For three years, every conversation about humanoid robots ran into the same wall: nobody would tell you the price. You'd watch a slick video of a robot sorting parts, ask what it costs, and get "contact sales" or a vague six-figure number that made the whole thing feel theoretical. That wall just came down in the most consequential way possible. Figure's commercial deal with BMW is structured at roughly $25 per robot-operating-hour — and suddenly robot labor has something it never had before: an hourly wage you can put next to a human's.
That single number changes the conversation from science fiction to spreadsheet. Automotive labor in the US runs roughly $50–70 an hour fully loaded, so a machine that bills at $25 isn't a futuristic curiosity — it's a line item that undercuts the alternative by half. And BMW didn't just run the pilot and write a nice press release; after 11 months at its Spartanburg plant, it doubled down and signed a commercial contract. Let me walk you through what actually happened, the pricing model that makes it work, and what a $25-an-hour robot means for your business, your investments, and your job.
What BMW actually got
Before the money math, the performance — because a cheap robot that doesn't work is just cheap. Figure's robots spent roughly 11 months at BMW's Spartanburg plant, and the numbers they put up are unusually concrete for this industry:
| Metric at BMW Spartanburg | Result |
|---|---|
| Placement accuracy per shift | over 99% |
| Cycle time target | met the 84-second target |
| Parts loaded | 90,000+ |
| Operational hours | ~1,250 |
| Vehicles the pilot contributed to | 30,000 |
Read that top row twice, because it's the one that matters most in a factory. Over 99% placement accuracy while hitting an 84-second cycle time means the robot kept pace with a production line built for humans, reliably, shift after shift. That's not a demo — that's industrial-grade repeatability. And the payoff was the follow-on order: BMW moved from test to a commercial contract covering an initial fleet of 40 Figure 03 units across body-shop and assembly-line workstations. Customers who run pilots and then expand are the only reliable signal in this whole sector; everything else is a highlight reel, which is exactly the skepticism I applied in Figure vs Tesla Optimus: real versus hype.
The pricing model that changes everything
Here's the part I think is genuinely more important than the robot itself. The BMW deal isn't a purchase — it's robot-as-a-service (RaaS). BMW doesn't buy a fleet, own it, depreciate it, and pray it doesn't break. It pays roughly $25 for every hour a robot actually works.
Think about what that removes. When I ran the numbers on Neura's €98,000 humanoid, the whole analysis hinged on a big upfront purchase — capital most businesses don't have sitting around, spent on a technology that might be obsolete in three years. RaaS deletes that problem entirely:
| Buy the robot | Rent by the hour (RaaS) | |
| Upfront cost | ~€98,000+ per unit | $0 |
| Who eats obsolescence risk | You | The vendor |
| Who handles maintenance/updates | You | The vendor |
| If it doesn't work out | You own an expensive paperweight | You stop paying |
| Compares directly to | A capital budget | An hourly wage |
That last row is the strategic masterstroke. By pricing per hour, Figure stopped selling a machine and started selling labor — and labor is a budget every company on Earth already has, understands, and approves without a board meeting. A plant manager doesn't need to win a capex fight; they need to show that $25/hour beats $60/hour. It's the same shift that took software from boxed purchases to subscriptions, applied to physical work, and it's how adoption goes from a trickle to a flood.
The money math: $25 versus $50–70
Let's do the arithmetic honestly, because this is the calculation now landing on operations desks everywhere. Take one workstation running two shifts, roughly 4,000 working hours a year:
| Line | Human worker(s) | Figure robot (RaaS) |
| Effective hourly cost | $50–70 (loaded: wages, tax, benefits) | ~$25 |
| Annual cost at ~4,000 hrs | $200,000–280,000 | ~$100,000 |
| Benefits, pension, overtime | Yes | None |
| Availability | Shifts, breaks, sick days, turnover | Continuous while it works |
| Upfront capital | — | $0 |
| Annual difference | — | ~$100,000–180,000 saved per station |
At those rates, a single workstation converted to robot labor saves somewhere between $100,000 and $180,000 a year — and you pay nothing upfront to find out. Multiply by 40 robots and you're looking at millions annually at one plant. That's why the pilot converted into a contract, and it's the same "the machine is cheaper than the payroll" logic I've documented in warehouse robot ROI and farming robots that pay back in a season — except now it's priced per hour, which makes it undeniable rather than theoretical.
One honest caveat before you extrapolate: $25/hour is a rate for a specific, well-defined industrial task at a flagship customer, and early strategic contracts are often priced to win the logo. The number may not hold across every task or customer. But the principle — humanoid labor priced by the hour, below human rates — is now established, and prices in this field are falling, not rising.
The supply side is finally real
A price only matters if there are robots to rent, and 2026 is the year the production lines actually started moving. Here's where the field stands right now:
| Maker | Status (August 2026) | Price signal |
| Figure | Figure 03 passed 1,000 units; BotQ producing ~1 robot/hour; 40 units commercial at BMW | ~$25/robot-hour (RaaS) |
| Unitree | Shipped 5,500+ in 2025 — more than everyone else combined; targeting 10,000–20,000 in 2026 | From ~$16,000 to buy |
| Tesla Optimus | Fremont line converted from Model S/X, targeting 1M units/year capacity | V3 cost >$60,000; targeting $20–30K |
| AgiBot | ~15,000 cumulative units | Chinese cost structure |
| Neura | €1B+ order book, scaling serial production | ~€98,000 to buy |
The pattern here is the one I flagged in why China builds humanoids cheaper and the Unitree price war: the Chinese makers dominate on volume and price, the Americans dominate on high-value deployed contracts, and everyone's costs are heading down. Tesla's target is the wildcard — if a Fremont line built for a million units a year delivers Optimus at $20–30K, the hourly cost of robot labor collapses again, and the $25 benchmark starts to look expensive. For the full board, I keep a running humanoid price comparison.
Why "per hour" is the number that breaks the dam
I want to dwell on why this pricing shift matters more than any spec sheet, because it's the thing most coverage misses.
Every previous humanoid conversation forced a business into an unfamiliar decision: should we make a six-figure capital investment in an unproven robot? That question has a hundred ways to end in "no" — budget cycles, board approval, depreciation schedules, fear of buying the Betamax of robots. RaaS reframes it into a question every operations manager answers weekly: can we get this task done cheaper? When the answer is "yes, at less than half the rate, with no money down," the decision stops being strategic and becomes routine.
That's how technologies actually cross over — not when they get impressive, but when they get easy to buy. And it creates a genuine benchmark for the first time: robot labor now has a market rate. Every task in your business can be measured against it. That's a profound change in how work gets priced, and it's arriving quietly, in procurement documents, rather than in the dramatic headlines about robot uprisings.
What this means for jobs
Let me be straight rather than either alarmist or dismissive, because you deserve the honest version. A machine that does a defined physical task at half the loaded cost of a human, across multiple shifts, with no benefits, is going to take that task. Not every job — a task. The BMW deployment is instructive: robots went to specific body-shop and assembly-line workstations, the repetitive part-loading work with clear cycle times, not to the whole plant.
So the displacement lands first exactly where the money math is cleanest: repetitive, physically defined, high-wage-region industrial work. That's the pattern I traced in which jobs AI and robots hit first and the pressure on middle-class work. But the same deployment creates work that didn't exist: someone deploys these fleets, integrates them into lines, monitors them, maintains them, and decides which stations to convert. That's the paid side of the great AI job split — the robots need humans who can direct them, and those humans are getting scarce and expensive. If you work in or near manufacturing, becoming the person who runs the fleet is the single most durable move available to you right now.
What this means for you
Depending on where you sit, here's the practical read.
If you run a business with physical labor, this is now a real procurement option, not a future one — and the RaaS model means you can test it without a capital commitment. Pick your single most repetitive, well-defined, high-cost station, get a quote on an hourly basis, and run the comparison honestly against your loaded labor cost. The worst case is you learn your number; the best case is six figures a year per station. Start with one station, not a plan to automate the plant.
If you invest, note that RaaS changes the whole shape of these companies. Selling robots is lumpy hardware revenue; renting them by the hour is recurring, high-margin, software-like revenue that compounds with every deployed unit — which is exactly why the market rewards it. Watch deployed-fleet counts and hourly rates, not demo videos. And remember the whole stack rides on the same silicon driving Nvidia's numbers; the broadest way to play it is the basket approach in how to invest in the robot boom.
If you work in manufacturing or logistics, take this as an early, actionable signal rather than a threat. The $25/hour benchmark tells you precisely which tasks are on the clock: the repetitive, defined ones. Move toward the work that surrounds the robots — deployment, integration, maintenance, quality, supervision. The fleet at BMW didn't eliminate the need for skilled people; it changed what those people do, and it made the ones who understand robots considerably more valuable.
The honest take
The thing that strikes me about this milestone is how boring it looks on paper. There was no breakthrough announcement, no humanoid doing backflips. There was a purchase order, an hourly rate, and a customer who ran the numbers for 11 months and decided to buy more. That's what actual adoption looks like — not a moment, but a line item that pencils out. Robot labor crossed from spectacle into procurement the day it got a price per hour.
The pattern worth keeping is the one that repeats across every technology this site covers: the capability arrives first and impresses everyone, but nothing really changes until somebody figures out the business model that makes it easy to buy. Cloud computing did it by turning servers into a monthly bill. Software did it with subscriptions. Humanoid robots just did it by turning a €98,000 machine into a $25 hourly rate — and in doing so, they made themselves directly comparable to the thing they're replacing. That comparison is now running on spreadsheets in every high-wage factory on Earth, and the arithmetic isn't close.
So here's the question worth carrying into your own operation: if a machine can now be hired by the hour for half of what the work costs you today, with nothing down — what's the honest reason your competitor won't do it first?
Sources: IIoT World — physical AI deployment ROI; IndustrialSage — 11 months at Spartanburg.
