embodied AI

BYD drew a robot. The drawing you should study is the dealership floor.

15 October 2026

BYD filed a design patent this week, and the drawing matters less than where BYD says the robot will go.

On 9 October, China’s National Intellectual Property Administration published an “Embodied AI Robot” design patent with BYD as the holder, according to CarNewsChina and Star News Korea. The drawings show a black-and-white humanoid with a rounded, screenless head and jointed torso, arms and legs, designed to walk, balance and interact with people. It looks very like “Xiao Di”, a robot a BYD executive teased on social media earlier this year. Figures disclosed earlier put Xiao Di at 1.61 metres and 58.5 kilos, with 31 degrees of freedom and a hand accurate to plus or minus one millimetre. BYD has not confirmed that the patent shows the production model.

Most carmakers now have a humanoid in a slide deck. BYD has also said where its one will work. According to CarNewsChina, its executives plan to deploy these robots across the company’s own dealership network, to greet customers, explain the cars and demonstrate the in-car infotainment. They are meant for the showroom, not the factory.

That makes this a retail story. For twenty years the dealership has been the weak link for premium car brands. Service varies from one site to the next, and a carefully built brand can be undersold by whoever happens to be on the floor that Saturday. BYD wants to automate the one role no carmaker has ever controlled: a salesperson who never has an off day, never improvises the pitch, never forgets the script, and demonstrates the infotainment the same way in Shenzhen as in Stuttgart. The robot is the headline. The bigger change is brand control reaching the last metre of the sale, the one part the brand has never owned.

China matters here because of how BYD builds. It is the world’s largest integrated maker of electric vehicles, and it is making this robot the way it makes everything else: in-house, at scale and from its own parts. It set up a dedicated robotics team in 2022, invested in AgiBot in 2023 and took a stake in Percibot in 2025. It has also been running Ubtech’s Walker S1 inside its own Changsha plant to collect operating data. BYD already makes the batteries, motors and actuators and runs the factory floor, so the robot is simply the next product down the same line. A Western carmaker partners with a robotics firm and negotiates over the margin. BYD owns both sides of that deal, and Western boards keep underestimating what that is worth.

There is reason for caution. A design patent is a drawing. There is no price, no reliability data and no deployment date, and Xiao Di may never get beyond the teaser video. A hand accurate to a millimetre still has to survive a showroom full of children and spilled coffee. So far BYD has shown renders and no revenue, and scepticism is fair.

The intent is still clear. BYD wants to know whether a humanoid can be the brand’s own face in the room where the sale happens. That is a far more commercial question than whether one can work in a warehouse, and almost no Western retailer or carmaker is seriously asking it yet.

What to watch. See whether BYD puts one of these robots on a dealership floor before it puts one in a factory. The order of deployment will show what BYD thinks the robot is for. If customers meet it first, BYD is selling brand control rather than cost savings, and that is the version retail executives should worry about.

The Roth Read. Picture the robot on your own shop floor, in your uniform, saying your words every time, without fail. BYD has said its humanoid belongs where the customer decides, in the showroom. If your brand still treats the last metre of the sale as somebody else’s problem, BYD is about to show you what that assumption costs.

The carmakers ran out of margin. So they went looking for a body.

1 September 2026

Read the balance sheet, not the backflip

A car company does not raise nine hundred million dollars for a robot because the robot is ready. It raises it because the car has stopped paying, and everyone in Shenzhen can read the same balance sheet.

This week Xpeng’s robotics unit raised more than $900 million at a post-money valuation above $6.3 billion, in a round led by IDG Capital with Tencent, Alibaba and Gaorong Ventures alongside. The company calls it the largest single private financing ever recorded in China’s embodied AI industry. It is not alone. AiMOGA, the robotics arm of Chery, is reported by Reuters to be preparing an IPO. BYD has unveiled a humanoid called Xiao Di. Changan, GAC, Li Auto, SAIC and Seres are all, per industry reports, building humanoids of their own.

The Western press files this as China chasing Tesla. Read the quote that matters instead. Michael Dunne of Dunne Insights, a man who has spent his career inside this market, told TechCrunch why Xpeng’s founder moved: “He sees razor-thin profit in cars on the near horizon. Robots look much more promising.” That is not ambition talking. That is a man doing arithmetic.

Now the part worth arguing with. A car is not naturally a commodity. It became one for these companies because they let it become one. A brand is a feeling, a badge, a way of being read at the school gates, and every point of that feeling is a point of price you do not have to give away. In China’s EV war some manufacturers gave it  away anyway, quarter after quarter, until the only ground left to fight on was cost. Once a company reaches that ground it is no longer a carmaker. It is a contract manufacturer of batteries, motors and control systems on wheels, and a humanoid is the same components in a different shape. Which is exactly Dunne’s point: “They have all the hardware to get the job done.” Batteries, actuators, motors, the control stack, all of it flows straight out of a mature EV supply chain that the West does not have and cannot conjure in a quarter.

So the structural lesson is this. When your product is treated as a commodity you cannot defend the margin, so you move the factory to the next product that still has one. The supply chain turns out to be the asset and the vehicle was only its current shape. A country that owns the batteries and the motors can change that shape far faster than a country that owns only the software.

But look at what a brand actually buys in that story. It buys the choice to stay. The manufacturers whose badge still commands a premium are not scrambling into robotics, because their cars still pay them. Brand is the thing that stops a category collapsing into cost, and it is the cheapest insurance a manufacturer will ever hold. The Chinese groups now pivoting are not proof that brand stopped mattering. They are proof of what it costs when you stop investing in it.

The other gap cuts both ways. Dunne names it plainly: the question is whether they can catch Tesla, and by extension the American labs, “on the AI side of the equation.” Hardware is China’s, for now. The brain is still contested. But notice which problem is easier to buy your way out of. You can hire researchers. You cannot hire a decade of battery plants.

For anyone running a store, the read is sharper still. The moment a humanoid works on a shop floor or in a stockroom, the retailer’s biggest line of cost stops being staff and starts being units. That is not a labour story you can hand to an HR memo. It is a question about what your store is for when the person who once greeted the customer is a leased machine amortised over three years. The answer is the same one it has always been. People go where they are known, and a machine on the floor only frees your people to do the part a machine cannot.

What to Watch The IPO, not the demo. AiMOGA filing to go public is the real signal, because a prospectus forces the question every backflip video dodges: where does the money come from, and when. A robot that can dance is a hobby. A robot with a revenue line is a business. The market is about to make several of these companies write the number down.

The Roth Read. Stop watching China’s robots and start reading China’s balance sheets, because the carmakers already have. The humanoid is rarely the threat. It is the tell: somebody with a better cost base is about to enter your category wearing a new shape, and they can only do it because that category let its brands become interchangeable. So ask which of your rivals owns the supply chain. Then ask the harder question. If your badge came off your product tomorrow, would anyone still pay more for it? That answer is the margin you are actually defending, and it is the only one nobody can build a factory to take from you.

China did not spend last week building video tools. It was building eyes.

5 August 2026

Every retailer who watched three Chinese labs ship video models in seven days filed the news under marketing. Cheaper ads, faster content, a problem for the agency. Wrong drawer. What landed last week was the perception layer for the machines that will one day walk your shop floor, and it landed open, and it landed cheap.

The launches came within days of each other. ByteDance’s Seedance 2.5 now generates thirty seconds of video with sound in a single pass, holding characters, scenes and camera logic together across a whole narrative rather than one lucky shot. MiniMax’s H3 does fifteen seconds with stereo audio generated jointly rather than bolted on, and on 3 August MiniMax put the weights on Hugging Face for anyone to download. Alibaba closed the week with Qwen3.8-Max, 2.4 trillion parameters, which now sits second in the world on the public leaderboard for reading images and visual material.

So run the telescope the other way. A model that keeps thirty seconds coherent, objects that persist behind an obstacle, weight that falls the way weight falls, a cup that is still on the table after the camera moves, has not learned to draw. It has learned how the world behaves. Generation is only the exam. The syllabus is physics, permanence and consequence. And the same weights that let a machine imagine a scene let it read one.

Reading a scene, fast, in bad light, with a person moving through it, is the entire job of a robot’s eyes.

That is the triangulation, and it is why this is a retail story rather than a media one. China already holds the other two legs. By industry counts it ships the overwhelming majority of the world’s humanoid robots, and TrendForce expects Chinese output to nearly double this year, with Unitree and AgiBot taking around eighty per cent of shipments. It holds the motors, the batteries, the rare earths, and the appetite to put machines in public before the ethics committee has finished its report. What it lacked was sight worth putting behind the visor. It is now building that in the open and giving it away.

Price finishes the argument. DeepSeek’s V4-Flash update, the least photogenic of the week’s launches and probably the most consequential, costs roughly three cents to run the full Artificial Analysis intelligence battery, against $3.15 for the Western frontier. A robot has a battery, not a data centre. Perception has to be almost free before it can live inside a body on a shop floor, and last week it became almost free.

A caution worth keeping. None of this means the machine understands anything. Video models still get physics wrong in ways that are amusing in a clip and unacceptable in a machine holding a bottle near a customer’s child. A convincing picture of a grasp is not a grasp, and that distance is where the next two years of the argument will be fought.

The direction, though, is not ambiguous. The robot that eventually watches your shelves, greets your customer and judges whether that customer is confused or annoyed will not be running perception you bought from a vendor you can name. It will be running weights someone downloaded, most likely trained in China, tuned by a supplier three tiers below the name above your door. And it will still be your brand doing the looking.

What to watch. Not the next video demo, and not the next leaderboard. Watch for the moment a Chinese maker ships a robot whose perception layer is one of these video models, and says so. When those two industries start sharing a checkpoint, the cost of machine sight collapses the way the cost of machine text already has, and every question retailers assumed they had until 2030 arrives early.

The Roth Read. Stop treating AI video as a marketing line item and start asking who supplies your machines’ eyes, because you are about to buy vision the way you buy electricity: from someone else, invisibly, with no say in how it was made. Put that question on your risk register this quarter. The machine watching your customer speaks for you, whoever trained it.