The Cost of Building Humanoid Robot Infrastructure
Robotics threads have moved past “will it work?” to a harder question. A composite of the version we keep seeing:
“Everyone quotes the sticker price, $16,000 or $20,000 a robot. But somebody has to build the factories, the parts supply, the data centers that train them. What does all of that cost? Who pays for it? And doesn’t that bill just get passed on to the rest of us?”
It’s the right question, because the sticker price is the smallest number in the story. Building a humanoid fleet means paying three bills: the body, the factory that makes bodies, and the training that makes a body worth having. And the reader’s last worry, that the bill lands on everyone else, depends less on any of those numbers than on how the robot gets priced.
The Body Is Getting Cheap
The average humanoid cost about $85,000 in 2023 and roughly $25,000 in 2025, a 70 percent drop in two years. Unitree sells its G1 for about $16,000. Tesla’s stated Optimus target is $20,000.
The best view of where that curve goes comes from a securities filing. When the paperwork to take Agility Robotics public through a SPAC merger was filed in September, the law made the company publish its own cost curve. The filing puts the cost to build a Digit at about $150,000 at commercial launch, with a bill of materials targeted at $75,000 at 1,000 robots a year and $30,000 at 10,000 a year. A robot costs what it costs because almost nobody is making them yet.
The Factory Is the Expensive Part
Volume needs factories, and factories are where the money goes. Tesla is spending more than $25 billion in capital in 2026 across AI, robotaxi and Optimus. It converted the Fremont lines that used to build the Model S and X, and Musk calls Optimus Tesla’s “hardest product to scale” because it needs an entirely new supply chain: about 10,000 unique parts, and hands and forearms that each hold more than 100 screws and small parts that people still put together by hand.
Everyone else is building too. Figure’s BotQ plant, designed for 12,000 robots a year, went from one robot a day to one an hour in under four months. Hyundai says it will build a US factory able to make 30,000 robots a year. In Germany, Bosch agreed to manufacture robots for the startup Humanoid, with Schaeffler covering more than half of its joint demand through 2031, a seven-digit number of actuators.
For scale, the Bootstrap Paradox article estimates one robot manufacturing facility at $500 million to $1 billion and two to three years, inside a $4-8 billion budget for an entire self-sustaining Free Zone. The same article warns against the shortcut of “robots building robot factories.” Self-replication works today for simple manufacturing. For complex machines it “won’t for decades.”
The Mind Is the Biggest Bill
The surprise of 2026 is the third bill. In August, Figure launched Index, an app that pays people in 108 countries to film themselves doing chores. It had paid out $15 million, roughly 94 cents a video, and committed to spending more than $1 billion on data and compute over the next twelve months. A week later, the AI-cloud provider Nscale agreed to supply Figure with at least $3.5 billion of GPU infrastructure over several years. The Humanoid Robot Revolution article sums it up in one line: “Teaching a body costs more than building one.”
From the Podcast: The Factory Finished First
The October 25, 2026 episode of Minds, Bodies & Terawatts, our daily podcast, put all three bills in one story. Its headline: Tesla Put Optimus Into Production on a Line Built for a Million Robots a Year. Musk Says the Hands Aren’t Finished, the Brain Isn’t Ready, and Customers Wait Until Late 2027.
On the earnings call, Musk said most of the equipment for a million-robot-a-year line is already in Fremont, and that reaching its rate of about 20,000 a week will take at least a year, maybe longer, because parts of the design are still changing. The episode sets that against IDC’s count of about 25,000 humanoids shipped worldwide in the whole first half of the year: one line doing close to six months of global output every week. Then it lists what the factory can’t buy. The hands aren’t finished. The software needs ten to twenty thousand robots practicing on each other in what Tesla calls an Optimus Academy. Berkeley’s Ken Goldberg put it plainly: you can produce them, but the software is lacking.
The guest’s summary fits the whole buildout: “It means the factory is further along than the product.” Or, shorter: “The factory finished first.”
Who Pays, and Who Gets the Change
So who pays? Today, investors. And they intend to be paid back by the customer, at a price that has nothing to do with the cost curve.
Agility’s filing states its pricing rule directly: subscription pricing “is generally structured at a discount to a customer’s fully burdened labor cost.” Digit rents for $8,500 a month, about $535,000 over a five-year life. The customer’s invoice stays the same whether the robot costs $150,000 to build or $30,000. The Wage-Anchored Pricing article does the arithmetic: eighty percent of the cost of building a Digit disappears between the first row of that cost curve and the last, and none of it is scheduled to reach the customer.
That answers the reader’s worry, and the answer is worse than the worry. What fails to get passed on is the savings. A buyer pays roughly what a worker costs, the vendor keeps the gap, and the deflation the abundance thesis depends on stays inside the vendor’s margin.
The article names three things that break the anchor. Competition, which the FCC weakened in July by keeping new Chinese humanoids off the US market. Ownership: the same filing shows that buying a Digit outright costs about $400,000 over five years against $535,000 to rent. And visibility, a buyer demanding cost-plus disclosure as a term of sale, the way utilities have opened their books to rate boards for a century.
On unscarcity.ai: Three Articles for This Question
The Humanoid Robot Revolution tracks the factory race: Fremont’s ramp, BotQ, Hyundai’s 30,000-a-year plant, and the training-data bill. Wage-Anchored Pricing: Why Cheaper Robots Don’t Mean Cheaper Goods reads the Agility filing line by line and asks who keeps the gap between cost and price. The Bootstrap Paradox prices a single Free Zone, robot factory included, and explains why the first one is the hardest to fund.
In the Book: Who Owns the Machines That Make Machines
Chapter 9 of Unscarcity, “The Enterprise EXIT (From Profit to Purpose),” takes up the question this post ends on. Once the factories exist, whose are they?
In the section “The Decision: Mission Guilds,” the book proposes a new organizational form built from four imperfect predecessors: Soviet planning, medieval guilds, open source and Mondragon’s cooperatives. “One Guild organizes physical production: factories, hospitals, infrastructure.” Open source proved that transparency and the right to leave can coordinate millions of people, “But try copying a car factory.”
The next section, “A First Day at the Sustainable Transport Guild,” follows the book’s invented scenario of an old Volvo plant in Ridgeville, South Carolina, where forty-seven welders, battery chemists and former managers sit around a mahogany table in the old executive dining room with no one in charge. The equipment changes status: “Volvo’s machines aren’t ‘owned’ by the Guild the way a landlord owns an apartment building. They’re held in stewardship, oriented toward the mission rather than the balance sheet.” If the Guild drifts, its resources can be reassigned “by published thresholds the members can see coming and contest, not by a planner’s verdict.” And every Guild pays surplus capacity into a solidarity fund, so that when drought hits West Africa, industrial capacity moves there. “Not as charity. As the system working as designed.”
That is the book’s answer to the reader’s last question. The robot factories will get built, with today’s money, at today’s prices. What gets decided is whether the machines inside them end up rented back at the price of a wage or held in trust and run at what they cost. The book works through how a company gets from one to the other over a decade.
More questions like this one are collected on the blog page. Follow the series on Substack.
