diamandis said he had expected Luckey, of all people, to be building humanoid robots. Luckey's second argument had nothing to do with performance. He said he opted out because he believes in humanoids. If he did not think humanoid robotics had such a powerful future, he said, he would probably build them himself. But he wants Anduril to "build things that wouldn't exist otherwise."
Wissner-Gross summed it up: "You want to maximize your marginal impact." Luckey agreed. "All the other people are already doing a great job," he said, and he wants to use their robots.
Humanoids at the controls of old radars
This is how Luckey can reject a universal combat robot and still welcome humanoids in defense. He does not see them replacing specialized weapons. He sees them operating equipment that already exists. He described humanoids as a powerful interface between AI and legacy platforms, the older systems built for a person to "pull the lever, push the button, read the screen." Robots could automate that equipment without anyone having to replace it.
His example was radar systems that cost millions, tens of millions or even hundreds of millions of dollars. Nobody wants to staff them during a hot conflict, he said, "because they're the first thing that's going to go." With humanoid robots inside, he said, those radars could be automated without putting people there.
The idea fits how Anduril describes its software. The company's developer documentation for its Lattice software says it can connect robots, sensors and programs made by other companies, which can share what they observe, receive information and take part in tasking. In congressional testimony in September 2023, Anduril chief executive Brian Schimpf described hardware as a way to deploy software in different environments, with shared information linking older equipment to newer autonomous systems. Neither document mentions humanoids, but both describe a company built to work alongside hardware it did not make.
Other markets left to others
Luckey said the same rule explains other gaps in Anduril's catalog. The company is not building small quadcopter attack drones because, he said, too many other companies are already doing a great job. It will not become a rocket launch provider either. Anduril makes space products and operates many of its own satellites and sensors in space, he said, but launch is "for all the other billionaires."