Humanoid Robot Forecast Jumps to 6.5 Million Units as Goldman Sachs Bets on Physical AI
Goldman Sachs has raised its humanoid robot forecast to 6.5 million units shipped in 2035, roughly five times its previous estimate, in an 80-page report that treats physical AI as an emerging industrial market rather than a research program. The revision, published this week, also lifts the bank's 2030 projection from 256,000 units to 890,000 and its 2026 figure from 51,000 to 75,000. Goldman now values the segment at about $138 billion by 2035.
The headline number is the least revealing part of the document. What changed is Goldman's explanation of why adoption accelerates: falling actuator, sensor and compute costs meeting a shortage of industrial labor that shows no sign of closing. Those two forces reinforce each other, and the suppliers sitting at their intersection are the ones the report treats as the beneficiaries.
What the New Humanoid Robot Forecast Says
A fourfold-plus upgrade to a decade-out estimate is unusual on its own. The nearer-term numbers moved with it, and those are harder to defend, because a 2026 figure can be checked against orders within a few quarters.
| Year | Previous forecast | New forecast |
|---|---|---|
| 2026 | 51,000 units | 75,000 units |
| 2030 | 256,000 units | 890,000 units |
| 2035 | 1.4 million units | 6.5 million units |
The 2030 revision carries the most weight. Moving that year from roughly a quarter-million units to nearly 900,000 means Goldman expects the inflection point to arrive well before the end of the decade, not in its final years. A forecast shaped that way depends on manufacturing capacity and component supply scaling at the same time as demand, a harder claim than a simple demand projection.
For scale, Goldman's earlier work put 2025 global shipments at about 20,000 units. The new 2026 estimate of 75,000 implies that volume roughly triples in a single year, a step change the report ties to structured environments rather than general-purpose use.
The Cost Math Behind the Ramp
The most concrete argument behind the upgraded humanoid robot forecast is a labor-cost comparison. JPMorgan estimates that a humanoid robot can operate for roughly $10 to $12 per hour, against about $30 per hour for comparable human factory labor. That gap is the core of the bull case, and it explains why overnight industrial shifts are treated as the first beachhead: unattractive hours are the hardest to staff and the easiest to automate.
The comparison needs one caveat that the numbers themselves invite. Robots remain less productive than human workers today. If a machine produces less output per hour, its effective cost per unit of work sits above the headline rate. The $10-to-$12 figure is therefore a target that arrives when reliability and throughput converge, not a price available today.
Capital cost sits outside the hourly comparison as well. Operating cost covers energy, maintenance and supervision; the purchase price of the units and the engineering required to integrate them into an existing facility do not appear in it. For a warehouse operator weighing a fleet purchase, the payback period depends on utilization rates over several years, not on a single shift's labor arbitrage. The hourly rate is a starting point for that calculation rather than the answer to it.
What makes the timeline plausible is that three separate cost lines are falling together. Actuators, sensors and onboard compute have all declined, and each one feeds directly into the bill of materials. Goldman's report connects that decline to the same industrial labor shortage that gives buyers a reason to act before the economics are fully settled.
Where the Value Lands
Goldman's most specific disclosure is the semiconductor content per unit: $3,000 to $6,000 per robot. At the forecast 2035 volume, that works out to roughly $20 billion to $39 billion in cumulative chip content. The robotics ramp is therefore an AI chip demand story as much as a manufacturing one, and it pulls chip suppliers into a market they have mostly served indirectly.
The supply-chain consequence is that robotics becomes a second demand channel for AI silicon. Data center accelerators have been the dominant driver of AI chip demand; onboard robot compute would add a separate buyer for the same fabrication and memory capacity. A second channel does not remove cyclicality, but it spreads exposure across markets that do not peak at the same time.
The implied revenue math is worth sitting with. A $138 billion market across 6.5 million units implies an average selling price near $21,000 per robot. Semiconductor content at $3,000 to $6,000 would then account for roughly 14% to 28% of the price of each unit, an unusually high component share for capital equipment and a signal of how much of the value sits in electronics rather than metal.
Geography matters as much as component mix. Goldman's analysts estimate that Korean companies will hold a 30% direct and indirect share of global humanoid production by 2035, with about 74,000 units built on Korean supply chains by 2030. Read against the global forecast, that implies Korean supply chains cover roughly 8% of 2030 volume before the share climbs toward 30% five years later.
On the demand side, Goldman points to structured environments such as warehouses and logistics for first adoption, and names Amazon and Tesla among the leading players in physical AI and robotics. Both companies control their own deployment sites, which lets them absorb early reliability problems without exposing outside customers to them.
Goldman's note identifies e-commerce warehouses and automotive production lines as the fastest-growing applications, which widens the initial beachhead beyond logistics. Vehicle assembly adds different constraints: cycle times measured in seconds, heavy payloads and strict safety certification. Those conditions are harder to satisfy than a warehouse aisle, so the ordering in the report is a sequencing claim rather than a simultaneous one.
The Trade-offs and Open Questions
Forecast dispersion is the clearest warning sign. Wedbush and SoftBank have described physical AI as a trillion-dollar market, a framing far larger than Goldman's $138 billion unit-revenue figure, and other Goldman notes published this year have put the 2035 market closer to $38 billion. Those gaps reflect different assumptions about unit prices and volume, and they are wide enough that any single number is better read as a scenario than an outlook.
The labor economics also cut both ways. A robot that costs less per hour but produces less per hour does not automatically displace a human worker. It displaces the specific task a human was performing, which is why the report's emphasis on warehouses and logistics is a limit as much as a starting point. Structured environments have predictable layouts, stable lighting and repetitive motion, the conditions where current reliability is adequate.
The 2026 figure is where the forecast becomes testable. If 75,000 units ship this year, the 2030 path turns into a capacity question; if deployments stay in pilot programs of tens or hundreds of machines, the later numbers rest on assumptions that no purchase order has yet confirmed.
Component suppliers face their own version of the trade-off. A faster ramp rewards whoever adds capacity first, but capacity built for a forecast that slips by three years is expensive to carry. The fivefold revision is a demand signal, and demand signals this early in a hardware cycle have historically been revised in both directions.
Why This Matters
The revised humanoid robot forecast matters less as a prediction than as a reallocation signal. If even part of the projected volume arrives, chipmakers, actuator suppliers and logistics operators face demand curves they did not plan for, and the labor-cost argument gives buyers a reason to move before the technology is fully mature. For decision-makers, the practical question is which parts of the supply chain are already being priced for that outcome.
Sources
South Korea's Growing Role in Humanoid Robot Development | Goldman Sachs
Related Articles
- Humanoid Robot Shipments Top 22,000, Yet Factory Use Lags [Update]
- OpenAI Humanoid Robot Confirmation Points to a Bigger Hardware Play
- How the Hyundai Boston Dynamics Acquisition Reshapes Physical AI Strategy
✔Human Verified
Researched and cross-referenced against primary sources by the Bytevyte editorial team. This article was generated with the assistance of artificial intelligence and reviewed by the Bytevyte editorial team.