Goldman Lifts Humanoid Robot Forecast to 6.5 Million Units

Goldman Sachs has multiplied its 2035 humanoid-robot shipment forecast nearly fivefold, turning a futuristic automation theme into a potentially $138 billion industrial market.

The bank now expects roughly 6.5 million humanoid robots to ship in 2035, up from its previous estimate of about 1.4 million, according to its latest physical-AI research. Its interim forecasts also rose sharply, with 2026 shipments lifted to 75,000 units from 51,000 and 2030 shipments raised to 890,000 from 256,000.

That pace would require more than better demonstrations at technology conferences. Manufacturers would need to drive down actuator, battery and sensor costs while making robots reliable enough for repetitive work in warehouses, factories and automotive plants. Goldman’s earlier research identified product design, affordability, use cases and public acceptance as the main barriers to mass adoption.

The revised forecast also changes the investment map. Tesla’s Optimus program gives the automaker a direct claim on the market, although the company still has to prove it can manufacture and deploy the machines at scale. Nvidia sits farther upstream, supplying the computing infrastructure that physical-AI systems need to interpret environments and make decisions outside the data center.

Goldman estimates that each humanoid could contain between $3,000 and more than $6,000 worth of semiconductor content, including compute, analog and mixed-signal chips, sensors and edge storage. At 6.5 million units, that implies roughly $19.5 billion to $39 billion in chip content tied to annual production at full scale, though the figure is an industry opportunity rather than revenue guaranteed to any one supplier.

The forecast is aggressive, but it is not without precedent. Goldman’s 2024 report had already lifted its projected humanoid market to $38 billion by 2035 from $6 billion, while warning that general-purpose robots had not yet proved commercially viable.

For investors, the bigger question may be less which robot wins than who sells the components, software and factory systems required by all of them. That favors the picks-and-shovels layer, but only if deployment moves from pilot projects to repeatable production.