Factories now run on a record installed base, and the labor data around it is messier than either the optimists or the doom forecasters claim. The International Federation of Robotics (IFR) counted 603,307 new industrial robot installations in 2025, up 11%, and an operational stock of about 5.08 million, up 9%. Meanwhile, the best-known causal estimates still come from older data. This guide pairs the 2026 hardware numbers with the econometrics, then gives engineers a model to quantify the effect on their own site.
Quick Takeaways
- Installations are accelerating. The IFR forecasts 655,000 installations in 2026 and 806,000 by 2029.
- Local displacement is real but concentrated. In the classic US estimate, one more robot per thousand workers lowers the employment-to-population ratio by 0.2 percentage points and wages by 0.42%.
- Firm-level and country-level results differ. Adopting firms often grow, while competitors and local labor markets can shrink.
- Humanoids are still a rounding error. The IFR counts only about 7,000 humanoids sold worldwide last year for industrial and professional use.
The 2026 Hardware Baseline
Start with what is physically deployed. The numbers below come from the IFR World Robotics 2026 report.
| Metric (2025 data) | Value | Note |
|---|---|---|
| Installations | ~603,000 | Up 11% year on year |
| Operational stock | ~5.08M | Up 9% |
| China share | 59% | Installations grew 20% year on year |
| US installations | ~38,500 | Up 12%; the US passed Japan as the second-largest market |
| Cobot installations | 66,000 | Up 7%, about 11% of all installations, and the first year cobots grew slower than the overall market |
| Humanoids sold | ~7,000 | Industrial and professional use only |
Two engineering details matter here. First, the IFR counts robots, not vacancies. Installations are not jobs. Second, the mix is shifting. Fixed 6-axis arms still dominate, while AMRs (autonomous mobile robots) and early humanoids target warehouse and intralogistics work that caged arms never reached.
The Economics: How Researchers Measure Robot Impact
The Shift-Share Exposure Model
The standard identification strategy builds a local exposure measure from national robot adoption and each region’s industry mix:
Exposure_c = Σ_i [ (L_ci / L_c) × (ΔR_i / L_i) ]
Where:
c = commuting zone
i = industry
L_ci = employment of industry i in zone c
ΔR_i = change in national robot stock in industry i
L_i = national employment of industry i
Researchers then regress the change in the employment-to-population ratio on that exposure. Acemoglu and Restrepo applied this to 19 industries and 722 US commuting zones. Their reported effects vary by specification: 0.18–0.34 percentage points of employment and 0.25–0.5% of wages per additional robot per thousand workers. Their headline figure also implies that each robot replaces about 6.6 jobs locally.
Two Effects in Tension
Every automation study separates two forces:
Net labor demand effect = Displacement effect (tasks taken from workers)
+ Productivity effect (cheaper output, more demand)
The authors argue robots produce a stronger displacement effect than the productivity effect. That is a claim about 1990–2007 manufacturing, not a law of nature.
What the Studies Disagree On
| Study / Context | Finding | Trade-off Signal |
|---|---|---|
| US commuting zones (Acemoglu & Restrepo) | Negative employment and wage effects | Strong displacement in routine manual work |
| France/Spain firm-level | Adopters expand at competitors’ expense | Gains at firm level, losses at industry level |
| Chinese firms (2026) | Positive link between robot adoption and firm employment, with falling labor share | Scale effect outweighs substitution |
| Chinese industry elasticities (2026) | Robots and labor are complements overall, with crowding-out in substitution industries | Task mix decides the outcome |
| US gender/race split (Lerch) | Employment fell 3.7 points for men vs 1.6 for women, and 4.5 for non-White vs 1.8 for White workers (1993–2014) | Unequal incidence |
The reconciliation is task-based. Where a robot substitutes for a task a worker performs, jobs fall. Where it complements labor and the firm grows, jobs hold or rise. Firm-level studies observe the winners. Regional studies observe the net.
Safety is a measurable upside. Italian data show robot adoption reduced workplace accidents, particularly for men in physically demanding jobs.
Technical Feasibility vs Economic Feasibility
Engineers know the gap between “can do” and “pays back.” Anthropic’s robot exposure index puts numbers on it: robots can perform 74% of US physical job tasks, covering 34% of working hours, yet are cost-competitive for only 0.3%. The same source reports that 13% of firms use robotics now and 22% expect to within three years.
That gap is the central fact of 2026. Capability is not the bottleneck. Payload, cycle time, integration cost, and uptime are.
Hardware Comparison: Where Each Robot Class Substitutes Labor
| Platform | Typical Task | Setup Cost | Flexibility | Latency/Cycle Profile |
|---|---|---|---|---|
| 6-axis industrial arm | Welding, palletizing | High (cell + safety cage) | Low | Sub-second, highly repeatable (~0.05mm class) |
| Cobot | Machine tending, light assembly | Medium | Medium | Slower, speed-limited near humans |
| AMR | Tote/pallet transport | Medium | Medium | Fleet-orchestrated |
| Humanoid | Tote handling, destacking | Highest per unit | Highest | Slow, pilot-stage |
Humanoid pilots are real but narrow. Amazon, GXO, Schaeffler and Toyota Motor Manufacturing Canada operate Agility Robotics’ Digit in live deployments, and Gartner expects fewer than 20 companies to scale humanoids to production in manufacturing and supply chain by 2028. Treat vendor volume targets as forecasts, not data.
Practical Workflow: Estimating Local Impact and Payback
Use this script to size both sides: what a robot cell does to local employment (using the 0.2 pp coefficient) and whether it pays back. All defaults are illustrative assumptions. Replace them with your own numbers.
#!/usr/bin/env python3
"""Robot cell payback + local employment exposure estimator."""
def payback_years(capex, opex_year, fte_replaced, fte_cost_year):
"""Simple payback: capex / net annual savings (no discounting)."""
savings = fte_replaced * fte_cost_year - opex_year # $/year
return float("inf") if savings <= 0 else capex / savings
def local_employment_effect(robots_added, local_workers, working_age_pop,
coef_pp_per_robot_per_1000=-0.2):
"""Apply the Acemoglu-Restrepo style coefficient.
coef: change in employment-to-population ratio (percentage points)
per additional robot per 1,000 workers.
"""
robots_per_1000 = robots_added / (local_workers / 1000.0)
delta_epop_pp = coef_pp_per_robot_per_1000 * robots_per_1000
jobs_change = (delta_epop_pp / 100.0) * working_age_pop # pp -> fraction
return robots_per_1000, delta_epop_pp, jobs_change
if __name__ == "__main__":
# Illustrative cell: $150k installed, $18k/yr upkeep,
# two shifts replacing 1.6 FTE at $48k fully loaded.
pb = payback_years(150_000, 18_000, 1.6, 48_000)
print(f"Payback: {pb:.2f} years") # -> ~2.55 years
# Illustrative commuting zone: 100k workers, 250k working-age pop.
r, d, j = local_employment_effect(50, 100_000, 250_000)
print(f"Robots/1000 workers: {r:.2f}") # 0.50
print(f"Delta EPOP: {d:.2f} pp") # -0.10 pp
print(f"Implied job change: {j:.0f}") # about -250
Read the output carefully. A 2.5-year payback explains why adoption accelerates. The regional coefficient shows why the cost lands on specific towns. Neither number includes the productivity spillovers that offset losses elsewhere in the economy.
Real-World Deployment Scenario: Warehouse Intralogistics
A mid-size 3PL adds AMRs for transport and a small humanoid pilot for tote destacking. A realistic sequence:
- Baseline pick, pack, and transport labor hours per shift.
- Deploy AMRs first. They run on mature navigation stacks and fit existing aisles.
- Pilot the humanoid on one bounded task. Humanoid and Siemens, for example, ran a proof of concept on tote-to-conveyor destacking.
- Reassign, don’t just cut. Fleet supervision, maintenance, and exception handling become the new roles.
- Re-measure uptime, throughput per labor hour, and injury rates after 90 days.
Vendors often model humanoid ROI at very low robot prices. One conference presentation assumed a USD 20,000 robot working two shifts and replacing two workers. That is a scenario, not a quote. Plug realistic installed cost and uptime into the script above before trusting any payback claim.
What Engineers Should Do With This
- Design for task complementarity. Cells that remove the heavy, repetitive step and keep humans on judgment steps match the evidence on where employment holds.
- Log displaced-hours explicitly. The studies above infer it indirectly. Your site data is better.
- Plan reskilling. Maintenance and integration roles appear wherever fleets scale.
- Track the distribution. Gains accrue to adopting firms. Losses are local and uneven by gender, race, and education.
FAQ
Do robots take more jobs than they create?
In US regional data from 1990–2007, exposure to robots reduced employment and wages in affected areas. Firm-level data from France, Spain, and China show adopting firms often grow. The net effect depends on task substitution and local industry mix.
How many industrial robots are in use in 2026?
The IFR reports a record operational stock of about 5 million industrial robots at the end of 2025, with roughly 603,000 installed that year.
Are humanoid robots replacing warehouse workers now?
Not at scale. About 7,000 humanoids were sold worldwide in 2025, and deployments are pilots or bounded tasks like tote handling and palletizing.
Which workers are most exposed to robots?
Evidence points to male, less-educated, lower-paid workers in routine manual and assembly roles. Studies also find larger employment declines for non-White workers than for White workers.
