Humanoid Robot Market Size: What Goldman Sachs and Morgan Stanley Actually Project

The headline numbers get repeated everywhere — and just changed dramatically. Here's what each bank's forecast actually measures, including Goldman's massive September 2026 revision.

Goldman Sachs made a dramatic revision on September 13-14, 2026: it now projects 6.5 million humanoid robot units shipped globally by 2035, valuing that market at $138 billion — nearly a 5x increase from its own prior estimate of 1.4 million units and $38 billion. The bank also raised its nearer-term 2030 forecast from 256,000 to 890,000 units, and its 2026 estimate from 51,000 to 75,000. Morgan Stanley's much larger $4.7 trillion figure remains a 2050 forecast for the total market including downstream services, assuming more than one billion humanoid robots are eventually deployed. These aren't contradictory numbers — they're measuring different scopes over very different timeframes, and Goldman's own number just moved substantially closer to the more bullish end of that range.

The forecasts, by source

SourceProjectionTimeframeWhat it measures
Goldman Sachs (revised Sept 2026)$138 billion, ~6.5M units shippedBy 2035Hardware market, commercially deployed units
Goldman Sachs (2030, revised)890,000 unitsBy 2030Up from a prior 256,000-unit estimate
Morgan Stanley$4.7 trillion (over $5T incl. supply chain)By 2050Total market incl. services, ~1B units deployed
CitiUp to $7 trillionBy 2050Long-range total market scenario
UBS$1.4 trillionBy 2050Long-range total market scenario
Morgan Stanley (China-specific)$2B in 2026 → $15B by 2030By 2030China market only, external sales
MarketsandMarkets$4-6B (2026) → $13.8B (2028)Near-termGlobal hardware market

Why Goldman raised its forecast nearly 5x

The bank's new 80-page "Physical AI" report attributes the upward revision to faster-than-expected AI progress, falling hardware costs, and growing investment in physical automation — with Goldman increasingly treating reliability, limited real-world training data, and autonomy as engineering problems being actively solved, rather than open-ended reasons to discount the market. At 6.5 million units, the bank estimates each robot carries $3,000-$6,000 in semiconductor content, which alone would generate $19.5-$39 billion in annual chip demand by 2035 — a meaningful new category of AI infrastructure spending layered on top of data-center buildout. Logistics, warehousing, and automotive assembly are flagged as the leading adoption categories. Nvidia CEO Jensen Huang has pointed to Amazon's planned adoption of Nvidia's full physical-AI stack (Omniverse, Cosmos, Isaac, and Jetson) for its warehouse robot fleet as exactly the kind of application-layer deal that could drive the ramp.

Still, Goldman's own report is explicit that 6.5 million units is a scenario, not an order book — the bank's own analysts have flagged the risk that a voluntary AI safety slowdown (a possibility raised publicly by leaders at both Anthropic and OpenAI) remains the single factor that could stall the entire build cycle before it reaches that scale.

Why the numbers still look so different across banks

Every one of these figures is technically correct for what it's measuring — they just aren't measuring the same thing. Some forecasts count only hardware sales; others fold in software, maintenance, and robotics-as-a-service revenue. Some include prototypes and pilot deployments, while others count only units that actually ship commercially. And the timeframes range from a couple of years out to a 2050 horizon, which alone accounts for most of the gap between "$138 billion" and "$4.7 trillion." Independent of the banks, publicly reported 2025 revenue across humanoid robot makers was still under $1 billion — a useful reality check against any of these longer-range projections.

The near-term signal: China is moving faster than expected

Morgan Stanley has revised its China humanoid shipment forecast upward twice within a single year — from an initial 14,000 units to 28,000, then to 50,000 units for this year, with annual shipments projected to reach 446,000 by 2030. That pace of upward revision is arguably more informative than any single long-range total, since it shows commercialization moving from demonstration to real deployment faster than the banks themselves initially modeled — and it's the same directional pattern Goldman's own September revision just confirmed on the global numbers.

What this means for investors

The wide range between forecasts isn't a reason to dismiss the theme — it's a reason to be precise about which claim you're actually underwriting when you invest in it. A bet sized around Goldman's revised $138 billion, 2035 hardware-market number implies very different assumptions than a bet sized around Morgan Stanley's trillion-dollar, 2050 total-market figure. Know which one you're actually holding — and treat any of these numbers as a scenario tied to specific technical and cost assumptions, not a guarantee, given how much Goldman's own number just moved in a single revision.

Frequently asked questions

What is Goldman Sachs' current humanoid robot forecast? As of September 2026, Goldman Sachs projects 6.5 million humanoid robot units shipped globally by 2035, valuing the market at $138 billion — up nearly 5x from its prior estimate of 1.4 million units and $38 billion.

Why did Goldman Sachs raise its humanoid robot forecast so much? The bank cited faster-than-expected AI progress, falling hardware costs, and growing investment in physical automation. It now treats previous obstacles — reliability, limited training data, and autonomy — as engineering challenges being actively solved rather than reasons to discount the market.

How much semiconductor demand would 6.5 million humanoid robots create? At $3,000-$6,000 in semiconductor content per robot, 6.5 million units would generate an estimated $19.5-$39 billion in annual chip demand by 2035.

Related reading

See how these market-size numbers translate into specific fund exposure: BOTZ vs. KOID vs. ROBO compared, and for the valuation context around the AI trade broadly: Grantham's CAPE framework applied to AI.

This is factual, comparative information for research purposes and is not investment advice. Forecasts cited here are from third-party research and are subject to revision — verify current figures directly with each source before making any investment decision.