So last night, humanoid robots did kung fu, drunken boxing, and triple backflips on the Spring Festival Gala — basically China’s Super Bowl, nearly a billion viewers. Naturally the internet lost its mind. Western media went straight to “Beijing’s industrial policy machine.” Musk called China “an ass-kicker next level.”

Look, I get it. The videos are wild. But having worked in robotics in both countries, I think almost everyone is getting the why wrong.

And there’s a fun irony here: a lot of the people now panicking about China’s robots are the same crowd that mocked China for not being able to build good foundation models. First it was “they can’t innovate,” now it’s “we’re doomed.” Neither reaction is useful.

China’s robots aren’t advancing this fast because of some grand government masterplan. It’s the same reason the U.S. leads in AI: infrastructure.

America has more GPUs, so American labs run crazier experiments. China has the world’s densest hardware supply chain, so Chinese teams can break more robots, try dumber stuff, and bounce back faster. That’s basically it.

Crediting political systems for innovation is silly — nobody thanks democracy for GPT-5, and nobody should thank the Five-Year Plan for robot backflips.

It’s the Infrastructure. Period.

Last year’s Gala, Unitree’s robots did a wobbly folk dance twirling handkerchiefs. This year: swarm coordination at 4 m/s, drunken boxing, and aerial backflips while swinging weapons near children’s heads. That kind of leap in twelve months doesn’t come from genius alone. It comes from the ability to fail cheaply and recover fast.

I’ve been in the U.S. robotics startup world. Let me tell you what that looks like on this side.

I know a really good American research team that was doing aggressive full-body control experiments with Unitree robots about a year ago. They fried three machines. Then they just... stopped. Not out of ideas — out of patience. Getting replacements shipped across the Pacific, clearing customs, coordinating repairs — the iteration cycle went from days to weeks to “you know what, let’s just not try that move anymore.”

At the YC robotics startup where I worked, we tried making parts locally in the Bay Area. It was slower than shipping from China. We found some U.S. suppliers eventually — scattered around the country, inconsistent quality, hard to count on. A replacement actuator that a Hangzhou team gets by tomorrow morning? For us, that was a multi-week adventure.

That’s not a talent gap. That’s a logistics gap. Chinese engineers and American engineers are both incredibly talented — I’ve worked with both, and the difference in raw ability is negligible. What’s not negligible is how fast they can turn ideas into physical prototypes and iterate on them.

The Pearl River Delta around Shenzhen, plus the Yangtze River Delta around Shanghai and Hangzhou — where Unitree is actually based — these two clusters produce motors, sensors, actuators, custom PCBs at speeds that feel like cheating. When Unitree wants to test a new joint design, they walk down the street.

That’s why China shipped 90% of the ~13,000 humanoid robots sold globally last year. That’s why Unitree’s G1 costs $16,000 while Tesla’s Optimus Gen2 will be over $20,000. That’s why Unitree plans to ship 10,000–20,000 units this year.

Not strategy. Not policy. Just two insanely dense manufacturing ecosystems doing what they do.

It’s Not Just the Money

Here’s the other narrative I keep hearing: “China is winning because of unlimited state subsidies.”

This is lazy thinking.

Yes, Chinese robotics companies have raised from state-backed funds — Unitree’s backers include CITIC, China Mobile, and others with government ties. But let’s be real: the U.S. is not poor. Silicon Valley has significantly more capital, much deeper pockets, and — frankly — far more sophisticated investors than the Chinese venture ecosystem right now. American VCs have decades more experience pricing risk, backing high-variance bets, and giving founders room to fail fast. If this were a pure spending contest, America would win handily.

In fact, China’s state capital comes with baggage. State money is classified as “state assets” (国有资产) — founders who take it face serious personal liability if things go south. This pushes a lot of capital toward politically safe bets rather than the most innovative companies.

But here’s the thing that actually matters: the difference isn’t how much money you have, it’s how far that money goes.

In the Bay Area, a million dollars buys you a few months of runway and maybe one major hardware iteration. In Shenzhen, that same million buys you ten iterations. Same engineers, same ideas — wildly different output.

The “state subsidy” argument is a coping mechanism. It’s easier to say “they have unfair government money” than to admit “their ecosystem lets them do R&D 10x cheaper than we can.”

So strip it all away: both sides have capital, both sides have talent. What’s left? The supply chain.

So What Now?

Step one: just accept it. China’s robotics industry is doing really well right now. You don’t need exotic explanations — not “state capitalism,” not “they’re copying,” not “it’s all hype.” It’s real progress, driven by real engineering, on top of real infrastructure.

The sooner the West stops looking for reasons to dismiss it, the sooner it can figure out how to compete.

Step two: think clearly about what competing actually means.

A friend who founded a U.S. humanoid startup recently tweeted that “nuking El Segundo and giving green cards to like 1000 of the right people would probably go a long way.” More “sharks in the water.”

I get the energy. But sharks need water.

Drop 1,000 brilliant robotics engineers into the Bay Area tomorrow and they’ll hit the exact same walls: multi-week waits for parts, robots they can’t replace quickly, half their time wasted on logistics instead of actual research.

This is America’s problem to solve, and it has to be solved by building American infrastructure. Foreign talent helps on the margin, but you can’t graft a supply chain onto green cards.

And this stuff takes time — years of boring, unglamorous work. Building a network of local suppliers and figuring out how to make them work together. Relentlessly driving down cost and delivery time the way Chinese manufacturers have been doing for decades. None of it makes for good tweets or pitch decks.

The goal is dead simple to say and brutally hard to do: make it so a robotics lab in SF, Pittsburgh, or Boston can get a replacement part in two days, not two weeks. That’s not a DARPA problem. Not an immigration problem. It’s an industrial base problem.

Until that changes, expect more viral robot videos from across the Pacific. Not because China cracked some code or stole it from the US. Because they made it cheap and fast to break stuff and try again.

In engineering, that’s most of the game.