Goldman Sees 6.5 Million Humanoid Robots Rolling Out by 2035
Goldman Sachs has put a number on the humanoid robot build-out: 6.5 million units by 2035. The forecast is rippling through chip, foundry and automation shares.
Goldman Sachs now forecasts 6.5 million humanoid robots in service by 2035, a projection that is already moving shares across semiconductor, foundry and factory-automation names.
Goldman Sachs has attached a hard number to one of the least disciplined narratives in technology investing. The bank now forecasts 6.5 million humanoid robots in operation by 2035 — a figure large enough to force analysts covering semiconductors, contract chip manufacturing and industrial automation to rework their end-market assumptions, and small enough, relative to the global industrial workforce, to keep the debate alive.
The forecast, reported by 24/7 Wall St, is already showing up in share prices across chips, foundries and factory-floor suppliers. That is the tell: when a sell-side unit forecast starts moving stocks, the market is treating it less as a thought experiment and more as a demand schedule.
Why a unit forecast, not a dollar forecast, carries weight
Total-addressable-market estimates in emerging technology are notoriously elastic. A dollar figure can be stretched by assuming a higher average selling price, or shrunk by assuming cost deflation, without anyone having to defend a shipment assumption. A unit number is harder to hide behind. Six and a half million machines by 2035 implies a specific production ramp, a specific set of factories capable of building them, and a specific bill of materials repeated millions of times.
That is what makes the number tradable. Every humanoid needs actuators, force and torque sensors, power management silicon, battery cells, high-bandwidth memory or its equivalent for on-device inference, and a processor capable of running vision and control models in real time. Multiply a bill of materials by millions of units and the arithmetic lands squarely on the income statements of companies that already exist and already report quarterly.
Goldman has revised this forecast before. The direction of travel in the bank's humanoid work has been upward as prototype demonstrations have improved and as component costs — particularly for actuators and rare-earth-based motors — have come down faster than earlier models assumed. Investors should treat the current 6.5 million figure as a waypoint rather than a settled estimate, because the inputs that drive it are still moving.
Where the exposure actually sits
The instinct when a robotics forecast lands is to buy the robot maker. That is usually the worst-risk-adjusted way to express the view. Humanoid platform developers are competing on software, industrial design and manufacturing scale simultaneously, and most of them are either private or embedded inside much larger companies where robotics is a rounding error on revenue.
The more durable exposure sits one and two layers down the supply chain:
- Logic and edge-inference silicon. On-device processing is non-negotiable for a machine that has to avoid hurting people. Latency budgets rule out sending everything to a data center.
- Foundries. Whoever designs the chip, someone has to fabricate it. Contract manufacturing captures volume regardless of which humanoid brand wins.
- Sensors and precision components. Encoders, harmonic drives, ball screws, force sensors and machine-vision modules are the parts that determine whether a robot can pick up a box without crushing it.
- Factory automation and integration. The firms that already sell industrial robot arms know how to install, service and finance capital equipment on a factory floor. That installed relationship is a moat humanoid startups do not have.
- Power and thermal. Battery chemistry, power semiconductors and thermal management scale directly with unit count.
Investors should note which of these categories already generate revenue today from unrelated end markets. A foundry whose humanoid exposure is zero in 2026 but whose capacity is fungible carries far less forecast risk than a pure-play component supplier whose valuation already discounts the ramp.
How the tape looked going into the call
The forecast landed against a firm but unremarkable equity backdrop. At the most recent close, on Friday 11 September 2026 at 20:00 GMT, the S&P 500 tracker (SPY) finished at $764.29, up 0.85% on the day from a previous close of $757.83, with a day range of $763.60 to $766.38. The Nasdaq 100 proxy (QQQ) closed at $714.88, higher by 0.87% against a prior close of $708.69. The Dow tracker (DIA) ended at $525.79, up 0.97% from $520.75.
At the most recent close, on Friday 11 September 2026 at 20:00 GMT, the S&P 500 tracker (SPY) finished at $764.
Shares in GS last traded at 1029.18, up 0.92% on the session from a previous close of 1019.77, having ranged between 1018.59 and 1042.99. Markets are closed; these are last-traded levels, not live quotes. Nothing in that session's move should be read as a market verdict on the robotics call itself — a sub-1% gain in line with the broad indices is beta, not conviction.
The gap between a forecast and a purchase order
Three things stand between 6.5 million units and the revenue that would justify current robotics-adjacent multiples.
The first is unit economics for the buyer. A humanoid has to beat the fully loaded cost of the labour it replaces, including downtime, maintenance and the cost of reconfiguring a workspace around it. Fixed-position industrial arms cleared that bar decades ago in automotive assembly precisely because the task was repetitive and the workspace could be designed around the machine. General-purpose humanoids are being sold on the opposite premise — that they slot into environments built for people — and that premise has not yet been proven at scale.
The second is manufacturing capacity. Building millions of complex electromechanical systems requires supply chains for precision gearing and high-performance magnets that are geographically concentrated and politically exposed. Any export-control action affecting rare-earth magnets or advanced logic would compress the ramp regardless of demand.
The third is the regulatory and liability framework. Machines that operate in the same physical space as human workers raise workplace-safety questions that no jurisdiction has fully answered. Insurers, not engineers, may end up setting the pace of deployment.
What to watch from here
The useful signals over the next several quarters are not robotics press releases. They are capital expenditure guidance from foundries, order backlogs at precision-component makers, and any line item in an industrial company's disclosure that separates humanoid-related revenue from the rest of the automation business. Until that disclosure exists, the market is pricing a forecast rather than a cash flow.
Investors positioning for the theme should also watch for revisions. If Goldman's estimate moves again — up or down — the same chips, foundries and automation names that responded this week will respond again, and the ones with the thinnest actual revenue exposure will move the most in both directions. That asymmetry is the risk, not the opportunity.
Key facts
- Goldman Sachs forecast: 6.5 million humanoid robots in operation by 2035
- GS last price: 1029.18, +0.92%, as of Fri 11 Sep 2026 20:00 GMT close
- Sectors reacting: Semiconductors, contract foundries, factory automation
- Market backdrop: SPY $764.29 (+0.85%), QQQ $714.88 (+0.87%), DIA $525.79 (+0.97%) at last close
Frequently asked questions
What exactly did Goldman Sachs forecast?
Goldman Sachs projects that 6.5 million humanoid robots will be in operation by 2035. It is a unit forecast rather than a dollar market-size estimate, which makes it harder to stretch with assumptions about pricing and forces a specific view on production volumes and component demand across the supply chain.
Why does a unit number move chip and foundry stocks?
Each humanoid robot requires a repeatable bill of materials: processors for on-device inference, power management silicon, sensors, actuators and battery cells. Multiplying that bill of materials by millions of units produces a concrete demand figure that analysts can add to semiconductor and contract manufacturing revenue models, which is why share prices respond.
Which parts of the supply chain have the most durable exposure?
Foundries, edge-inference silicon designers, precision component makers such as encoders and harmonic drives, factory automation integrators, and power and thermal suppliers. These layers capture volume regardless of which humanoid brand ultimately wins, whereas platform developers face simultaneous competition in software, design and manufacturing scale.
Has Goldman changed this estimate before?
The bank has revised its humanoid robot work previously, with the direction of travel moving upward as prototype capability improved and component costs fell faster than earlier models assumed. Investors should treat 6.5 million as a waypoint rather than a settled figure, since the underlying cost and capability inputs continue to shift.
What could slow the humanoid robot ramp?
Three constraints stand out: buyer unit economics, since a humanoid must beat the fully loaded cost of the labour it replaces; manufacturing capacity for precision gearing and high-performance magnets, which is geographically concentrated; and workplace safety regulation and liability, where insurers may effectively set the pace of deployment.
How did GS shares trade around the forecast?
GS last traded at 1029.18, up 0.92% on the day from a previous close of 1019.77, with a session range of 1018.59 to 1042.99, as of the market close on Friday 11 September 2026 at 20:00 GMT. That move was broadly in line with the major index trackers that session.
Sources
Photo: Pavel Danilyuk · Pexels Licence — source
