Humanoid Robots Will Create Jobs Before They Destroy Them. Here Is the Math.

Every time a new humanoid robot video goes viral, the same headlines appear. “Robots are coming for your job.” “The end of human labor.” “Mass unemployment is inevitable.” The framing is always the same: robots destroy jobs, humans lose.

But there is a number that almost never appears in those stories. It is the number of people required to design, build, program, deploy, maintain, regulate, and supervise each robot that enters the workforce. That number is not zero. It is not small. And when you multiply it across the 250,000 humanoid robots Goldman Sachs projects by 2035, you get a finding that makes both automation optimists and pessimists uncomfortable.

The humanoid robotics industry will need to create hundreds of thousands of jobs before it can displace any at scale. And right now, the workers to fill those jobs do not exist.

The question nobody is doing the math on

Ask an economist how many jobs humanoid robots will displace and you will get a number. McKinsey estimates that up to 30% of hours worked globally could be automated by 2030. The World Economic Forum says 85 million jobs could be displaced by automation by 2025 (a figure that has since been revised, since 2025 has come and gone without anything close to that happening). Goldman’s own model assumes humanoid robots will fill 250,000 positions by 2035.

But ask how many jobs it takes to put a humanoid robot into a factory and keep it running, and you will get silence. Not because the answer is unknowable. Because almost nobody has tried to calculate it.

The displacement forecasts get all the attention because fear drives clicks. The creation side of the equation is harder to model and less dramatic to report. But it is where the real story is.

Counting the jobs behind each robot

To model the job creation per robot deployed, we need to trace the full lifecycle of a humanoid robot from concept to retirement. Each phase requires distinct human skills, and many of those skills are in critically short supply.

We estimated the labor requirements by analyzing publicly available workforce data from established industrial robotics companies, adapting for the additional complexity of humanoid platforms. The International Federation of Robotics, Deloitte’s robotics skills surveys, and the Association for Advancing Automation all provide baseline data. We cross-referenced those with hiring patterns at humanoid-specific companies and academic labor economics research.

Here is the lifecycle and the labor it requires.

Phase 1: Research and development

Before a single robot ships, hundreds of engineers spend years developing the platform. A typical humanoid robot company employs 50-500 engineers across mechanical design, electrical engineering, embedded systems, control theory, computer vision, and AI/ML. These are highly specialized roles that draw from a global talent pool.

But R&D jobs scale differently from the other categories. You do not need 10x more R&D engineers to produce 10x more robots. R&D costs amortize across units. For the purpose of job creation math, we estimate that the global humanoid robot R&D workforce will need to grow from roughly 5,000 today to 25,000-35,000 by 2035 across all manufacturers. That is significant, but it is not the biggest category.

Phase 2: Manufacturing and assembly

This is where the numbers start getting large. Humanoid robots are among the most complex consumer-scale machines ever mass-produced. A single humanoid has 20-40 actuators, dozens of sensors, custom wiring harnesses, precision-machined joints, and integrated computing hardware. Current assembly is heavily manual.

The International Federation of Robotics estimates that traditional industrial robot manufacturing requires approximately 3-5 direct manufacturing workers per 100 robots produced annually. Humanoid robots, because of their complexity, require roughly 8-12 workers per 100 units at current maturity levels.

To reach Goldman’s 250,000 cumulative units by 2035, the industry needs to produce roughly 50,000-80,000 units per year by 2033-2035 (accounting for the ramp curve). At 10 workers per 100 units, that is 5,000-8,000 direct manufacturing workers. Add supply chain, logistics, and quality assurance, and the total manufacturing workforce needed is 15,000-25,000.

Phase 3: AI training and data operations

This is the category that most people underestimate. Every humanoid robot that performs useful work in the real world depends on AI models trained on enormous datasets. Those datasets do not create themselves.

Humanoid robot AI training today relies on three main approaches: simulation (which requires engineers to build and maintain virtual environments), teleoperation (where humans physically demonstrate tasks that the robot learns from), and real-world data labeling (where humans annotate sensor data to train perception models).

Teleoperation alone is enormously labor-intensive. To teach a robot to perform a single manipulation task reliably, current approaches require 50-200 hours of human demonstration data. A robot that can perform 20 different tasks in a warehouse environment might need 1,000-4,000 hours of teleoperation data. At scale, the industry will need thousands of teleoperators, a job category that did not exist five years ago.

Data labeling for robotics is similarly demanding. Unlike labeling images for a web search algorithm, labeling 3D spatial data from LiDAR, depth cameras, and force sensors requires specialized training. The ILO and Stanford HAI both note that robotics data work is emerging as a distinct labor category that is harder to offshore than traditional data labeling.

We estimate AI training and data operations will require 30,000-50,000 workers globally by 2035.

Phase 4: Deployment and integration

A humanoid robot does not arrive at a factory and start working. Each deployment requires site assessment, environment mapping, safety zone configuration, task programming, integration with existing systems (conveyor belts, WMS software, safety interlocks), and validation testing. For early deployments, this process takes weeks to months per site.

Based on current deployment patterns from companies like AgiBot and Figure, each site deployment requires 2-4 integration engineers for 2-8 weeks. As the technology matures and deployments become more standardized, this ratio will improve. But even at scale, the industry will need an estimated 10,000-20,000 deployment and integration specialists by 2035.

Phase 5: Fleet operations

Once robots are deployed, someone has to manage them. Fleet operations involves remote monitoring, task scheduling, performance optimization, over-the-air software updates, and exception handling (what happens when a robot encounters a situation it cannot handle).

The emerging model looks similar to autonomous vehicle fleet management, where each human operator oversees 5-20 vehicles remotely. For humanoid robots in industrial settings, early ratios are closer to 1 human per 3-5 robots. As autonomy improves, this could reach 1 human per 15-25 robots by 2035.

At 250,000 deployed robots with an average ratio of 1 operator per 15 robots, that is roughly 17,000 fleet operations jobs. With support staff, supervisors, and escalation teams, the total is closer to 20,000-30,000.

Phase 6: Field maintenance and repair

This is the biggest single category, and the one with the most severe talent shortage. Humanoid robots are complex machines operating in dynamic environments. They break. They wear out. They need calibration.

The Bureau of Labor Statistics reports that existing industrial robots require an average of 40-60 hours of maintenance per year per unit. Humanoid robots, with their more complex kinematic chains, higher actuator counts, and exposure to more varied environments, will require more. Early data from pilot deployments suggests 80-150 hours per unit per year during the current maturity phase, declining to 40-80 hours as reliability improves.

At 250,000 deployed robots requiring an average of 60 hours of maintenance per year, that is 15 million maintenance hours annually. At 1,800 productive hours per technician per year, that is approximately 8,300 full-time maintenance technicians. But field service is inefficient. Travel time, parts ordering, diagnostics, and administrative tasks typically double the headcount requirement. Realistic estimate: 15,000-25,000 field maintenance workers.

And here is the problem. These workers need a skill set that combines mechanical repair, electrical troubleshooting, software diagnostics, and basic AI/ML understanding. That training pipeline does not exist in most countries. It takes 2-4 years to develop a competent humanoid robot field technician, and almost no vocational programs teach this combination of skills today.

Phase 7: Safety, compliance, and regulation

Putting a bipedal robot weighing 35-80 kg into a workspace alongside humans requires safety certification, regulatory compliance, insurance underwriting, and ongoing auditing. Each jurisdiction has its own requirements. The EU Machinery Regulation, updated in 2023, now includes specific provisions for autonomous mobile robots in shared workspaces. The U.S. is developing ANSI/RIA standards for humanoid systems. China has its own GB/T standards.

This creates demand for safety engineers, compliance officers, auditors, insurance analysts, and regulatory specialists who understand both robotics and workplace safety law. We estimate 5,000-10,000 jobs in this category by 2035.

Adding it all up

Our estimate: the humanoid robot industry will directly create 120,000-195,000 jobs by 2035 to support Goldman’s projected 250,000 deployed units. The midpoint is roughly 155,000 direct jobs.

That does not include indirect employment effects: the machine tool companies that build actuator manufacturing equipment, the university programs that train robotics engineers, the consulting firms that advise on automation strategy, the insurance companies that develop new product lines, or the content creators and journalists who cover the industry. Standard economic multiplier effects for advanced manufacturing suggest that each direct robotics job creates 1.5-2.5 indirect jobs. Applied conservatively, that adds another 180,000-390,000 indirect jobs.

Total direct and indirect job creation: 300,000-585,000 by 2035.

Now compare that to the displacement side

Goldman’s base case says humanoid robots will fill 250,000 positions by 2035. “Fill” is doing a lot of work in that sentence. It does not mean 250,000 people lose their jobs overnight. It means humanoid robots will be performing work equivalent to 250,000 full-time positions.

But many of those positions are in sectors that already face chronic labor shortages. Manufacturing in the U.S. alone has 600,000-800,000 unfilled positions. Japan’s manufacturing sector faces a projected shortfall of 4 million workers by 2030 due to demographic decline. Germany’s Mittelstand manufacturers report unfilled vacancy rates of 5-8%.

The International Labour Organization notes that a significant fraction of early humanoid robot deployments will fill positions that employers cannot staff with human workers at any wage, particularly in physically demanding manufacturing roles, hazardous environments, and regions with acute demographic decline.

The comparison reveals something that neither side of the debate talks about. In the near term (2026-2030), job creation from the humanoid robotics industry will likely exceed job displacement. The industry is in its building phase. It needs workers to design, manufacture, train, deploy, and maintain robots faster than those robots can displace workers elsewhere.

The crossover point, where displacement begins to exceed creation, is somewhere around 2031-2033 in most models. After that, displacement accelerates while creation plateaus (you do not need proportionally more maintenance workers per robot as the fleet grows, because tooling, training, and processes improve).

But even after the crossover, a large fraction of the “displaced” positions are in sectors with chronic labor shortages. The robot is not taking someone’s job. It is filling a position that has been vacant for months or years.

The skills gap is the real crisis

If the industry needs 120,000-195,000 workers by 2035, where do they come from? This is where the math gets genuinely alarming.

The Deloitte Global Robotics Skills Gap Survey, published in 2025, found that 73% of robotics companies report difficulty hiring qualified technicians. The Association for Advancing Automation reports that the U.S. alone will need 40,000 new robotics workers by 2028, but university and vocational programs are producing roughly 8,000-10,000 qualified graduates per year across all robotics disciplines.

The problem is especially acute for field maintenance. There are excellent mechanical engineering programs. There are excellent computer science programs. There are almost no programs that produce people who can diagnose a faulty actuator, reflash a motor controller, recalibrate a force-torque sensor, and update a behavior model, all in the same service call.

Boston Consulting Group’s 2025 report on the robotics workforce estimates that retraining an experienced industrial maintenance technician for humanoid robot service takes 6-12 months of intensive training. Training someone from scratch takes 2-4 years. At those timelines, even aggressive workforce development starting today cannot fully close the gap by 2035.

What skills are actually needed?

The skills gap is not one gap. It is several overlapping gaps across different roles. Here is what the industry actually needs.

Mechatronics technicians. People who understand the intersection of mechanical systems, electronics, and software. This is the core skill for manufacturing and field maintenance. Traditional trade schools teach these disciplines separately. The integration is what matters.

AI/ML operations specialists. Not researchers. Operations people who can manage training pipelines, monitor model performance in production, debug data quality issues, and handle the unsexy operational work that keeps AI systems running. Universities produce plenty of ML researchers. They produce almost no ML operations specialists.

Teleoperation specialists. An entirely new job category. These are people who control robots remotely to generate training data, handle exceptions, and provide real-time guidance. The skill profile is unusual: fine motor control, spatial reasoning, patience, and enough technical understanding to know what the robot is trying to learn from each demonstration.

Integration engineers. People who can make a humanoid robot work within an existing factory or warehouse environment. This requires understanding both the robot system and the legacy infrastructure it needs to interact with (PLCs, SCADA systems, warehouse management software, conveyor systems). These people are rare.

Safety engineers with robotics expertise. Workplace safety is a mature field. Robotics is a maturing field. The intersection is tiny. Most safety engineers have no experience with autonomous mobile robots. Most robotics engineers have limited safety certification experience.

What previous automation waves tell us

This is not the first time a new technology has triggered panic about mass unemployment. The historical pattern is instructive, though not perfectly reassuring.

The pattern from previous waves is consistent. In the near term (first 5-15 years), new technology creates more jobs than it destroys, because building and deploying the technology requires enormous human effort. In the medium term (15-30 years), displacement catches up and eventually exceeds creation in the directly affected sectors. In the long term, the economy restructures around the new capability, and total employment recovers, though not necessarily in the same sectors, regions, or skill levels.

The IFR’s data on industrial robots is particularly relevant. Countries like South Korea, Japan, and Germany that deployed the most industrial robots per capita in the 1990s and 2000s did not see higher manufacturing unemployment than countries that deployed fewer robots. In many cases, they saw lower unemployment, because the productivity gains from automation made their manufacturing sectors more competitive globally, which sustained employment.

The uncomfortable middle ground

Both extremes of the debate are wrong.

The techno-optimists who say “robots will create more jobs than they destroy” are right in the near term but wrong in the long term, at least within directly affected sectors. If humanoid robots become as capable as their makers promise, they will eventually displace more workers in manufacturing and logistics than the robotics industry itself employs.

The doomers who say “mass unemployment is coming” are wrong in the near term and likely wrong in the long term, because they consistently underestimate the economy’s ability to create entirely new categories of work. But they are right about the transition period being painful for specific workers in specific industries.

The real finding is this: the near-term crisis is not robots taking jobs. It is the robotics industry being unable to find workers. The medium-term crisis is not mass unemployment. It is a brutal skills mismatch where hundreds of thousands of people need retraining and the systems to retrain them do not exist at scale.

What needs to happen

If the analysis above is correct, the policy response should be obvious. But it is not happening fast enough.

Vocational training programs need to add mechatronics and robotics maintenance tracks now. Not in five years. Now. The training pipeline is 2-4 years long, and the industry needs graduates by 2028-2029 to support the production ramp. Germany’s dual education system is the closest model, where apprentices split time between classroom and factory floor. Most countries have nothing comparable for robotics.

Companies need to invest in retraining before they deploy. Every major humanoid robot deployment should include a workforce development component. If you are replacing 50 warehouse workers with 20 robots and 10 human operators, the math says you should retrain 10 of those 50 workers to be the robot operators and maintenance techs. Some companies are doing this. Most are not.

Universities need to stop treating robotics as a subdiscipline of CS or ME. Robotics is its own field. It requires integrated curricula that combine mechanical engineering, electrical engineering, computer science, AI/ML, and human factors. The best robotics programs already do this, but there are fewer than 50 graduate programs globally that qualify, and almost no undergraduate or vocational programs at that level of integration.

Governments need to fund transition programs, not try to slow deployment. The worst policy response is to restrict humanoid robot deployment to “protect jobs.” This just ensures that other countries capture the robotics industry jobs while yours loses both the robotics jobs and the manufacturing jobs (because your factories become uncompetitive against robotics-enhanced competitors).

The math, summarized

Goldman Sachs projects 250,000 humanoid robots deployed by 2035. Here is what getting to that number requires in human labor.

For every robot deployed, the industry creates roughly 0.6 direct jobs. Including indirect economic effects, that number rises to 1.2-1.8 jobs per robot. Against Goldman’s 250,000 displaced positions (many of which are currently unfilled), the net employment effect through 2035 is roughly neutral, possibly slightly positive if you account for unfilled manufacturing vacancies.

The story changes after 2035. If humanoid robots reach the kind of capability and cost levels that Goldman’s bull case envisions, displacement will accelerate faster than job creation. But that is a different problem for a different decade.

The problem for this decade is straightforward. The robotics industry is hiring, and there is almost nobody to hire. The people who should be training to become robotics technicians, fleet operators, and AI data specialists in 2035 need to start that training in 2026 or 2027. Most of them do not know these jobs exist.

The real crisis is not a jobs gap. It is a skills gap. And every month we spend arguing about whether robots will take our jobs is a month we are not spending building the training programs that would let people take the robot jobs instead.