Booster Robotics builds the humanoid body. You bring the brain.

Booster lets you use whatever physical AI you want

Most humanoid robot makers want to build both the body and the brain. Booster Robotics, the Beijing startup that is now one of the three biggest humanoid shippers in the world, only wants to build the body.

The brain is up to you:

"Actually, Booster, we produce the hardware … but we do not focus on the software," Booster's Clarence Huang told me when we met at Web Summit this week. "We provide a open platform. That means you can come in and put your own foundation model, you can do whatever you want. Because we give full access of API and SDK."

That is a very different bet from the one Tesla and Figure and many other robot manufacturers are making. They build their own robot AI in-house and plan to sell a finished worker. Booster sells a programmable humanoid and the tools to program it, and then gets out of the way.

You start with sample code

Booster supplies sample code so developers aren't starting from zero. It also bundles the whole development workflow into one environment, Booster Studio, which is free.

"Nowadays we know that the developers, they use one app for coding, another app for debugging, another app for sim to real," Huang says. "But we combine all in one platform … that means all this process, they can complete it in our platform. No need to change apps or change their work platform."

The hardware includes the T2.

Booster launched it in July, and it stands 1.4 meters tall, has up to 75 degrees of freedom, and carries 10 kilograms across both arms. Its brain is Nvidia's Jetson Thor, rated at 2,070 TFLOPS. A magnetic charging connector lets the T2 keep working while it charges. The one on Booster's stand was docked and working full time. Untethered, Booster says it gets a bit more than two hours of walking.

The T2 I saw was wearing very basic hands, and there is a reason for that.

"The hands are very expensive," Huang says. "You can spend $100,000 on hands."

Booster doesn't make its own. Customers who need dexterity buy from specialists like BrainCo and Sharpa, and the hands get integrated onto Booster's arms.

This is working for Booster

So does the build-the-body-only model actually work?

So far, yes. In the first half of 2026, global humanoid shipments hit roughly 25,000 units, up 432%, according to IDC. Booster ranked third with about 1,000 robots. Only Agibot and Unitree shipped more, as I reported for Forbes and covered in Humanoid (almost) Daily. Chinese coverage of the IDC data notes that Booster is the only open-platform company in the top three. Its robots go out one or two at a time to universities, research labs, plus RoboCup teams and developers in more than 40 countries, not in big fleet orders.

Investors like it too. In April, Booster closed a Series B of nearly 1 billion yuan, about $139 million, and said first-quarter shipments were up 500% year over year.

The smaller robot may matter just as much for the long game.

That is the Booster K1, a 95-centimeter humanoid that won the KidSize league at RoboCup 2025 and starts at around $6,000. Huang says its main market is younger students, not university researchers.

"The main use case is for the education for primary or middle schools," he told me. "So that the young students can have the basic knowledge of the physical AI … Like 20 or 30 years ago, I knew the computer in school. It's the same thing nowadays."

A kid who learns to program a Booster in grade seven is a pretty good candidate to build on Booster's platform in grad school. (Apple ran a similar playbook with classroom Macs decades ago.)

All of this happens in the most crowded humanoid market on earth. China has hundreds of humanoid robot makers, and many of the biggest hedge their bets. Unitree and Agibot both sell quadrupeds alongside their humanoids, for instance.

Booster doesn't.

"We only focus on humanoids," Huang says. "Not like other companies, they have quadrupeds or whatever else."

Will it work for the long term?

I think the hardware-only focus is a smart bet, at least for now.

Robot foundation models are improving fast, and startups are now offering robot brains on demand. GR00T 2 is very impressive.

So a company that doesn't have to win the AI race only has to build bodies the winners want to run on. The risk is that bodies become commodities, and hardware margins in China's price war are not exactly generous.

As for what's next, Huang wasn't announcing anything.

"Maybe we will have other models," he says. "But we will focus on the hardware … and then to engage more developers in our platform."