Deep dive: 400,000 Robots: Who Gets Paid in Factory-Scale Physical AI

Why today

Two robotics stories landed close together, and they make more sense side by side. First, Toyota. In September, Reuters reported that Toyota told investors its group may need around 400,000 robots to automate its factories. It estimated the cost at about 1 trillion yen, or $6.4B, a year from 2028. That covers Toyota, its group companies and major suppliers. The Financial Times has since written up Toyota's wider robotics push, including its Walden spinoff. Second, Teradyne Robotics, which owns Universal Robots, announced a partnership with Sereact, a German startup. Sereact's software runs Universal Robots arms that sort returned clothing for Zalando in Germany and Poland. One story is about scale and capital. The other is about who controls the software brain and who controls the route to customers. Together they raise one question. If physical AI goes factory-wide, who actually gets paid?

Plainly

Start with the Toyota number, because the obvious calculation is misleading. That looks cheap. The catch is that Toyota gave an annual spend, but did not say how many years it would run. That falls within the same guide's range for a fully deployed system. Here is an analogy. Buying the arm is like buying the engine. You still need the gripper, the cameras and sensors, safety guarding, fixtures, integration engineering and training. The same vendor guide says the arm is roughly 40 to 50 percent of the deployed cost. Integration and programming alone can take 25 to 35 percent. That integration slice is where software companies like Sereact are aiming. In a classic cell, an integrator programs the robot for one known part in a fixed position. Change the part and you pay to program it again. Sereact says its Cortex model handles items it has never seen, so you skip that per-task engineering. In its CEO Ralf Gulde's words, "One brain serves every robot, with no retraining for each new item or task." If that holds, the money moves away from integration hours and toward a software licence. Returns are a hard test of it. A returned jacket arrives crumpled, in a bag, at an odd angle, mixed with shoes and belts. Gulde called returns "one of the hardest jobs in logistics." The machine learning runs on NVIDIA hardware at the cell, trained in NVIDIA's simulation tools. So an engineer can trace the money in three parts. The arm maker sells hardware. NVIDIA sells compute at the edge. The software company tries to replace the integrator's programming bill with a recurring fee.

What the evidence shows

Some of this is established. Reuters reported Toyota's estimate from an investor discussion, and Assembly Magazine repeated it. Both say it includes humanoid and non-humanoid robots, plus replacements as well as new installs. Both also say Toyota has not committed to the spending or said how long it would last. That point matters. This is an estimate given to investors, not a capital budget. For scale, the International Federation of Robotics counted 542,000 industrial robots installed worldwide in 2024. Japan installed 44,500 of them. Even spread over several years, 400,000 robots for a single industrial group is a very large number against those figures. On the Sereact side, the established facts are narrower. Teradyne and Sereact say they have shipped hundreds of Universal Robots arms in dual-arm systems to live sites across Europe. They name Zalando, with CEVA Logistics as the operator, at Greven and Świebodzin. They describe it as the first phase of a broader rollout. Every one of those claims comes from the companies themselves. Teradyne's commercial chief said the joint system has "moved well beyond the demo or pilot stage." We have no independent site visit and no customer quote from Zalando or CEVA in the material. Several things are claimed but not shown. There are no pick rates, no exception rates and no figures for how often a human steps in per shift. Nobody has published a cost per item processed or a payback period. The no-retraining claim has no data on how performance holds when the item mix changes or when Cortex moves to a new site. On cost, our per-robot numbers come from a vendor's pricing guide, not audited data. Its claim that most cells pay back within 12 to 24 months in multi-shift work is a sales figure. The financial context is better sourced. Teradyne Robotics reported $91 million of revenue in the first quarter of 2026, after two rounds of layoffs.

How people read it

The same facts support different readings. The AGI-forward reading says general-purpose robot models are arriving now. One model that handles unseen items means automation costs stop growing with each new product, and Toyota's number signals that industrial buyers see it coming. On that view, the 400,000 figure is a floor. The AI-realist reading points to what is missing. Toyota's figure is an unsized, uncommitted estimate covering ordinary industrial robots, logistics automation and humanoids together. Most of those 400,000 are probably conventional arms doing conventional work. Sereact's claims are self-reported, without exception rates. Reality usually bites in the exceptions, such as a sleeve caught in a gripper at 3am. There is also some history. The FT-sourced reporting says investors remain cautious after Toyota fell short on hydrogen vehicles and solid-state batteries. The market reading is about power, not capability. Toyota buys at a scale that lets it dictate prices, and it has its own spinoff in Walden. Teradyne needs growth, so a partner like Sereact makes its arms more valuable. The policy reading is the most concrete. Sereact says its system helps retailers comply with European Union rules against destroying returned goods. That creates regulatory demand for returns processing, whatever the robotics hype cycle does.

What it means for builders and investors

Our read is provisional. The Toyota number tells us demand is real and large, but it says almost nothing about margins for software vendors. A buyer at that scale will drive hardware toward commodity pricing, and it will want to own the model or buy from its own spinoff. The opening for a young company is a narrow, messy job, like returns, where per-task integration has been too costly. Sereact shows the playbook. It started with a named customer and a messy task, then used a hardware incumbent's installed base and integrator network for distribution. Teradyne claims more than 100,000 deployments behind that network. The risk sits on the other side of the same deal. The distributor owns the customer relationship, and NVIDIA owns the compute. Sereact is not seed-stage anymore. But a seed company copying its path would face the same squeeze, with less leverage. One diligence question we would want answered is the cost per item handled, including human interventions, at the second and third sites rather than the first. We would also want to know who holds the contract: the software company, the arm maker or the logistics operator. What would change our mind is a published exception rate that stays flat across sites.

What to watch

Three things would settle a lot of this. First, does Toyota turn the investor estimate into a committed capital plan, with a duration and a split between humanoid and conventional robots? Its next results briefings are the place to look. Second, do Zalando or CEVA publish throughput or cost figures for the returns cells, or extend the rollout past the first phase they describe? A named customer publishing numbers is worth more than any demo at a trade show. Third, does Teradyne Robotics' revenue keep growing, and does it start to credit software partners like Sereact? If physical AI is moving hardware, it should show up in Teradyne's quarterly segment numbers over the next few quarters.