Some thoughts, reflections, and judgments on robots.

"We don't even have a dedicated chip for embodied intelligence."

Right now, robots are where autonomous driving was in 2016.

Recently, many friends in the Chinese-speaking internet have been discussing this sector. The A-share robot index surged 6% in a single day, with over 40 stocks hitting their daily limit, and 13 stocks doubling.

UBTECH's 990k-yuan humanoid robot has pre-sold 13k units. Tesla's Optimus is set to start mass production at the end of the month. Unitree's IPO registration has just been approved. Morgan Stanley has raised its shipment forecast twice in six months, from 14k units to 50k units.

The industry seems red-hot, but do you know what the people actually building robots think?

Zhou Jian, founder of UBTECH, said something a couple of days ago, to the effect: We don't even have a dedicated chip for embodied intelligence. We have no learning data from the physical world, and we haven't built a physical AI yet.

What we're seeing now is just a pile of weak machines standing up.

That includes Unitree. A bunch of videos show Unitree's robots doing all kinds of cool moves — flipping, running, dancing — getting people pumped up. Then someone noticed there was a person behind it controlling it with a remote. I recall a comment from a netizen saying, "What's the difference between this and the remote-control cars we played with as kids?"

Of course, this jab isn't entirely fair. The robot's ability to maintain balance and perform those actions is already an impressive engineering feat — that underlying capability isn't granted by the remote control.

But the disappointment is real. The gap between "remote execution" and "autonomous decision-making" is exactly what Zhou Jian was talking about: dedicated chips, physical AI models, and massive amounts of real-world data. Right now, all of those are blank slates.

When I saw this information, one image came to mind: autonomous driving in 2016.

It was the same back then. Baidu, Waymo, Uber were burning cash. Demos were everywhere. The capital markets gave sky-high valuations. Everyone was shouting that it would be rolled out within three years. A decade later, only Tesla and Waymo have truly found a closed-loop business model. The companies that died along the way could fill a documentary.

Robots are standing at that exact spot now. The direction is absolutely right. But the road from "standing up" to "actually working on its own" is far longer than what those excited people on Twitter think.

UBTECH itself has said that fully functional household robots are still 5 to 10 years away. The U1 it's selling now can't walk or do housework. What you're buying for 990k yuan is essentially an advanced silicone doll that can chat.

A few days after the launch, the government issued regulations on anthropomorphic product management, restricting emotional companion AI, blocking that path as well.

Betting on which robot company will survive at this stage is like betting on which autonomous driving company would win ten years ago — the odds aren't great.

It's worth paying attention to. But you need to have the right posture.