The Parity Calendar: Humanoid Robots Already Undercut Factory Workers in 22 Countries

BMW's Leipzig plant will put 40 Figure 03 humanoid robots on its logistics floor this summer at a contract rate of $25 per robot per hour, the same fully loaded cost as the plant's lowest-paid line workers. That single data point launched a thousand think pieces about "the future of work," but futures are vague and numbers are not, so we ran the numbers instead.

Using Bank of America's 2026 humanoid cost analysis, BLS manufacturing wage data, Destatis labor-cost comparisons, and the documented 40% annual hardware cost decline from Goldman Sachs (cited in Deloitte's 2026 Tech Trends), we built what nobody seems to have built yet: a country-by-country timeline of when humanoid robots become cheaper per hour than the humans they're designed to work alongside. We call it the Parity Calendar, and the finding that surprised us is that for most of the developed world, the crossover year isn't 2030 or 2035 but has already happened.

The Model

A humanoid robot's effective hourly cost isn't just the purchase price divided by hours worked, any more than a car's cost is the sticker price divided by miles driven. You need total cost of ownership: amortization, integration engineering, maintenance, energy, and the human supervisor who babysits a pod of ten robots because autonomy isn't perfect yet.

We used Bank of America's 2026 midpoint of $95,000 for a Western-built unit (their range is $90,000 to $100,000) and $35,000 for a Chinese-manufactured unit using Goldman Sachs' bill-of-materials estimate. Depreciation runs three years, which is generous since industrial equipment typically depreciates over five to seven, but humanoid hardware is evolving fast enough that three-year obsolescence seems realistic. Maintenance comes in at 12% of unit cost annually, drawn from industrial robotics benchmarks. Integration engineering, meaning the custom work required to fit a robot into a specific production line, adds 25% of unit cost in Year 1 and is amortized across the full life. One human supervisor per ten robots, at the local supervisory wage rate, rounds out the model.

We ran two operating scenarios because uptime is the swing variable. Our conservative case assumes two shifts, five days a week, 80% uptime, which yields 3,328 operating hours per year and roughly matches what you'd expect in a cautious early deployment. Our aggressive case assumes three shifts, seven days a week, 85% uptime, producing 7,446 hours per year, the kind of utilization continuous-operation factories like semiconductor fabs and Amazon fulfillment centers target.

The Math

At a $95,000 Western unit on conservative two-shift utilization, total cost of ownership works out to $17.18 per hour. On aggressive three-shift utilization, it drops to $7.71 per hour.

According to the U.S. Bureau of Labor Statistics, the average hourly earnings of all manufacturing employees reached $36.71 in June 2026. Even production and nonsupervisory workers, the lower-paid cohort, average $30.27 per hour. Team assemblers, the workers whose jobs most overlap with what humanoid robots can currently do, earn a median of $22.19, and the lowest-paid manufacturing helpers still come in at $18.92.

$17.18 is below every single one of those numbers.

Germany tells an even starker story. Destatis reports German manufacturing compensation at €49.50 per hour as of Q4 2025, roughly $54 at current exchange rates, which means a humanoid robot at $17.18 per hour is operating at 32 cents on the dollar. Denmark, the EU's most expensive manufacturing labor market at €55 per hour, shows an even wider gap. France at €47.10, Austria at €51.30, and Sweden at €47.20 face the same arithmetic: the robot undercuts them by factors of two to three.

The Parity Calendar

Here's what the crossover looks like, country by country, using the conservative two-shift model and applying a 40% annual unit-cost decline, the rate Goldman Sachs documented between 2023 and 2024 that Bank of America projects will continue as production scales.

Country / Cohort	Mfg. Wage ($/hr)	Robot TCO at Crossover	Parity Year
Denmark	~$60	$17.18	2026 ✓
Germany	~$54	$17.18	2026 ✓
France	~$51	$17.18	2026 ✓
Austria	~$50	$17.18	2026 ✓
Sweden	~$47	$17.18	2026 ✓
EU-20 Average	~$44	$17.18	2026 ✓
United States (all mfg.)	$36.71	$17.18	2026 ✓
Italy	~$36	$17.18	2026 ✓
United States (production)	$30.27	$17.18	2026 ✓
Japan	~$27	$17.18	2026 ✓
South Korea	~$25	$17.18	2026 ✓
U.S. (team assemblers)	$22.19	$17.18	2026 ✓
U.S. (helpers)	$18.92	$17.18	2026 ✓
Eastern Europe avg.	~$12	$11.14	2027
China	~$7.50	$7.56	2028
Mexico	~$5.00	$5.41	2029
Vietnam	~$3.00	$2.43	2029–2030
India	~$2.50	$2.43	2029–2030

Read that table carefully, because every OECD manufacturing country, all 22 of them, has already crossed parity under our conservative model. China, the last major bastion of labor-cost advantage in high-volume manufacturing, crosses in 2028. At three-shift utilization, it's already there: a $95,000 Western robot at $7.71 per hour matches a Chinese factory worker at roughly $7.50.

The Production Problem

Cost parity without supply is an academic exercise, and until this year supply was the wall that made the entire argument theoretical rather than practical. That wall is cracking open.

In 2025, the entire planet shipped approximately 13,000 humanoid robots. China accounted for about 90% of them, with AGIBOT alone shipping 5,168 units to claim 39% of the global market according to research firm Omdia, while Unitree shipped 5,500 and the entire United States managed roughly 450 combined with Figure AI, Agility Robotics, and Tesla Optimus each contributing about 150.

That was 2025, and in the first half of 2026 the production ramp has been staggering. AGIBOT's cumulative production hit 15,000 units in late June, doubling from 5,000 to 10,000 in just three months and then reaching 15,000 three months later, a pace that reflects manufacturing velocity rather than laboratory curiosity. During a six-day livestream from Longcheer Technology's tablet plant, the company showed G2 robots running inline quality inspection at production pace alongside human workers, accumulating over 100 cumulative hours of continuous factory operation.

Figure AI's trajectory is different but equally telling, with their BotQ factory going from producing one Figure 03 per day to one per hour in 120 days, a 24× throughput improvement that has already delivered over 350 units while targeting 12,000 per year and scaling to 100,000 over four years. Unitree is forecasting 20,000 units for 2026 alone, more than the entire global market produced the previous year.

And then there's the demand signal that makes all of this concrete rather than aspirational. State Grid Corporation of China disclosed in its 2026 Embodied Intelligence Development Plan a ¥5.8 billion ($800 million) procurement for 8,500 embodied intelligence devices, including 500 humanoids at ¥2.5 billion, 5,000 quadruped robot dogs, and 3,000 dual-arm inspection robots for power grid inspection and live-line operations. That is a single customer spending nearly a billion dollars on robots it intends to deploy rather than demo.

What the Deployment Data Actually Shows

Production scale means nothing if the robots can't do useful work, and the evidence here is thin but growing real in ways that matter for the economic argument.

Figure 02's 11-month pilot at BMW's Spartanburg plant produced verifiable results: the robot contributed to the assembly of more than 30,000 BMW X3 vehicles by handling over 90,000 sheet-metal components across roughly 1,250 operating hours. Inserting components into welding fixtures, the task required high precision, speed, and consistency, and the robots performed it alongside human workers under real production conditions rather than in a cordoned-off demonstration area.

Agility Robotics' Digit has moved over 100,000 totes at GXO Logistics in Flowery Branch, Georgia, marking the industry's first formal commercial humanoid deployment under a Robots-as-a-Service model, while Boston Dynamics' electric Atlas has its entire 2026 production allocation committed to Hyundai and Google DeepMind.

Its successor, the Figure 03, now being deployed, ran for 200 consecutive hours sorting packages with zero failures and zero human intervention. When a robot's battery ran low, it messaged a colleague, walked itself to a charging dock, and let the line keep running, which is autonomous industrial behavior and not a scripted demonstration of potential.

The Capability Gap Is Real

Here's what the Parity Calendar does not tell you, and this matters more than anything in the table above.

A humanoid robot at $17 per hour that can perform exactly three tasks (pick, place, inspect) is not a substitute for a $30 per hour human who can do 200 tasks, troubleshoot a jammed conveyor, shout to a colleague about a quality defect, and adapt when the production schedule changes at 2 PM on a Tuesday. Cost comparison is real, but capability comparison remains overwhelmingly in favor of the human worker, and conflating the two is the central error in most commentary about the "robot workforce."

Figure 02's 1,250 operating hours at BMW were impressive, yet they represent one task at one station in one factory where the 90,000 components handled were all the same category of work: inserting sheet-metal parts into fixtures. That robot didn't switch between tasks, didn't respond to unexpected production changes, and didn't do anything it hadn't been specifically trained to do. Industry analysts estimate humanoid robots can currently handle 15 to 20% of manufacturing tasks that are structured and repetitive, while the other 80% demands the kind of adaptive intelligence, cross-functional problem-solving, and physical improvisation that no robot on the market can deliver.

Our projected 40% annual cost decline is extrapolated from a single year of data (Goldman Sachs, 2023 to 2024). Cost curves in hardware manufacturing do tend to follow learning-rate declines, with solar panels dropping 99% over four decades and lithium-ion batteries dropping 97% over three, but the slope varies enormously in early years and humanoid robots have moving parts, literally, that batteries and panels do not. Actuators remain the primary bottleneck: harmonic drive suppliers are capacity-constrained, lead times for high-torque actuators exceed six months for new customers, China controls 26% of the global actuator market, and the United States controls 5%.

And the operating hours in our model are aspirational relative to what has actually been demonstrated in the field. BMW's Figure 02 achieved 1,250 hours over 11 months, roughly 114 hours per month, while our conservative scenario assumes 277 hours per month and our aggressive scenario assumes 620 per month. A gap between demonstrated performance and modeled performance running 2.4× to 5.4× will almost certainly close over time, but pretending it has already closed would be dishonest.

Why It Matters Anyway

The Parity Calendar matters not because it predicts mass displacement, which it emphatically does not, but because it reveals where the economic incentive has flipped. When a technology becomes cheaper than the alternative for even a narrow slice of tasks, adoption doesn't creep but rather compounds. Solar didn't replace coal by becoming cheaper across all use cases simultaneously; it became cheaper for specific grid segments first, then volume drove costs down further, costs drove adoption up further, and within a decade the crossover was total.

Humanoid robotics is entering that first segment now, and the tasks where parity holds, like structured pick-and-place, sequential loading, and quality inspection on a consistent line, represent a $340 billion slice of the global manufacturing labor market according to McKinsey's 2025 automation-scope analysis. That is the addressable wedge, not the whole workforce and not tomorrow, but the wedge is enormous and the cost curve moves in one direction only.

Production numbers tell you the industry already knows this, because you don't build a factory that spits out one robot per hour, commit $800 million in government procurement, and sign contracts at wage parity with human workers unless the math works for a product you intend to sell at scale rather than demonstrate at conferences.

What We Didn't Prove

Our model uses official wage data (BLS, Eurostat, Destatis) but estimated figures for Japan, South Korea, China, Mexico, Vietnam, and India drawn from OECD and ILO reporting rather than current-quarter national statistics. The Chinese wage estimate of roughly $7.50 per hour is the most consequential and the least precise, since manufacturing wages in China vary enormously by region from Shenzhen to Chengdu and our single number flattens that spread into a blunt average. The 40% annual cost decline is one data point extrapolated forward: if that rate slows to 25%, China's crossover moves from 2028 to 2030; if it accelerates to 50%, plausible given AGIBOT's production ramp, the crossover pulls in to 2027. Our operating-hours assumptions have not been validated at scale in any deployment, integration costs are estimated from traditional industrial-robotics benchmarks that may understate or overstate the cost for humanoid-specific adaptations, and we have not modeled software licensing, firmware updates, or the cost of retraining robots for new tasks, all of which could meaningfully affect TCO.

The Bottom Line

The Parity Calendar says humanoid robots already cost less per hour than factory workers across the developed world, a statement that is true on paper and narrow in practice because the robots can do a fraction of what humans can do, the production volumes can't satisfy real demand yet, and the deployment data covers thousands of hours rather than the millions needed to prove reliability at industrial scale. But the economics have crossed, and the question has shifted from "will humanoid robots ever be affordable?" to "how fast can they become capable?" Affordability depends on cost curves and production scale, which are slow, predictable, and already measurable; capability depends on AI, which is fast, unpredictable, and accelerating in ways that make five-year forecasts unreliable. If you're planning a factory, a workforce strategy, or a career in manufacturing, the conservative bet is to assume the capability curve catches up to the cost curve within five years, not because that outcome is certain but because the penalty for being wrong in the other direction, assuming it won't happen and being surprised when it does, is far more severe.

Here's what you can do with this analysis right now. If you manage a manufacturing line, run our TCO model against your actual wage bill for the 10 most repetitive stations, and you'll find the savings case is already there for at least two or three of them. If you're negotiating with a humanoid vendor, know that $25 per hour is the launch price and not the floor, so demand volume pricing tied to utilization commitments and push for contract structures where the vendor absorbs integration risk. If you work on a production floor, the most durable skill you can build right now is the ability to supervise, troubleshoot, and retrain robotic systems, because the 1-to-10 supervisor role in our model is the job category growing fastest in every country on the calendar, and starting that training before the robots arrive rather than after gives you leverage the market will reward.