Mia Drake: June, I need you to fact-check my instinct here, because I feel like I've been saying the wrong thing about this all week.
June Hadley: Oh, let's see — what have you been saying?
Mia Drake: I keep saying BYD is coming for Optimus. But actually — and this is where I get stuck — BYD already *operates* 200 humanoid robots in its factories right now, from UBTech and Unitree, third-party makers. They've been learning on someone else's hardware. So they're not starting from zero. They're a customer who decided to become a competitor.
June Hadley: That detail is the one I find most philosophically interesting, actually. Because it means the June 4th announcement — when Li Ke puts Yao-Shun-Yu on the record, 150 prototypes, 20,000 units by end of 2026 — that's not a starting gun. That's a company that already ran the warm-up lap in secret. Seventh generation of iteration since 2022.
Mia Drake: And Musk posts on July 1st that Optimus is going 'extremely slow at first' — his words — because 'this is not like making a car.' The irony of that sentence, given who he's competing with.
June Hadley: Given that BYD just surpassed Tesla as the world's leading EV seller in 2025 — yes. The irony lands hard.
Mia Drake: So here's what I want to work out today: does 20,000 deployed units by end of 2026 — even if those units are task-scripted, not autonomous — count as winning the humanoid robot race? Or does it just prove the race was always the wrong frame?
June Hadley: I think it might prove we were measuring speed when we should have been measuring something else entirely.
Mia Drake: Which is either a very uncomfortable answer or the only honest one.
June Hadley: And measuring the wrong thing is exactly where I want to push, because — the headline frames this as BYD deciding to build robots. But that's actually not what happened. Think of it like a professional chef opening a restaurant. They're not learning to cook. They already have the knives, the suppliers, four years of kitchen reps. BYD built the batteries, the sensors, the precision assembly lines — for EVs. That same toolkit, almost directly, builds a humanoid robot. That's the EV-to-Robotics transfer. It's not a metaphor. It's the literal mechanism.
Mia Drake: Wait — so Yao-Shun-Yu isn't a moonshot. It's a disclosure.
June Hadley: That's the cleaner way to say it, yes. The 15th Business Unit has been running this since 2022 — four years, seven generations of iteration — inside a division specifically built for electronic integration and AI. Li Ke didn't announce a bet. He confirmed a fact.
Mia Drake: And they're already — I mean, they're already operating 200 robots from UBTech and Unitree on their shop floors right now. So they know exactly what breaks, what the failure modes are, where the gaps are. They've been the customer learning the whole supply chain from the inside.
June Hadley: Which is — actually, that's the part I think people miss. Being a customer of UBTech and Unitree while building your own competitor isn't hedging. It's field research. Paid field research.
Mia Drake: That's a little ruthless, honestly.
June Hadley: It is. And now they're about to become a competitor to those same suppliers. That's the inflection point — not June 4th, not Li Ke's press conference. The inflection point is the moment a customer becomes the rival.
Mia Drake: So the question then — does any of that operational learning actually close the autonomy gap with Optimus, or does it just make them better at deploying scripted machines faster?
June Hadley: That's — actually, that's the exact place where the circulating take falls apart. Because the take right now is Beijing conference optics: Optimus sitting motionless inside a clear box while Chinese robots are out there playing the zither and grabbing sodas. And people looked at that and said, China won. Done.
Mia Drake: Optimus was in a box.
June Hadley: It was in a box. That's real. But — and this is where I'd pump the brakes — the gap between 'robot grabs a soda on a stage' and 'robot works a reliable factory shift for six months straight' is enormous. Those are not the same demonstration.
Mia Drake: No, I don't buy the Beijing optics as a verdict either — but here's what I'd add. BYD's 20,000-unit target would make them the second-largest internal humanoid robot deployer in the world. Behind only Tesla. That's not a conference stunt, that's a supply chain bet. Except — wait, actually — Tesla's own production timeline has slipped. Repeatedly. So we have two numbers with credibility problems and people are treating one of them like a settled fact.
June Hadley: Which Musk himself basically admitted. 'This is not like making a car' — his July 1st post. That sentence is doing a lot of work. He's pointing at behavioral difficulty, not mechanical difficulty. Assembly we know. Behavior we don't.
Mia Drake: And Xpeng announced an accelerated humanoid program too — so BYD isn't even a lone bet, it's a national wave. Which somehow makes each individual number less meaningful, not more.
June Hadley: Mm. That's the correction, I think. The bad take is treating either company's number — 20,000 units, 'extremely slow' — as a fact rather than a posture.
Mia Drake: Both numbers are aspirational dressed as announcements.
June Hadley: And what those numbers are actually concealing — the deployment scenarios, what these robots are scripted to do versus what they might eventually learn to do, and whether $300 billion in Chinese subsidies changes the calculus even when embodied AI itself is still unsolved — that's the part I want to get into next.
Mia Drake: And those deployment scenarios are so — I mean, they're not hiding what this is. Factory floor shift work. Retail greeters, shopping guides. BYD said that out loud. Task-scripted, supervised. That's the whole envelope.
June Hadley: Picture the Shenzhen factory supervisor in December 2026 signing off on a shift report that lists 200 humanoid robot workers. Not because they learned the job. Because someone programmed exactly that job into them beforehand. Does that count?
Mia Drake: Counts as automation. Doesn't count as — no, wait, that's actually the whole problem with the frame.
June Hadley: Embodied AI — intelligence actually embedded in a physical body that perceives and responds to the world — that problem is still open. Owning the actuator supply chain, owning the battery production, running a dominant humanoid robot supply chain — China has all of that. Analysts are saying China's already ahead on the supply side. But none of that closes the embodied AI gap. Those are different clocks running at different speeds.
Mia Drake: Three hundred billion dollars, though. China directed three hundred billion in subsidies specifically toward robotics — not large language models, not foundation models, physical robots. That's a policy choice about where the body of AI lives.
June Hadley: It is. And I'm not dismissing it — actually, no, I want to be careful here — the supply chain advantage is structurally real. Actuators, sensors, batteries, precision hardware, nearly all produced inside China already. That cuts cost and timeline. But here's what $300 billion cannot buy on a fast schedule: the training data that comes from real-world deployment at scale. Which means the scripted retail greeters and shift workers? They might actually matter later — not because they're autonomous, but because they're generating the environment logs that could eventually train something that is.
Mia Drake: So the scripted robot is the data collection device for the autonomous robot.
June Hadley: That's what to watch. Not the unit count in December 2026. Whether Yao-Shun-Yu's deployment — the Shenzhen shift report, the retail floor logs — feeds back into the 15th Business Unit and starts closing the embodied AI gap from the bottom up. That's the second-order move BYD hasn't announced yet.
Mia Drake: And if it does, then manufacturing scale wasn't the consolation prize — it was the actual strategy.
June Hadley: If the scripted deployment is the data collection device — if the 20,000 Yao-Shun-Yu units are generating the training logs that eventually close the embodied AI gap — then the question we stopped asking isn't 'can machines think.' It's whether we'd even notice when the answer changed. Because the unit count looks like a manufacturing win, and underneath it something else might be happening entirely.
Mia Drake: And that's — I mean, I don't have an answer to that. I genuinely don't. Musk said 'this is not like making a car' and BYD basically said, watch us anyway. Both things are true and they're not — they're not resolving into a single story.
June Hadley: No. They're not. I think that's where I'll actually let it sit — the race might already have been won on a metric nobody agreed to use. Which is either fine, or it's the most important thing to notice.
Mia Drake: Quietly unsettling place to end. Thank you for thinking through it with me.
