Andrew Kang runs a Nasdaq-listed fund whose entire reason for existing is owning robot companies before they go public. Ask him which American humanoid robot startups are actually going to deliver, and the list is short.

“That is Tesla Optimus, that is Figure AI, that’s Apptronik,” Kang, CEO of RoboStrategy, told me on the Humanoid Daily podcast . “Those are really the three key companies that we have strong conviction are going to deliver a real product to the market where a humanoid can manipulate things around us, can understand the physical world and that they will grow to become really massive enterprises.”

By my current (though continually incrementing) count, there are 112 companies in the United States that make humanoid robots, supply humanoid robots with parts, software, or AI, or are humanoid robot adjacent. Kang is also tracking “100-plus” in China alone, but isn’t currently looking at the European humanoid startups. Kang was careful not to say humanoid robot companies he isn’t picking will completely fail, and adds that there are some other strong competitors in the space.

But he doesn’t think there will be scores or dozens of winners here.

“The reason why we believe there are so few is because this is one of the most difficult technological challenges in the world," he says.

Why is the list so short?

It’s not about money or funding. It’s about expertise … specifically, the scarcity of it.

“I need an expert in robot learning, I need an expert in robot behaviors, I need an expert in computer vision, in hand engineering, in battery engineering, and fleet management systems,” Kang says, speaking as the CEO of a humanoid robot startup. “And then there are only a few hundred people, maybe, in all these categories that are really, really competent. And then putting them together in one company and raising billions of dollars of capital, that’s not very easy.”

In other words, software scales, and there’s plenty of available capital, but hiring the world’s best robotics engineers – or physical AI specialists – does not.

The one name he mentioned that I’m not personally too hyped about is Tesla with its Optimus robot. What Tesla has shown publicly is nowhere near Figure, nowhere near Apptronik, probably not Agility’s Digit, certainly not 1X’s NEO. Tesla should know motors and actuators and batteries from its car business and it has plenty of capital and compute, but we just haven’t seen Optimus perform at the level of the other American competitors yet.

So what’s the basis for the optimism?

“Actually, you’re exactly right,” Kang said. “At this point in time, Figure is ahead of Tesla on the AI side. You can see it.”

He pointed at Figure’s marathon livestreams , where its Figure 03 robots sorted packages in public for days at a stretch. The May run went 200 hours across three robots and roughly 250,000 packages without a catastrophic hardware failure. Kang’s explanation for why Tesla hasn’t put up anything comparable: “Elon’s had a million other things on his plate.”

He has been busy: taking SpaceX public, orchestrating the xAI merger, buying Cursor for $60 billion, dipping toes if not whole legs into U.S. politics repeatedly.

The question is whether Tesla can pull another Model Y out of its hat. Musk is never not going to have a million other things on his plate, but Kang thinks it will happen eventually.

“When he turns his attention, I think he gets a lot of stuff done, and can get as much stuff done as the combination of maybe five other CEOs put together,” Kang said. “I think it’s a matter of time before he really flips the switch.”

Kang’s reasoning: past performance is the best predictor of future delivery: “Even though he’s kind of been slow on deadlines sometimes, he’s delivered technology that no one else has been able to throughout his companies.”

Interestingly, RoboStrategy’s disclosed portfolio held $276.7 million in net assets as of August 31. Apptronik was 14.2%. Figure was 13.8%. Two of the three companies on the conviction list are, together, roughly 28% of the book. Tesla, on the other hand, is not there. That’s probably smart given that the company is currently down just under 20% year-to-date. Kang didn’t mention it, but there’s probably a price at which RoboStrategy buys in.

Of course, there’s much more to robots than just humanoids. The single largest position isn’t a humanoid company at all. Standard Bots, industrial automation, was 31.4% of net assets. Dyna, Dexmate, Maven Robotics, Path Robotics, Eccentric Machines and a handful of others fill out the rest. Close to a third of BOT, therefore, is standard conventional industrial robotics.

Kang disclosed his positions in our podcast, including Dexmate, whose seed and seed-plus rounds RoboStrategy led at $123 million and $216 million post-money, and actuator startup Eccentric Machines, whose pre-seed it led. He isn’t, however, very interested in Agility Robotics, which is going public via a $2.5 billion SPAC . On first glance that looks cheap against the $66 billion valuation Unitree touched on its Shanghai debut last month.

“I’m not sure if I would agree with them being cheap as a valuation,” Kang told me, adding that “there’s a lot of capital in the private capital markets that is now only just beginning to kind of turn on to robotics. And if your technology is really improving, if you have a good pace of development, if you are really winning core contracts and people believe that you’re going to solve general robotics, then there’s no lack of funding for it. So if you have to kind of go public via a SPAC, maybe there’s a reason.”

He’s incredibly bullish on the overall robotics and physical AI category, however, with an emphasis on humanoid robots. Kang’s ecosystem valuation for the eventual market is “clearly going to be tens of trillions of dollars. And I don’t say that as a joke.”

Amortize one humanoid over seven years, run it three shifts, get roughly three humans’ worth of output, and you land at “something like $5,000 to $7,000 per year for humanoid labor,” he says. At that price a lot of work that isn’t economical in America becomes economical, which is the whole reindustrialization argument — and the reason the FCC’s Covered List update in July , restricting imports of new foreign-made robots, matters more than it appears.

Over time, Kang expects the FCC to soften regulations the way the drone restrictions did. “I think there will be accommodations that will be made.”

His timeline for humanoid robots?

Millions of units around 2028 to 2029, humanoids “commonly on the street, in a lot of homes, I would say by 2030.” That’s less than four short years away.

There is a caveat, however.

“I can snap my fingers and you can generate a million instances of an AI chatbot, but for robots, we’ve got to build out the infrastructure, which is happening right now," he adds. "And then you’ve got to actually build the robots, and that’ll take a little bit of time.”

Global humanoid shipments were up nearly 300% year over year in the first half of 2026, to about 30,000 units. But the biggest IPO so far, from Unitree, has given back more than half its first-day pop, though it still trades at roughly triple the IPO price.

So there’s growth, and there’s excitement. But there’s still a lot of proof to be provided.