The Humanoid Economy: Robots, Revenue, and Reality

Zach Martin

This is K M K T I R Hub Radio, and this is Robot Capital. I'm Zach Martin. Today's subject is the humanoid and physical AI economy: the machines leaving the screen, the contracts that pay for them, the capital that funds them, and the magnets, motors, and power that limit them. One note before we start. Everything you hear on this show is informational market intelligence. It is not investment advice. Do your own homework, and talk to a licensed professional about your own situation. Alright. Let's build this from the floor up.

Zach Martin

For years, AI lived behind glass. It answered questions, drafted emails, and made pictures. Nothing it did had weight. Now it's walking into buildings that do have weight, and gravity, and doors that stick. That's the beginning of the humanoid economy. The interesting part is where these machines have to work. They have to work in brownfield spaces, meaning buildings that already exist and were designed around a human body. Doors, stairs, aisles, shelving, carts. A humanoid is a bet that if a machine fits the human body plan, you don't have to rebuild the building around it. Traditional industrial robots are usually task specific and need a controlled setup. Humanoids try to shrink that integration burden. Whether they can is the open question, and we'll keep coming back to it.

Zach Martin

Why now? Part of the answer is demographics, and I want you to hold onto these numbers. Across the OECD, the old age dependency ratio, that's people aged 65 and over per 100 working age people, was about 19 percent in 1980. By 2023 it was about 31 percent. The projection for 2060 is about 55 percent. Fewer workers supporting more retirees, right when demand for logistics, healthcare, and elder care keeps rising. So the pitch for humanoids is synthetic labor capacity: a way to keep output up when the workforce can't grow. That's a pitch. It's not a result. But it explains why serious capital is paying attention.

Zach Martin

Now, how many robots are we talking about? Here's where I'd ask you to stay disciplined, because the forecasts are all over the map. Goldman Sachs has a bear case of 700,000 humanoids a year by 2035, and a base case of 1.4 million. The Bank of America Institute projects 10 million a year by 2035. And Elon Musk has predicted 10 billion humanoid robots by 2040, which would imply close to a billion built per year through the 2030s. When forecasts span that range, the honest reading is that nobody knows yet. The players are real, though. Tesla with Optimus. Figure AI. Boston Dynamics with its electric Atlas. Agility Robotics with Digit. Apptronik with Apollo. Sanctuary AI with Phoenix. And in China, companies like Unitree and Fourier. Real teams, real prototypes. The question is who gets paid, and how.

Zach Martin

To see how, look at warehouses, where robots are already earning money. In 2012, Amazon bought Kiva Systems for 775 million dollars in cash. The message to every other merchant was clear: automate or fall behind. But selling robots the old way was brutal. A workcell might cost 75,000 to 250,000 dollars or more. That's a capital expense, and capital expenses go to the CFO. Sales cycles ran 12 to 18 months. Many deals died in what the industry calls pilot purgatory, where a risk averse finance chief says, let's test a little longer, and the test never ends.

Zach Martin

So the business model changed. Robots as a Service turned that big capital request into a monthly operating expense, something a facility manager can approve. Locus Robotics is a good example of the structure. Agreements are typically multi year, often three year minimums, paid in advance, with implementation fees on top for integrating with the warehouse management system, planning the layout, and getting the site ready. Think about what that means. The robot isn't the product. The product is a pick rate. The vendor isn't selling hardware once. It's running a service for years, and it's responsible for uptime.

Zach Martin

There's a smart wrinkle in this, which I think of as the operational Trojan horse. A warehouse leases extra robots for the holiday surge in the fourth quarter. The robots help pickers walk less, by as much as 12 miles a day in some deployments. The operations team adapts. The warehouse management system adapts. Pick rates go up. Then January comes, and returning the robots means going back to a slower baseline that everyone now knows is slower. So the temporary lease becomes a higher permanent baseline. That's stickiness, and stickiness is what recurring revenue is made of.

Zach Martin

Some vendors go further and bolt themselves into the building. Mobile robots are flexible. You can move them. But a system like Symbotic's, with high density grid structures and autonomous robots running on them, is anchored to the warehouse's concrete floor. Taking it out means demolition and months of downtime. That's a very different kind of moat, and it shows up in the numbers. Symbotic has reported more than 22 billion dollars in contracted backlog. Flexible robots give you easy adoption. Embedded systems give you lock in. Investors should know which one they're looking at.

Zach Martin

Now let's talk about the part that keeps founders up at night, the subscription cash crunch. I'll use an illustrative number, not any one company's. Say it costs 100,000 dollars to build and install an enterprise robot: the arm, the sensors, the compute, the safety integration. The customer pays somewhere between 1,500 and 5,000 dollars a month. At a midrange 2,500 dollars a month, payback takes about 40 months. The range people talk about is 36 to 48 months. That's three to four years before a single machine has paid for itself, and that's before field support, software maintenance, and depreciation.

Zach Martin

Scale that up and you see the trap. Deploy 100 units and you've put 10 million dollars into the field. At 2,000 dollars a month per unit, that fleet brings in about 2.4 million dollars in its first year, gross. Software companies get to say that growth creates cash. Physical subscriptions do the opposite. Growth consumes cash first, and the faster you grow, the deeper the hole. It's like buying rental houses with the rent coming in slowly. Each house is a fine asset. Buy too many too fast, and the mortgage payments arrive before the rent does.

Zach Martin

What do you fund that with? Venture equity is expensive, with an implied cost of capital of 20 to 40 percent. Using it to buy depreciating metal and wiring dilutes founders badly. So smart hardware companies move their capital stack toward asset backed debt, special purpose vehicles, and equipment leasing, the same tools used to finance trucks and aircraft. There's a reward for getting it right. Data from SVB and S2G suggests hardware as a service companies can command a valuation multiple premium of about 59 percent over transactional hardware peers, because recurring revenue is better quality revenue. But that premium disappears if the company runs out of cash before it reaches breakeven. A great multiple doesn't pay the electric bill.