Why Elon Would Rather Rent Optimus Than Sell It

Most people assume Tesla will sell Optimus like a car. You pay, it leaves the lot, and Tesla moves on. One short answer from Elon Musk points to a completely different business.

At the 33rd Annual Baron Investment Conference in New York City on November 6, billionaire investor Ron Baron interviewed Musk over a live video feed. Baron suggested a robot costs less to build than a car. Musk said it would start out about the same and get much cheaper over time. Baron put the cost at 30,000 on the way to 20, and Musk said yeah. Then Baron asked whether Tesla would rent them, sell them, or do something else.

Musk answered, "I think I think probably rent them." I think that's the most important business model call in the whole Optimus story, because it changes what Tesla sells.

Probably rent them

When a manufacturer sells a machine, the relationship ends the moment it leaves the loading dock. When the manufacturer keeps ownership and rents it out, the company stops acting like a gadget maker. It becomes a labor utility, selling hours of work.

So try the hypothetical from the video. What would the US economy look like if you could hire infinite people at $2 an hour per person? I think that's what humanoid robots will do. And if that's right, Optimus gets priced against a wage instead of a sticker.

Five workers in one body

Before Baron's question, Musk explained how he thinks about physical AI, meaning software intelligence inside a moving body. Any economy, he said, is the number of people in it times the average productivity per person.

Now add a robot that works 24/7. Musk put that at 168 hours a week, with no vacation, no bathroom breaks, no food, and it moves faster than a person. A person works maybe 40 hours, takes breaks and moves slower. So Musk said a minimum productivity expectation for a humanoid robot would be five times that of a human.

Take that at face value and one robot delivers the weekly output of at least five people. In the US, government data put average pay for factory workers near $30 an hour and warehouse workers near $26, before benefits. If a machine does that work for a few dollars an hour in upkeep, selling it once gives away almost all of its value.

Why outsiders have to wait

So why not open the doors? When The San Francisco Standard reported on October 30 that Tesla would open Optimus reservations in early December, Musk answered a post about it on X with one word. No.

In its third-quarter 2026 update, Tesla said "Optimus production began in Fremont on our high-volume lines" and called these "the first production-quality bots that we have manufactured." The same update listed the California line as an early ramp and the Texas facility as under construction, with no unit count or customer price. On the October 21 earnings call, Musk said he expects the AI, design, reliability and usage life issues to be solved around the end of 2027, and that's when to expect deliveries to companies outside Tesla.

Why hold back? Because the cheapest, most reliable labor unit you have belongs in your own plant first. Tesla runs car and battery plants staffed by tens of thousands of workers. Every time Optimus takes a station moving sheet metal or sorting battery cells, a human wage becomes a robot hour. And Optimus builds Optimus.

There's a learning reason too. A walking robot can't pause midstep to ask a distant server where to step, so its brain runs on an onboard inference chip. That brain needs millions of hours of real-world trial and error, and a stumble in a family kitchen is a safety hazard. A factory floor can be roped off and watched by engineers. So the order is factory, then warehouse, then much later kitchen or bedroom.

The Colossus template

The answer is already running at Musk's other company. In February 2026, SpaceX acquired xAI, which builds the Grok model and owns X. Reuters reported the deal valued the combined private company at $1.25 trillion. Inside it sit giant Memphis-area supercomputers called Colossus, packed with GPUs, the chips that train and run AI.

In the same interview, Baron walked through the data center math. A conventional builder, he said, takes about two and a half years, spends something like $60 billion, and needs five-year leases to raise the capital. Speaking as an investor, Baron said "we build them" because "we're vertically integrated for our cars, for our robots, for our rockets," so the risk can be taken without outside contracts. Baron said that takes SpaceX eight or nine months and about $35 billion, and that SpaceX will rent compute out until Grok needs it. He said SpaceX revenues are "about to explode." Those are Baron's figures, not company disclosures.

SpaceX's prospectus, filed with the Securities and Exchange Commission in June, spells out the contract. SpaceX said it uses its compute for its own AI work, including training Grok-5, while renting select capacity to outside customers. Anthropic agreed in May to pay $1.25 billion a month through May 2029 for compute across the Colossus clusters. Then comes the key sentence. "This structure allows us to monetize a portion of the compute capacity in our infrastructure, while still permitting reallocation of that capacity for our own internal initiatives if needed in the future."

After an initial three months, either party can end the deal on 90 days' notice. On tape I put gross margins on rented tokens around 80 to 90%, with rental payments paying back construction in under a year. If SpaceX needs every chip later, it gives notice and takes the capacity back.

Now swap chips for robots. My view is that data centers, self-driving cars and humanoid robots are the same kind of object, hardware with an inference chip that generates intelligence and acts on it. Token factories for the physical world. When Optimus output outruns Tesla's own needs, I think Tesla rents the surplus hours to outside warehouses and factories while keeping title and the right to pull the robots back.

The quarter that explains the rental

Why does recurring revenue matter so much right now? Look at Tesla's third-quarter 2026 results. About $28 billion in quarterly revenue, almost flat year over year. Net income under standard accounting down 43%. My read is the drop came from the upfront cash autonomy and robotics require. R&D spending jumped 69% from a year earlier, and capital expenditures more than doubled to about $4.7 billion.

What carries earnings meanwhile? Full Self-Driving subscriptions. In North America the FSD attach rate crossed 75% of new deliveries, and active FSD customers, counting upfront buyers, hit 1.74 million. On tape I put that software's margins above 80%, because the car hardware is already paid for.

Now run that math on Optimus. Sell a robot for $30,000 and Tesla pockets a small hardware profit, then never sees another dollar from its work. Rent it and the robot stays on the balance sheet producing cash. On tape I put humanoid operating cost around $3 to $5 an hour early, falling toward $2 at scale. At about $2, any job the robot can do, it gets hired for, including work no human wants.

Rent a unit to a warehouse at half a worker's wage and the customer halves its labor cost while Tesla bills every hour. On tape I put global output near $118 trillion a year and human physical labor north of $40 trillion. One percent of $40 trillion is $400 billion a year.

Atoms don't forgive

So why doesn't every startup copy this? Because when you own the fleet, every broken joint, worn motor and dead battery is your bill. A chatbot's clumsy sentence gets a patch. In the physical world, atoms don't forgive the way bits do.

On Tesla's October 21 earnings call, Musk called Optimus growth an S-curve whose early part has a shallow slope and is bumpy, because it's such a radically new product. He said there are something like 10,000 things that have to go right, and production moves as fast as the slowest, least lucky of them. The robot has to be safe and reliable and can't wear out after a month, he said, and there's still a lot of AI work before it's truly useful for open-ended tasks.

For rental to pencil out, a robot has to survive thousands of hours of industrial shifts without a standing repair crew. That means solving the brain and the hand together. A human hand can lift a raw egg and then torque a heavy wrench. The robot brain has to read cameras in real time and balance every joint so a heavy machine doesn't slam onto concrete.

Nobody else is tooling up

Look across the 2026 humanoid industry. Agility Robotics rents its bipedal warehouse robot Digit as a service and also sells it, and SEC filings put its launch parts cost near $150,000 per robot, with a conditional rental order for 1,000 units. Figure, which tested robots at BMW, said in April its factory had delivered more than 350 robots, and in July it built its 1,000th. Hyundai, which owns Boston Dynamics, plans a plant aimed at 30,000 robots a year. In China, Unitree shipped over 5,500 smaller humanoids in 2025. Shipments and contracts sit in the hundreds or low thousands, and parts still cost a fortune.

Tesla built about 1.76 million vehicles in the four quarters through September. Musk said the vast majority of a million-unit-per-line production line has landed in Fremont, and it'll take at least a year, maybe longer, to reach 20,000 units a week. He said he'd be stunned if anyone has a robot remotely competitive with Optimus when deliveries to customers start, likely toward the end of 2027. Tesla's Vaibhav Taneja added that a factory under construction in Austin could build a lot more robots than Fremont.

Wall power for muscle

Nobody pedals a bike in the backyard to keep the fridge cold. You flip a switch and pay the electric company. I think physical labor moves to that model, with people paying a labor company for work done by autonomous robots.

The hours math is stark. A typical employee works about 1,800 hours a year. A humanoid on daily shifts with short charging breaks logs about 7,300. That's roughly four humans per robot. Musk ran it for the US. With about half of people in school or retired, he landed around 120 million Americans with net positive output, matched by about 20 to 30 million robots.

My view is humans will deploy physical labor instead of providing it. I think humanoid robot orchestrator becomes a giant job over the next 20 years. Pair Optimus with digital workers like Grok handling scheduling and supply chains, and one person can run a construction business renting compute by the token and labor by the hour. A small shop could rent three robots for a busy week and send them back.

The landlord problem

Now the hard part. A farmer who buys a tractor keeps driving it if prices rise next year. Rent Optimus and the customer never owns the worker. If one company builds most of the world's humanoids and keeps title, it becomes the private landlord for physical labor, and its hourly rate effectively becomes the baseline wage for physical work. Remember the 90-day clause. A warehouse that swaps its staff for rented robots can't rebuild a human workforce overnight if Tesla raises rates or recalls units.

Then there's control. Every rented robot carries cameras and sensors feeding data back to train the fleet, and because it connects to headquarters for updates, the owner can pause or shut it down remotely. On tape I noted Tesla has published no policy on who holds an Optimus kill switch, whether a customer can cut the network link, or how a remote operator is authenticated. If power cuts out, an actively balanced machine falls straight down, and ISO 25785-1, the international safety standard for such robots, is still a draft.

And the brain and body live in different companies. SpaceX owns xAI, Grok and Colossus. Tesla builds the body. Special board committees are reviewing closer ties, including a possible merger, but no deal exists, so who captures each rented hour is a real fiduciary test.

Castles, seashells and the 58%

Near the end, Baron asked where the demand for all this power and all these chips comes from. Musk said, admitting it sounds circular, that demand for digital superintelligence will come from digital superintelligence. He said it will bring universal extremely high income and medical care better than the best on earth today. You could wish for a castle. Our money system, he said, will seem like exchanging seashells.

In an October statement to the IMF and World Bank annual meetings, International Labour Organization Director-General Gilbert F. Houngbo noted that about 58% of the world's workforce is in informal jobs and warned that AI concentrated in a few firms and countries could widen gaps unless productivity gains are broadly shared.

Renting sharpens that risk. A displaced worker could finance a $20,000 robot, put it to work and collect its income. If the robot is rental only, that worker loses the paycheck and never owns what replaced it. So I'd want ordinary people to feel abundance early, through free self-driving miles and free robot hours, plus an AI dividend fund that helps workers become fleet orchestrators.

Who owns the fleet

Those only bridge the gap. My view is ownership is the real answer, because whoever owns the robots that build homes, run factories and build bases beyond Earth owns the engine of civilization. That can mean owning robots outright or holding a stake in the company that rents them. I'm describing who captures the surplus. This isn't investment advice. Musk said "probably rent them." The open question is who gets to own the fleet.