When robots beat humans on cost (and when they never will)

The short version

• We built cost calculators for 21 jobs and ran them for the US, Germany, the UK, Japan, China, India, Brazil and Mexico, with sourced wages and sourced robot prices. Where we had to guess, the calculator says "assumption" next to the number. Try them.
• In our model, at default inputs, robots already cost less than people in 15 of the 21 jobs in the US. They win where the work is repetitive, runs long hours, and happens somewhere awkward or expensive for humans; several of those wins are thin. Wind turbine inspection by drone costs between a quarter and a third of rope access today.
• They lose, and keep losing to 2035 in our model, where the work is seasonal, short or cheap: driveway snow clearing, commercial mowing on a lease, roof inspections and small fireworks shows.
• Where wages are low, the crossover moves out by a decade or never comes. In India only work at height (drone inspections, facade cleaning) pays today in our model, and pool cleaning is the only other job that crosses before 2035 in the base case (2031 (2029–after 2035)).
• Every projection here is a Whirrwork modelled estimate, not a fact. We give ranges, not dates, and we will score ourselves against what actually happens.

The Robot vs Human Index

Each cell is the year a robot option first costs no more than the human baseline, in our base scenario, with the robots-fast to robots-slow range in brackets. "Now" means it is already cheaper at 2026 prices. Click a job to change any input.

Read it as a map of where the economics already work, not a forecast of what will happen. Plenty of cheaper robots never get bought, for reasons the last section covers.

Five findings

1. At height, drones won years ago. Rope-access wind turbine inspection runs $1,500–$3,000 a turbine against $300–$600 by drone (SkyVisor, a drone software company, so a vendor number; a French operator quotes €350–€700 within a farm campaign, C-Drone). Our model puts the drone at 0.28× the human cost in the US. The same holds in every country we modelled, including India, because the expensive part is the access, not the wage.

2. Cobots tending machines pay back in about a year where wages are decent. A UR5e arm lists at $39,083–$43,544 (Automation Distribution). Running two shifts, the default machine-tending model puts the robot at $10.78 an hour of work against $28.32 for an operator. Crossover: US now, China now, Mexico now, India not by 2035.

3. Warehouse picking is the genuine toss-up. Collaborative picking robots on subscription (Locus has advertised $2,000 a month, CIVL) are close to break-even against pickers walking with carts in the US today. Crossover in the US: 2029 (2028–after 2035). That range is wide because it depends on the learning rate, and nobody knows the learning rate.

4. Floor-cleaning robots win in high-wage countries only, and only just. At US and German wages a robot scrubber already costs slightly less per square foot than a cleaner with a walk-behind machine (US: now, at 0.93× the human cost; Germany: now; UK: 2029 (2028–2033)). A margin that thin flips on one bad assumption about uptime or edge work, so treat it as a toss-up. In Japan, China, India, Brazil and Mexico it does not cross before 2035 in any scenario, because a scrubber robot costs the same in yen or reais but the cleaner does not.

5. Some jobs never cross. Driveway snow clearing (not by 2035): a Yarbo with a snow module costs $7,299 (Yarbo) against $30–$70 a visit for a person with a plough (HomeGuide), and it snows a few dozen days a year. Drone light shows against fireworks (not by 2035): shows average about $52,500 in North America (Cyberdrone, citing SPH Engineering); a small-town fireworks display can be under $10,000 (City of Watertown). People choose drones for other reasons (no fire risk, reusable, quieter), and that is fine. It just isn't cost.

Why the lines move (and why slowly)

Robot hardware gets cheaper with volume. The clean example is the battery: pack prices fell from above $1,100 per kWh in 2010 to $108 in 2025 (BloombergNEF; BloombergNEF), about 20% cheaper for every doubling of cumulative production (Ziegler & Trancik, 2021). That pattern is Wright's law, first written down for aircraft in 1936 (Wright), and it hindcasts technology costs better than most alternatives, with errors that grow with the horizon (Nagy et al., 2013).

Finished robots fall much more slowly than their parts. A UR5 cost $35,000 in 2015 (IEEE Spectrum); its successor lists at about $39,000–$43,500 today. Vendors spend the savings on better robots, and integration, support and margins do not ride the learning curve. So our scenarios use 10%, 15% and 20% per doubling, applied to hardware only, with integration and service costs moving with wages. Interact Analysis expected 10–15% for mobile robots (Robotics 24/7).

Meanwhile wages for the jobs robots compete with grow, but not fast. In the US, real hourly earnings in elementary occupations grew about 1.2% a year from 2015 to 2024 (ILOSTAT, deflated by World Bank CPI; our calculation). Average earnings in China (all industries, 2012–2022, the series ILOSTAT has) grew far faster, about 6% a year in real terms, which is part of why China crosses over on palletizing (2027 (2027–2028)) while India and Brazil do not. The details are on the trends page.

The hidden costs that decide most cases

The robot is rarely the biggest number. In the calculators, the inputs that swing the five-year NPV most (the tornado chart on each page) are usually not the robot's price but:

• Edge share: the part of the work the robot can't do (corners, odd items, exceptions). A person still does it, at full cost.
• Supervision: minutes of staff time per robot-hour to start, empty, refill and rescue. Six minutes an hour sounds like nothing; across a fleet it is a job.
• Uptime and failures: a robot that is charging, stuck or broken is doing nothing, and the work it drops still gets done by someone.
• Integration: mapping, Wi-Fi, docking, IT. Our defaults are mid-range guesses, labelled as such. Get a quote.
• Utilisation: fixed costs only pay back on hours. Each calculator shows the break-even utilisation, the share of available hours a robot must work just to cover itself. Above 100% means no schedule saves it.

There's a longer list, with receipts, in the hidden costs of robots.

When the cheaper robot still fails

Three verified cases, each a reminder that cost is not the only test:

• Henn-na Hotel, Japan (2019). The robot hotel retired more than half of its 243 robots because they created more work for staff than they saved (The Verge).
• Walmart shelf scanning (2020). Walmart ended its contract with Bossa Nova after finding simpler ways to get the same inventory information (Reuters). Our inventory-scanning model says the robot is cheaper; Walmart found something cheaper still.
• iRobot (2025). The company that made the robot vacuum mainstream went through a prepackaged Chapter 11 in December 2025 (iRobot). A five-year payback assumes your vendor is still there in year five.

People, ethics and redeployment

Cheaper is not the same as better, and the research on robots and jobs is mixed. One more robot per 1,000 US workers reduced the local employment-to-population ratio by about 0.2 percentage points and wages by about 0.42% (Acemoglu & Restrepo, 2020). In Germany, losses in manufacturing were offset by new service jobs, with the adjustment falling mostly on young people entering the labour market (Dauth et al., 2021). Spanish firms that adopted robots grew employment about 10%, while their competitors lost jobs (Koch, Manuylov & Smolka). The evidence piece has the full table.

Our view, which is a view and not a finding: the cost case for a robot should include the cost of redeploying or retraining the people it affects, and the plan should exist before the robot does. Each calculator lists the non-cost factors (safety, quality, flexibility, customers, jobs and regulation) under the numbers, for exactly this reason.

How to use this

1. Open the calculator for your job, pick your country and region, and replace every input labelled "assumption" with your own figure.
2. Look at the tornado chart. If the answer flips on an input you can't pin down, run a pilot before you sign anything.
3. Check the projections for whether waiting a few years changes the answer, and the methodology for what the model leaves out.
4. Agents can ask the same question through the API and MCP tool. Related Whirrwork research lives in the research hub.

Cite this

Whirrwork. (2026). When robots beat humans on cost (and when they never will) (Version 1.0.0) [Data set and interactive model]. https://whirrwork.com/guides/when-robots-beat-humans-on-cost/

Whirrwork. “When robots beat humans on cost (and when they never will).” Version 1.0.0. 2026. Accessed October 25, 2026. https://whirrwork.com/guides/when-robots-beat-humans-on-cost/.

Whirrwork. “When robots beat humans on cost (and when they never will).” Whirrwork, 2026, whirrwork.com/guides/when-robots-beat-humans-on-cost/. Accessed 25 Oct 2026.

@misc{whirrwork2026when,
  author = {{Whirrwork}},
  title = {When robots beat humans on cost (and when they never will)},
  year = {2026},
  version = {1.0.0},
  howpublished = {\url{https://whirrwork.com/guides/when-robots-beat-humans-on-cost/}},
  note = {Accessed 2026-10-25. Modelled estimates, not facts.}
}

Charts on the trends page and each calculator have an embed code with the source line built in. Data and model are CC BY 4.0: download them.