Humanoid Market Forecast 2030–2035: Which Numbers Should Industrial Buyers Believe?
Physical AI Journal Research Team — Independent analysis by Sekason Research Limited, United Kingdom
Published: physicalaijournal.org | 3 September 2026 | Report Type: Executive Intelligence — Premium Research
1. Executive Intelligence Synthesis
Automation and procurement buyers deciding which humanoid market forecast to budget against in 2026–2028 face a 10× range — from approximately $15 billion (Interact Analysis, May 2026) to $154 billion (Goldman Sachs blue-sky) — that reflects five different market definitions, not competing measurements of the same market. The central decision this report supports: which forecast assumptions are credible enough to plan against, which serve as stress-test inputs, and which belong only in upside optionality models. The five signals below are the most operationally consequential findings from the 2026 evidence base.
Signal 1 — The $15 billion to $154 billion range is an assumption dispute, not a measurement error.
Interact Analysis' ~$15 billion revenue projection for 2035 counts commercial humanoid hardware shipments in a conservative near-term commercial scenario, with the inflection point arriving only after 2032. Goldman Sachs' $154 billion figure is an explicitly labelled blue-sky scenario, contingent on — in Goldman's own framing — the complete resolution of every major barrier to adoption, including product design, use cases, technology, affordability, and public acceptance. These are not two estimates of the same thing. A procurement model built on the $154 billion figure as a base-case planning assumption has misread the source material.
Signal 2 — Only approximately 10% of produced humanoids were in real-world industrial operations in 2025.
Interact Analysis (May 2026) found that approximately 10% of humanoids produced were deployed in real-world operations — the remaining 90% allocated to research, development, demonstrations, education, and non-commercial applications. Commercial inflection is projected only after 2032. For a buyer constructing a deployment timeline, this figure is more operationally useful than any headline market-size number: it establishes how far the industry currently sits from the deployment density that aggressive forecast scenarios require.
Signal 3 — China's government procurement is not the same as commercial industrial demand.
Chinese government procurement of humanoids and related equipment reached at least $230 million in the first half of 2026 alone, according to Reuters — compared with $62 million across full-year 2025 and approximately $6 million in the comparable period of 2024. Observed Chinese unit volume is growing at a striking rate. A buyer extrapolating global commercial demand from that growth rate should first subtract the policy-procurement component. Subsidised state purchases are not commercially-justified industrial deployments, and they should carry different weight in any demand-signal analysis.
Signal 4 — No major platform has published a full public ROI dataset from an independent industrial customer.
The most operationally detailed public evidence in the record is BMW's Figure 02 deployment at Spartanburg, South Carolina: 10 months of operation, 10-hour shifts Monday through Friday, more than 90,000 components moved, approximately 1,250 operating hours, and support for production of more than 30,000 BMW X3 vehicles. BMW has not publicly disclosed the robot's total cost, labour savings, payback period, or comparative economics against alternative automation. That disclosure gap is not incidental — it is the most commercially important data point in the landscape: confirmed operation exists; confirmed economics do not.
Signal 5 — Country-of-origin risk entered the procurement calculation in July 2026.
The United States announced restrictions on imports of new Chinese humanoid and quadruped robots in July 2026, citing national-security, cybersecurity, and supply-chain concerns, according to Reuters, with Unitree identified as a company likely affected. A buyer whose evaluation includes Chinese-manufactured platforms must now factor country-of-origin risk into vendor selection — a variable absent from most pre-2026 procurement frameworks. This restriction also creates an asymmetry in market-demand forecasts that weight Chinese production capacity as a proxy for global commercial adoption.
2. The Humanoid Forecast Landscape: $15 Billion to $154 Billion and What Each Really Means
Humanoid robot market forecasts for 2030–2035 range from approximately $15 billion to more than $154 billion because they measure fundamentally different things. Interact Analysis' May 2026 projection counts near-term commercial hardware revenue in a conservative commercial-market scenario. Goldman Sachs' $154 billion figure is an explicitly labelled blue-sky scenario. Morgan Stanley's $5 trillion estimate encompasses the entire 2050 ecosystem — hardware, software, services, and broader economic value. Comparing these numbers without decomposing their definitions produces procurement conclusions that are systematically and materially wrong.
Industrial buyers facing the humanoid robot market forecast debate in 2026 encounter a problem that is structural, not statistical. The headline numbers — $15 billion, $38 billion, $154 billion, $5 trillion — do not represent competing measurements of the same market. They represent five different markets, at three different time horizons, with four different scoping methodologies. A planning framework that treats them as competing predictions about the same market has asked the wrong question.
Interact Analysis published its 2035 commercial humanoid forecast in May 2026 — the most recent major benchmark in the field. It projects approximately $15 billion in revenue and more than 700,000 annual shipments by 2035, with the commercial inflection point arriving only after 2032. The methodology is grounded in commercial deployment data and explicitly counts hardware shipments into real-world industrial and commercial applications. Current growth is driven by small-scale pilots, subsidies, and strategic partnerships, not yet by broad commercial industrial adoption. China and the United States are projected to account for more than 85% of demand in 2035, with China alone representing more than 65% of real-world application shipments.
Goldman Sachs published its humanoid robot analysis in February 2024, projecting $38 billion in market revenue and 1.4 million units by 2035 in its central scenario. This is Goldman's central estimate. It represents a credible acceleration scenario — one that requires meaningfully faster cost reduction and broader industrial adoption than the Interact Analysis base case, but one grounded in explicit commercial assumptions. This is the number a buyer should use to stress-test against, not the number that has dominated headlines.
Goldman's $154 billion figure is, in Goldman's own framing, a blue-sky scenario. Goldman describes it as contingent on the complete resolution of barriers around product design, use cases, technology, affordability, and public acceptance. It is not Goldman's prediction. It is Goldman's ceiling — the figure that results if every major friction disappears on an accelerated timeline. A strategy document that treats $154 billion as a base-case planning assumption has made a reading error with direct capital-allocation consequences.
UBS published its humanoid forecast in June 2025, projecting more than 2 million humanoids installed by 2035 and 300 million by 2050, with approximately 63% of 2030 use cases remaining industrial. This is a unit-count forecast, not a revenue figure. The revenue equivalent depends critically on average selling price — itself one of the most contested variables in the market. The UBS forecast is useful as a unit-count lens that is broadly consistent with the industrial-leadership assumption visible in 2026 deployment evidence.
Morgan Stanley's $5 trillion market estimate, published in May 2025, encompasses the entire humanoid ecosystem by 2050: hardware, software, services, integration, and the broader economic value generated by humanoid deployment. It is a long-range total addressable market scenario, not a 2035 commercial revenue forecast. Applying it as a procurement planning number requires disaggregating 25 years of ecosystem development — a step that requires assumptions the source document does not provide.
Citi projected almost 650 million humanoids by 2050 and a potential $7 trillion market after price declines are factored through, published in December 2024. Like Morgan Stanley's estimate, this is a long-range, technology-curve-dependent scenario with a 24-year horizon. It is not a 2026–2035 procurement benchmark.
Table V1 — Humanoid Forecast Comparison Matrix
Organisation | Forecast Figure | What It Counts | Scope | Timeline | Key Exclusion | Procurement Classification | Source Date |
Interact Analysis | ~$15B revenue; >700,000 annual shipments | Commercial hardware revenue + shipments | Commercial market, conservative scenario | 2035 | Consumer/long-range scenarios | High Procurement Confidence | 19 May 2026 |
Goldman Sachs (central) | $38B; 1.4M units | Revenue + units | Humanoid robot market | 2035 | Blue-sky contingencies | Moderate Procurement Confidence | 27 Feb 2024 |
Goldman Sachs (blue-sky) | Up to $154B | Revenue scenario | Humanoid robot market | 2035 | Requires all barriers overcome | Speculative / Upside Only | 26 Feb 2024 |
UBS | >2M humanoids by 2035 | Installed-population units | Installed humanoid fleet | 2035 | Revenue; service/software ecosystem | Moderate Procurement Confidence | 20 Jun 2025 |
Morgan Stanley | >$5T | Broad ecosystem TAM | Hardware + software + services + ecosystem | 2050 | Near-term commercial market | Speculative / Upside Only | 14 May 2025 |
Citi | ~650M humanoids; ~$7T market | Units + potential TAM | Installed population | 2050 | Revenue; technology assumptions | Speculative / Upside Only | 5 Dec 2024 |
Note: A Barclays optimistic-scenario figure (~$200 billion by 2035) was identified in preliminary research but excluded from this matrix because source-URL verification was not completed prior to publication. All forecasts classified as Verified analyst projections.
The procurement implication of this definitional analysis is direct. A buyer whose capex model rests on the $154 billion blue-sky number is testing their organisation against Goldman's ceiling scenario — not Goldman's expectation. A buyer using Morgan Stanley's $5 trillion as a market-size reference is using a 2050 ecosystem estimate to justify a 2027 deployment decision. Neither is analytically defensible without explicit acknowledgement of what those numbers require to be true.
3. Deployment Evidence vs. Forecast Assumptions
The commercial deployment evidence available in 2026 does not yet support the assumptions required by the accelerated or blue-sky forecast scenarios. Interact Analysis (May 2026) found that only approximately 10% of produced humanoids were in real-world operations in 2025, with commercial inflection expected only after 2032. A buyer whose planning horizon ends before 2032 is building to an assumption that requires the adoption curve to accelerate materially beyond what current deployment evidence shows.
Testing forecast assumptions against deployment evidence is the analytical step that most existing coverage skips. The evidence is specific enough to be useful — and specific enough to raise questions that the aggressive scenarios cannot currently answer.
The most important single figure is the 10% deployment rate. Interact Analysis assessed in May 2026 that approximately 10% of produced humanoids were in real-world operations in 2025. The remaining 90% were allocated to research, development, demonstrations, education, and non-commercial use. Manufacturing and warehousing currently lead real-world application because those environments offer more structured conditions and have attracted early technology adopters. For a buyer whose capex model assumes broad-based industrial deployment pre-2030, this figure is a direct empirical challenge to that assumption.
The autonomy constraint is equally specific. Reuters' August 2026 investigation into China's humanoid manufacturing landscape found that even robots in industrial settings were still relying substantially on choreographed movement sequences and human trainers for basic factory tasks. In some training environments examined, inexperienced trainers achieved approximately one usable movement in roughly 300 attempts. This observation reflects specific conditions — it is not a universal characterisation of every platform in every environment. It does, however, illustrate the gap between demonstration capability and the sustained, low-intervention autonomous industrial operation that aggressive forecast scenarios assume as a precondition for broad adoption.
The China subsidy signal requires careful interpretation. Government procurement of humanoids and related equipment reached at least $230 million in the first half of 2026 — up from $62 million across full-year 2025 and approximately $6 million in the comparable period of 2024, per Reuters reporting. Chinese unit volumes are growing at a rate that will influence headline shipment figures used in market-size forecasts. A buyer extrapolating global commercial demand from Chinese production data should first disaggregate the state-procurement component from organic commercial industrial demand. Those are not interchangeable signals. Demand driven by government subsidy does not validate the commercial ROI assumptions embedded in aggressive forecast scenarios.
The Deployment Conversion Ratio — the proportion of announced pilot commitments that have converted into sustained, paid, production-environment deployments with verified operational outcomes — is the metric most useful for tracking the industry's commercial maturity. No industry source currently tracks both the numerator and the denominator with methodological consistency. That absence is itself a procurement signal: the industry's commercial-deployment evidence base is not yet deep enough to compute a reliable conversion rate.
Table V2 — Forecast Assumption vs. Deployment Evidence Test
Assumption Category | What Aggressive Forecasts Require | What 2026 Evidence Shows | Classification | Gap Assessment |
Commercial deployment rate | Rapid expansion from 2026–2028; inflection pre-2030 | ~10% of produced humanoids in real-world operations in 2025; inflection expected post-2032 | VERIFIED (Interact Analysis, May 2026) | Significant — aggressive scenarios require 5–8 years of acceleration beyond current trajectory |
Robot autonomy | Sufficient for diverse unstructured tasks with minimal supervision | Reliance on choreographed routines and human trainers for basic factory tasks reported in some environments | VERIFIED reporting (Reuters, August 2026) | Significant — autonomy gap constrains task-coverage breadth aggressive forecasts require |
Task coverage breadth | Broad application across manufacturing, logistics, and service sectors | Manufacturing and warehousing lead; consumer/home applications commercially unproven at scale | VERIFIED (Interact Analysis, UBS) | Moderate — industrial-sector focus is consistent with evidence; service-sector breadth is not |
Industrial vs. subsidised demand | Organic commercial demand driving shipment growth | China government procurement: $230M H1 2026 vs. $62M full-year 2025 | VERIFIED (Reuters, August 2026) | Significant — subsidy contribution to observed volume is material and accelerating |
Pilot-to-repeat conversion | High conversion from pilots to fleet-scale repeat orders | No major platform has publicly confirmed fleet-scale repeat orders from multiple independent industrial customers | Verified absence of evidence | Significant — conversion evidence required by aggressive scenarios does not yet exist |
4. Economics and ROI Analysis: Reverse-Engineering the Forecasts
For each major forecast scenario to materialise by 2035, specific economic conditions must hold at scale — conditions that verified public evidence in 2026 does not yet confirm. Interact Analysis' conservative ~$15 billion scenario implies an average revenue per commercial unit of approximately $21,000 by 2035 (Physical AI Journal calculation from verified Interact Analysis projections). Goldman Sachs' $38 billion central scenario implies a higher average across more valuable use cases (Physical AI Journal calculation from verified Goldman Sachs projections). A procurement buyer's most productive question is not 'which forecast is correct?' but 'which economic conditions are plausible given the evidence verifiable in 2026?'
The implied unit economics of each forecast scenario reveal more useful procurement intelligence than the headline revenue figures. The ~$15 billion / >700,000 annual shipments projection from Interact Analysis implies an average revenue of approximately $21,000 per commercial unit by 2035 (Physical AI Journal calculation from verified Interact Analysis projections) — a number that assumes meaningful price decline from current industrial-grade platform pricing, but not the price compression that consumer-technology adoption curves sometimes generate.
The Goldman Sachs central scenario of $38 billion / 1.4 million units implies an average closer to $27,000 (Physical AI Journal calculation from verified Goldman Sachs projections), suggesting a market where higher-value industrial applications command premium pricing at meaningful scale. The $154 billion blue-sky scenario implies a market structure that requires both volume and per-unit economics that have no verified precedent in the current deployment data.
Current publicly listed prices — all company-claimed and not independently verified — illustrate the wide range the market currently spans.
Unitree's R1 is listed from $4,900 (company-claimed), intended primarily for research and education applications. The G1 is listed from $13,500 (company-claimed). The H2 is listed at $29,900 (company-claimed). 1X's NEO carries a list price of $20,000 (company-claimed) with a subscription option at $499 per month (company-claimed). NEURA Robotics' 4NE1 Gen 3.5 is priced at €98,000 for orders of 1–19 units and €60,000 for orders of 20 or more (company-claimed). At the industrial end, Figure AI's Figure 03, Apptronik's Apollo, and Agility Robotics' Digit v5 carry no publicly disclosed purchase price. The Physical AI Journal humanoid robot pricing analysis (physicalaijournal.org/humanoid-robot-pricing-2026) covers the available commercial-platform pricing data in detail.
The price-threshold observation from Jörg Burzer, Member of the Board of Management of Mercedes-Benz responsible for Production, Quality & Supply Chain Management, is the most operationally grounded pricing signal in the public record. Reporting from Reuters in March 2025 indicated that Burzer viewed the point at which costs reach a two-digit-thousand-dollar range as the level that would make large-scale industrial adoption significantly more attractive to Mercedes-Benz. That threshold begins to overlap with the lower end of the Unitree price range — but Unitree's research- and education-class platforms serve different market segments from the industrial-grade applications that automotive production demands. For the industrial-grade systems with no published price, the two-digit-thousand-dollar threshold remains an aspiration rather than a current market reality.
Marco Wang, Research Analyst at Interact Analysis, identified in May 2026 that 'technology readiness remains a primary constraint' on economic deployment — alongside embodied-AI capability, physical data scarcity, hardware durability, manufacturing consistency, safety standards, and certification. The constraint list is long enough to explain why the economics analysis returns consistently to the same conclusion: no major industrial customer has published a complete deployment economics dataset covering total acquisition cost, maintenance, integration, uptime, productive task cycles, human intervention frequency, and payback period against a conventional-automation baseline. Without that dataset, a buyer cannot construct a defensible cross-platform TCO model from public evidence alone.
The Sector ROI Model and the warehouse ROI analysis (physicalaijournal.org/warehouse-humanoid-roi-2026) provide the most granular available modelling frameworks for specific use-case assessments.
Table V3 — Three-Scenario Economics Matrix
Scenario | Market Size 2035 | Implied Annual Units | Implied Revenue / Unit | Key Enabling Condition | Procurement Planning Use |
Conservative | ~$15B (Interact Analysis) | >700,000 | ~$21,000 (Physical AI Journal calculation from verified Interact Analysis projections) | Commercial inflection after 2032; gradual reliability and cost improvement | Base-case capex assumption |
Accelerated | ~$38B (Goldman Sachs, central) | 1.4M units | ~$27,000 (Physical AI Journal calculation from verified Goldman Sachs projections) | Faster cost reduction; improved autonomy; broader industrial adoption by 2030 | Stress-test scenario |
Blue-Sky | Up to $154B (Goldman Sachs) | Not specified | Substantially higher | Complete resolution of all major barriers: product design, use cases, technology, affordability, acceptance | Upside optionality model only |
Per-unit revenue figures are Physical AI Journal calculations from cited analyst projections, not directly stated figures. Analyst forecasts classified as Verified.
Table V4 — The Five Forecast Variables
Variable | Conservative Assumption | Accelerated Assumption | Current 2026 Evidence | Classification |
Robot average selling price | Gradual decline; premium pricing persists through 2032 | Rapid reduction toward $10K–$20K industrial-grade range by 2028–2030 | NEURA 4NE1 from €60,000–€98,000 (company-claimed); Unitree G1 from $13,500 (company-claimed, research class) | COMPANY-CLAIMED; market not price-settled |
Annual shipments | >700,000 by 2035; modest near-term growth | 1.4M+ by 2035; meaningful pre-2030 commercial volume | ~10% of produced humanoids in real operations in 2025 | VERIFIED (Interact Analysis) |
Industrial adoption rate | Manufacturing/logistics lead; slow broadening | Rapid multi-sector adoption: automotive, logistics, healthcare | Manufacturing and warehousing lead; consumer/home commercially unproven | VERIFIED (Interact Analysis, UBS) |
Productive utilisation | Indicative ~40–60% uptime in real operations (editorial modelling assumption) | Indicative ~70–80%+ at commercial scale (editorial modelling assumption) | BMW Figure 02: ~1,250 operating hours across 10 months of Mon–Fri shifts — partial-shift deployment | VERIFIED (BMW press release) |
Task coverage / autonomy | Structured, repetitive tasks; limited dexterity breadth | Broad autonomous capability across variable tasks | Choreographed routines and human trainers still required in some environments for basic factory tasks | VERIFIED reporting (Reuters, August 2026) |
Utilisation percentages in Table V4 are editorial modelling assumptions for illustrative purposes, not directly sourced figures from Step 2 research.
5. Named Deployment Case Studies: What Happened After the Press Release
Four named industrial deployments provide the strongest publicly available evidence of humanoid robots in operational settings as of 2026. They sit at four different stages of the PAJ Evidence Pipeline — Production, Commercial Deployment, Pilot/Testing, and Announcement/Planned — and the distinction between those stages is the most important analytical discipline in the forecast debate. Only the first two generate operational evidence. The latter two generate none. All four are drawn from Step 2 verified research; no case study claims have been added or extrapolated.
Case 1 — BMW × Figure AI: Spartanburg, South Carolina (Automotive Manufacturing)
BMW Group deployed Figure AI's Figure 02 at its Spartanburg, South Carolina manufacturing facility during 2025, subsequently progressing to a Figure 03 logistics and sequencing application. The Figure 02 deployment ran for 10 months on 10-hour shifts Monday through Friday. It moved more than 90,000 components, accumulated approximately 1,250 operating hours and approximately 1.2 million steps, and supported production of more than 30,000 BMW X3 vehicles, according to BMW Group press releases. BMW then announced the deployment of Figure 03 at Spartanburg in a logistics and sequencing role.
BMW has not publicly disclosed the robot's total acquisition cost, labour savings, payback period, or comparative economics against alternative automation options for these tasks.
PAJ Evidence Pipeline classification: PRODUCTION. This is the most operationally mature named humanoid deployment in the public record. It confirms that a humanoid robot operated in a sustained production environment over a 10-month period at a major industrial facility. It does not establish cross-platform ROI economics, confirm that the deployment was economically superior to purpose-built automation for these tasks, or provide a replicable commercial template.
The deployment evidence analysis (physicalaijournal.org/humanoid-deployment-evidence-report-2026) provides context on how this case compares with other reported deployments.
Case 2 — GXO × Agility Robotics Digit: Flowery Branch, Georgia (Logistics)
GXO Logistics began what Agility Robotics describes as the industry's first commercial humanoid Robots-as-a-Service deployment (company-claimed by Agility Robotics) at its Flowery Branch, Georgia facility in June 2024, following a proof-of-concept pilot. Agility reports that Digit has moved more than 100,000 totes at the facility (company-claimed). The arrangement is structured as a multi-year RaaS agreement, and Agility subsequently secured a $2.5 billion pre-money valuation through a SPAC transaction in June 2026 (verified through SEC filing) and announced more than $300 million in contracted Digit v5 orders (company-claimed).
Neither GXO nor Agility has publicly disclosed RaaS pricing, operational uptime rates, cost comparisons against conventional warehouse automation, or customer-level ROI. For buyers evaluating the RaaS commercial structure in detail, the RaaS economics analysis (physicalaijournal.org/robots-as-a-service-vs-leasing-vs-buying-humanoid-robots) covers the financial and operational trade-offs of RaaS versus leasing versus capital purchase.
PAJ Evidence Pipeline classification: COMMERCIAL DEPLOYMENT. The underlying commercial deployment is independently confirmed. Outcome performance figures originate from the vendor and carry company-claimed status. Customer-level economics are not in the public record.
Case 3 — Mercedes-Benz × Apptronik Apollo: Berlin-Marienfelde and Kecskemét (Automotive)
Reuters reported in March 2025 that a small number of Apptronik's Apollo robots were under testing at Mercedes-Benz production facilities in Berlin-Marienfelde, Germany, and Kecskemét, Hungary. Applications included moving components and quality checks, with teleoperation used to train the robots for production tasks. Mercedes-Benz had also made a low-double-digit million-euro investment in Apptronik. By February 2026, Apptronik closed a $520 million Series A extension, bringing total Series A funding above $935 million and establishing a valuation of approximately $5 billion, according to Reuters reporting.
No operational outcome data from the Mercedes-Benz testing programme — productivity figures, task success rates, deployment decision, or any economic comparison — has been publicly reported.
PAJ Evidence Pipeline classification: PILOT/TESTING. A confirmed industrial testing deployment at a tier-one automotive manufacturer. No verified production-scale outcome data exists. The commercial investment by Mercedes-Benz in Apptronik is a strategic signal; it is not ROI evidence.
Case 4 — Schaeffler × Humanoid and Hexagon Robotics: Global Manufacturing Network (Planned)
Schaeffler announced a purchase agreement with Humanoid for its global production network in January 2026. In April 2026, Schaeffler announced plans to deploy at least 1,000 Hexagon Robotics humanoids (company-claimed) across its global production system over approximately seven years. Neither deployment has been reported as commenced. The Humanoid company had separately announced 34,000 pre-orders representing a claimed $2.4 billion in future annual recurring revenue (company-claimed) as of May 2026.
PAJ Evidence Pipeline classification: ANNOUNCEMENT/PLANNED. This is the largest announced industrial humanoid commitment in the public record. It must not be counted as 1,000 deployed robots in any market-sizing model. Announced commitments are procurement signals about industrial appetite — they are not operational evidence. The seven-year deployment horizon means any productivity or economics data from this programme lies well beyond the current decision window for buyers acting in 2026–2028.
6. Friction, Risk, and Unresolved Issues: Seven Constraints That Determine Which Forecast Scenario Plays Out
Seven specific constraints currently limit humanoid robot adoption at commercial scale, and each one directly determines whether the market trajectory tracks the Conservative, Accelerated, or Blue-Sky forecast scenario. A buyer whose capex assumptions require the Accelerated or Blue-Sky case should treat these seven constraints as monitoring conditions — if any one of them persists beyond 2028, the aggressive adoption curve becomes materially less likely. None of these seven constraints was resolved in the 2026 evidence available at the time of publication.
Constraint 1 — The deployment-to-production gap.
Interact Analysis assessed in May 2026 that current growth is driven primarily by small-scale pilots rather than large-scale commercial projects, with commercial inflection expected only after 2032. Pilot count is a poor proxy for addressable commercial demand — and the gap between pilots announced and pilots converted to production deployments is the central uncertainty in every near-term forecast model. The buyer implication: an announcement pipeline does not equal a revenue pipeline.
Constraint 2 — Autonomy remains commercially incomplete for many industrial tasks.
Reuters' August 2026 investigation found that Chinese humanoids in factory settings were still relying heavily on choreographed routines and human trainers for basic industrial tasks — with some training environments achieving approximately one usable movement in roughly 300 attempts. This applies to specific conditions observed, not to all platforms universally. The constraint is real regardless of its scope: the autonomy threshold required for broad commercial deployment across variable-task environments has not yet been publicly demonstrated at scale by any major platform.
Constraint 3 — Subsidy-driven demand cannot be extrapolated as organic commercial demand.
Chinese government procurement reached $230 million in the first half of 2026 — roughly 4× the full-year 2025 figure and approximately 38× the H1 2024 figure (calculated from Reuters-reported figures), according to Reuters reporting. The volume signal is visible. A forecast model that treats state-subsidised procurement as equivalent to commercially-justified industrial demand will overstate the organic adoption rate that underpins long-term revenue projections.
Constraint 4 — Hardware durability, consistency, and manufacturing yield are unresolved at commercial-cycle standards.
Interact Analysis specifically identifies hardware endurance, durability, manufacturing consistency, and efficiency as constraints on economic deployment at scale. A buyer commissioning a multi-year deployment needs evidence measured in sustained operating hours, successful task cycles, and mean-time-between-failures under real production conditions — not demonstration performance.
Constraint 5 — Unresolved economics relative to conventional automation.
Available evidence from 2026 — including Interact Analysis' deployment data and Reuters' August 2026 investigation — establishes that conventional industrial automation remains the primary cost comparator humanoids must beat for any specific task. The operationally correct procurement question is not 'can a humanoid perform this task?' but 'can it outperform the cheapest credible alternative automation architecture on a lifetime-cost basis for this specific task?' No widely available public dataset currently answers that question for more than a handful of specific applications.
Constraint 6 — Unit price against the commercial adoption threshold.
Jörg Burzer of Mercedes-Benz indicated in March 2025 that cost reaching a two-digit-thousand-dollar range would make large-scale industrial adoption significantly more attractive. Most currently available industrial-grade systems — those for which commercial deployment evidence exists — carry no public price, and the available comparable pricing data suggests that the two-digit-thousand-dollar range is not yet the market norm for industrial-grade capability. Price decline is occurring, but the evidence does not yet confirm it is occurring at the rate required by the Accelerated scenario.
Constraint 7 — Geopolitical and supply-chain risk is now a procurement variable.
The United States announced restrictions on imports of new Chinese humanoid and quadruped robots in July 2026, with Unitree identified as a company likely affected. This restriction creates a direct procurement implication for any buyer evaluating Unitree or other Chinese-manufactured platforms for U.S. operations: country-of-origin risk — including both regulatory access risk and supply-chain continuity risk — must now be priced into vendor selection. A global buyer's procurement framework must incorporate geopolitical assumptions alongside technology and economics assumptions.
7. Humanoid Forecast Reliability Matrix: Scoring the Major Forecasts for Procurement Usefulness
When evaluated against five buyer-relevant criteria — definition clarity, evidence base, assumption realism, methodology transparency, and procurement usefulness — the major humanoid market forecasts divide into three tiers. Interact Analysis' May 2026 projection receives the only High Procurement Confidence classification: the most recent major benchmark, most directly grounded in commercial deployment evidence, and most consistent with verified 2026 deployment data. Goldman Sachs' $38 billion central forecast and UBS' unit-count projection receive Moderate Procurement Confidence, suitable for stress-testing. Long-range and blue-sky scenarios carry Speculative / Upside Scenario Only classification.
Of the six major forecasts evaluated, only Interact Analysis' May 2026 projection earns a High Procurement Confidence classification — the one forecast grounded in current commercial deployment evidence with an adoption-timing assumption consistent with verified 2026 data. The three-tier output — High Procurement Confidence / Moderate Procurement Confidence / Speculative / Upside Scenario Only — is a classification for procurement planning purposes, not an assessment of any analyst organisation's research quality.
Table V5 — Humanoid Forecast Reliability Matrix
Forecaster | Forecast | Definition Clarity | Evidence Base | Assumption Realism vs. 2026 Evidence | Procurement Usefulness | Classification |
Interact Analysis | ~$15B revenue; >700K annual shipments by 2035 | HIGH | HIGH — grounded in commercial deployment data; most recent benchmark (May 2026) | HIGH — conservative; post-2032 inflection consistent with verified evidence | HIGH | HIGH PROCUREMENT CONFIDENCE |
Goldman Sachs (central) | $38B; 1.4M units by 2035 | HIGH — clear scope | MODERATE — credible; requires acceleration beyond current trajectory | MODERATE — faster adoption than evidence currently shows | HIGH — useful as stress-test | MODERATE PROCUREMENT CONFIDENCE |
Goldman Sachs (blue-sky) | Up to $154B by 2035 | HIGH — explicitly labelled blue-sky | LOW — requires all barriers completely overcome | LOW — conditional on outcomes with no 2026 evidence base | LOW — not appropriate as base case | SPECULATIVE / UPSIDE ONLY |
UBS | >2M units by 2035 | MODERATE — unit count, not revenue | MODERATE — industrial leadership assumption consistent with evidence | MODERATE — different metric; revenue implication depends on ASP | MODERATE — useful as unit-count lens | MODERATE PROCUREMENT CONFIDENCE |
Morgan Stanley | >$5T by 2050 | MODERATE — broadest scope | LOW for 2026–2030 decisions | LOW for near-term — 24-year horizon | LOW for 2026–2030 capex | SPECULATIVE / UPSIDE ONLY |
Citi | ~650M units by 2050 | MODERATE — unit count, 2050 horizon | LOW for 2026–2030 decisions | LOW for near-term | LOW for 2026–2030 capex | SPECULATIVE / UPSIDE ONLY |
All forecasts classified as Verified analyst projections. Classifications reflect procurement planning suitability, not research quality.
Why Interact Analysis receives High Procurement Confidence:
Three factors converge. The May 2026 publication date makes it the most current major benchmark available — two years more recent than the Goldman Sachs forecasts. The post-2032 commercial inflection assumption is directly consistent with the verified 2026 deployment evidence: only 10% of produced humanoids in real operations, commercial contracts driven by pilots and strategic partnerships rather than mass commercial orders, and autonomy constraints not yet resolved for broad task coverage. The methodology explicitly grounds projections in commercial deployment data rather than theoretical production capacity.
Why Goldman Sachs' $38 billion receives Moderate Procurement Confidence:
Goldman's central scenario is methodologically credible and explicitly scoped. It requires meaningfully faster cost reduction and broader industrial adoption than the 2026 evidence base currently supports — but it is a defensible acceleration scenario, not a speculative ceiling. A procurement programme that cannot survive a $38 billion stress test is more fragile than the buyer may realise.
Why Goldman Sachs' $154 billion receives Speculative / Upside Only:
Goldman's own language is unambiguous — this is a blue-sky scenario requiring complete resolution of all major barriers. The classification mirrors Goldman's own framing. Using it as a planning assumption without validating each of the barrier-resolution assumptions constitutes a methodology error that carries direct capital-allocation consequences.
Why UBS receives Moderate Procurement Confidence:
The unit-count lens and the industrial-adoption leadership assumption are broadly consistent with 2026 evidence. The revenue implication of the UBS forecast depends on average selling price trajectory — itself one of the five most contested variables in the market — making direct revenue comparison with the other forecasts unreliable without additional assumptions.
Why Morgan Stanley and Citi receive Speculative / Upside Only:
Both forecasts have 2050 horizons and broad ecosystem scope that make them investment-thesis tools for capital markets, not procurement planning benchmarks for operations and automation buyers making decisions in 2026–2028. They are analytically defensible as long-range exercises; they are simply not the right tool for the decision this report addresses.
Table V6 — Forecast to Procurement Use: Recommended Application
Planning Purpose | Recommended Forecast | Why |
Capex base case (2026–2030) | Interact Analysis: ~$15B / >700K annual units by 2035 | Most recent; most commercially grounded; consistent with 2026 deployment evidence |
Board-level stress test | Goldman Sachs central: $38B / 1.4M units by 2035 | Credible acceleration scenario; tests whether early deployment commitment is justified against a higher-growth environment |
Upside optionality modelling | Goldman Sachs blue-sky: $154B | Represents maximum plausible market if all barriers are overcome — not a planning assumption |
Long-range investment thesis | UBS units / Morgan Stanley $5T | Useful framing for capital investors with 10+ year horizon; not a 2026–2030 procurement benchmark |
Access the Humanoid Platform Index (physicalaijournal.org/platform-index) for current platform specifications and commercial-deployment status tracking across major platforms.
8. Strategic Recommendations: The 2026–2028 Buyer Decision Framework
Industrial buyers deciding whether to invest in humanoid robots in 2026–2028 should apply a three-tier decision framework grounded in the verified evidence from this report: Pilot Now if specific task and vendor evidence criteria are simultaneously met; Wait 12–24 Months if price trajectory or autonomy improvement makes deferral economically rational; Do Not Proceed if the use case cannot be defined with sufficient precision or vendor evidence is at announcement stage only. The central procurement conclusion: budget against the conservative case; stress-test against the accelerated case; treat the $154 billion blue-sky figure as an upside option, not a base-case capex assumption.
Table V7 — 2026–2028 Buyer Decision Framework
Decision Tier | When This Applies | Key Criteria | Risk of Acting Against This Signal |
PILOT NOW | All five criteria are simultaneously met | Task is structured, repetitive, and precisely defined; economics work at current pricing on a downside-utilisation assumption; vendor has verified sustained production deployment (not announcement-only); human intervention rate is bounded within operational tolerance; downside ROI remains acceptable | Risk: over-investment in a pilot where technology is not yet commercially mature for the specific task |
WAIT 12–24 MONTHS | Any one of four conditions applies | ASP is declining materially toward buyer's threshold; competing vendors entering the task category with meaningfully different economics; current vendor production evidence is announcement- or pilot-stage only; conventional automation cost advantage exceeds the robot improvement curve | Risk: ceding first-mover learning advantage if adoption accelerates faster than the Conservative scenario |
DO NOT PROCEED | Any one of four conditions applies | Use case requires general-purpose dexterity not commercially demonstrated in comparable environments; ROI depends on theoretical labour savings with no deployment comparator; vendor cannot provide acceptance criteria, uptime commitments, or maintenance economics; all available deployment evidence for the task is at announcement stage | Risk: capital destruction on a deployment that generates no measurable commercial return |
The PILOT NOW decision requires all five criteria to be present simultaneously — not a majority, not a weighted average. The task must be structured, repetitive, and precisely defined: not a request for a robot that can 'help in the warehouse' but a specific sequence of defined movements in a defined environment with defined acceptance criteria. The economics must work on a downside-utilisation assumption — if the deployment only makes financial sense at 80% uptime and the BMW evidence shows an initial deployment at partial-shift rates, the model is fragile.
For vendor assessment methodology, the vendor due diligence framework (physicalaijournal.org/humanoid-robot-vendor-due-diligence-2026) sets out 20 questions that should be answered before a contract is signed.
For buyers uncertain whether humanoid investment is appropriate to their context at all, the framework for rejecting humanoid investment (physicalaijournal.org/when-not-to-buy-humanoid-robot) sets out the conditions under which the answer should be no.
The WAIT decision is warranted when the economics of deferral outperform the economics of early commitment. Hardware prices across the market are not yet settled for commercial-grade industrial applications — the available published pricing spans from $4,900 (company-claimed, research class) to €98,000 (company-claimed, industrial). If prices are declining on a trajectory that delivers the buyer's threshold within 12–18 months, committing now pays a premium that may not be recovered through first-mover learning. Where a vendor's production evidence consists primarily of announced partnerships and pre-order claims rather than verified operational deployments, deferral limits the risk of committing to a vendor whose manufacturing and support infrastructure has not been tested at commercial scale.
The DO NOT PROCEED conditions are absolute. If the use case requires general-purpose dexterity that no commercially deployed system has demonstrated in a comparable environment, the deployment risk cannot be bounded at an acceptable level. If the ROI calculation depends on theoretical labour savings that have no deployment comparator in the verified evidence base, the model is not a forecast — it is an assumption stack. No board-level enthusiasm for the humanoid market, and no vendor's claimed customer-ROI figure (company-claimed, unverified) that appears in a commercial presentation, changes the analytical status of a poorly-bounded deployment case.
Five quarterly indicators that signal when to revise the scenario upward:
1. Paid production deployments as a share of announced pilot/commitment count — is the Deployment Conversion Ratio improving?
2. Published robot operating hours from named, independent industrial customers — are operational benchmarks accumulating in the public record?
3. Documented repeat orders from independent industrial customers — are the same customers coming back?
4. Average selling price trend across commercial-grade platforms — is the price curve moving toward the commercial threshold?
5. Share of reported deployments generating operating revenue from paying industrial customers rather than from subsidised procurement or R&D funding — is the organic commercial demand signal strengthening?
Wang Xingxing, Founder and CEO of Unitree, captured the market's current duality precisely in August 2026 when he said his company is
'marching towards a 'ChatGPT moment' in embodied intelligence'
— while simultaneously, in the same Reuters report, acknowledging that humanoids are not yet capable enough for mass deployment, with robot decision-making and interaction identified as major bottlenecks. Both halves of that statement are accurate. The buyer who holds only the first half will over-invest on the wrong timeline. The buyer who holds only the second half will miss the procurement window when the evidence does shift. The five quarterly indicators above are the signals that tell the buyer which half is becoming more true.
Budget against the Conservative case. Stress-test against the Accelerated case. Treat the $154 billion blue-sky figure as an upside option — not a procurement assumption. A buyer who leaves this report having made that distinction has the most operationally defensible foundation available in the 2026 evidence base.
9. Executive FAQ
Which humanoid robot market forecast for 2030–2035 should I use as my procurement base case?
Interact Analysis' May 2026 projection of approximately $15 billion in revenue and more than 700,000 annual shipments by 2035 is the most appropriate base case for procurement planning. It is the most current major benchmark, grounds its projections in commercial deployment data, and projects the commercial inflection point after 2032 — consistent with verified 2026 evidence that only 10% of produced humanoids were in real-world operations in 2025. Goldman Sachs' $38 billion central forecast is the correct stress-test scenario. The $154 billion figure is explicitly labelled a blue-sky scenario by Goldman and should not be used as a base-case planning assumption under any circumstances.
Why do humanoid market forecasts for 2035 range from $15 billion to more than $150 billion?
The 10× range reflects five radically different assumptions about five contested variables: when commercial inflection arrives, how quickly robot prices fall, how autonomous humanoids become for real industrial tasks, how many tasks they can perform economically relative to conventional automation, and whether pilots convert to fleet-scale repeat orders. The forecasts also define their market differently — hardware revenue, unit installed base, and total ecosystem value are not the same metric. Goldman Sachs' $154 billion is an explicitly labelled blue-sky scenario; Interact Analysis' $15 billion is a near-term commercial hardware forecast. Comparing them without decomposing those differences produces analytically unreliable conclusions.
What assumptions would have to be true for the $100 billion-plus humanoid market scenarios to materialise?
Goldman Sachs itself describes the conditions required for its $154 billion scenario as the complete resolution of barriers around product design, use cases, technology, affordability, and public acceptance. In practice, that requires a faster price decline toward commercial-grade pricing in the two-digit-thousand-dollar range for industrial-grade systems, autonomy sufficient for broad unstructured-task coverage with minimal supervision, and pilot-to-fleet conversion rates that current deployment evidence does not yet show. Each condition is individually possible; none is currently confirmed. A buyer should not commit capex to a scenario whose enabling assumptions cannot be validated against current evidence.
What does actual 2026 deployment evidence tell us about industrial humanoid adoption?
The most operationally important figure is that only approximately 10% of produced humanoids were in real-world industrial operations in 2025, according to Interact Analysis (May 2026). The most detailed public evidence is BMW's Figure 02 deployment: more than 90,000 components moved across approximately 1,250 operating hours over 10 months. No major industrial customer has publicly disclosed a complete economics dataset covering cost, payback period, and ROI against a conventional-automation baseline. Reuters' August 2026 investigation found autonomy limitations in some Chinese manufacturing environments. The evidence confirms that sustained production-environment operation is achievable; it does not confirm the commercial economics.
When should an industrial buyer move from a pilot budget to fleet-scale humanoid procurement?
Fleet-scale commitment is justified when five conditions are simultaneously met: the specific task is structured and precisely defined; the economics work at current pricing on a downside-utilisation assumption; the vendor has verified sustained production deployments — not announcements — with accessible service infrastructure; the human intervention rate per shift is bounded within operational tolerance; and the downside ROI remains acceptable at conservative utilisation. Missing any one of these conditions is a signal to extend the pilot rather than scale. The vendor due diligence framework (physicalaijournal.org/humanoid-robot-vendor-due-diligence-2026) covers the 20 questions a buyer should answer before a fleet commitment.
How much credibility should I give vendor production targets and pre-order announcements?
Weight them as directional signals about manufacturing ambition and market appetite — not as evidence of commercial demand. The analytical chain matters: pre-orders do not equal purchase orders; purchase orders do not equal shipments; shipments do not equal deployments; deployments do not equal productive utilisation; and none of the above equals verified ROI. Humanoid's 34,000 pre-orders representing a claimed $2.4 billion in future annual recurring revenue (company-claimed) is a commercially significant signal about industrial appetite — but it is not evidence of 34,000 robots generating commercial operating revenue. Vendor production targets similarly represent manufacturing ambition; a target of 1 million robots per year (company-claimed, Tesla) is a capacity aspiration, not a demand forecast.
10. Scope and Disclaimer
This report is published by Physical AI Journal, operated by Sekason Research Limited, United Kingdom (Company Registration: 14339910), and provides market and adoption intelligence for informational purposes only.
Nothing in this report constitutes legal, financial, investment, engineering, safety-certification, or professional advice of any kind. Physical AI Journal does not evaluate, certify, or endorse the safety, technical conformity, or fitness-for-purpose of any specific robotic platform or deployment. Company-claimed figures are labelled as such throughout and have not been independently verified by Sekason Research Limited; they should not be relied upon as confirmed facts. Analyst forecasts are reproduced as published projections from named organisations and are not treated as guaranteed outcomes.
Readers should obtain independent professional, technical, legal, and financial advice before making procurement, deployment, safety-certification, or investment decisions in relation to humanoid robotics or physical AI systems.
For the full disclaimer applicable to all Physical AI Journal publications, see physicalaijournal.org/disclaimer.
11. References and Strategic Sources
The strategic landscape and market evaluations for humanoid robotics are built upon a comprehensive foundation of tier-one research, verified market reports, industry announcements, and official manufacturer specifications.
Interact Analysis: Published "Humanoid robot revenue to reach $15bn by 2035" on 19 May 2026, classified as Verified — Tier 1 (URL).
Goldman Sachs Research: Released "Humanoid robot: The AI accelerant" on 26 Feb 2024, classified as Verified — Tier 1 (URL).
Goldman Sachs Research: Published "Global humanoid robot market could reach $38B by 2035" on 27 Feb 2024, classified as Verified — Tier 1 (URL).
UBS: Authored "Is the world ready for 1bn robots?" on 20 Jun 2025, classified as Verified — Tier 1 (URL).
Morgan Stanley: Released "Humanoids: A $5 Trillion Market" on 14 May 2025, classified as Verified — Tier 1 (URL).
Citi: Published "The Rise of AI Robots" on 5 Dec 2024, classified as Verified — Tier 1 (URL).
IFR: Published the paper "Humanoid Robots: Vision and Reality" on 14 Aug 2025, classified as Verified — Tier 1 (URL).
Reuters: Reported on "Apptronik raises $520M" on 11 Feb 2026, classified as Verified — Tier 1 (URL).
Reuters: Covered "Mercedes-Benz tests Apptronik Apollo" on 18 Mar 2025, classified as Verified — Tier 1 (URL).
Reuters: Published "Humanoid targets U.S. IPO by 2030" on 13 May 2026, classified as Verified — Tier 1 (URL).
Reuters: Reported on "Humanoid to deploy up to 2,000 robots at Schaeffler" on 13 May 2026, classified as Verified — Tier 1 (URL).
Reuters: Covered "Apptronik launches Robot Park and Apollo 2" on 30 Jun 2026, classified as Verified — Tier 1 (URL).
Reuters: Reported on "U.S. bans new Chinese humanoid robots" on 28 Jul 2026, classified as Verified — Tier 1 (URL).
Reuters: Covered "DeepSeek invests in Unitree IPO" on 6 Aug 2026, classified as Verified — Tier 1 (URL).
Reuters: Published "Unitree CEO and ChatGPT moment" on 20 Aug 2026, classified as Verified — Tier 1 (URL).
Reuters: Investigated how "China humanoids struggle with factory work" on 27 Aug 2026, classified as Verified — Tier 1 (URL).
BMW Group: Outlined the "Figure 03 project at Spartanburg" in 2026, classified as Verified — Tier 1 (URL).
BMW Group: Announced "Humanoid robots in production / Figure 02" on 27 Feb 2026, classified as Verified — Tier 1 (URL).
Tesla Investor Relations: Released the "2025 Annual Report / 2026 Outlook" in Jan 2026, classified as Verified — Tier 1 (URL).
Agility Robotics: Documented the "Digit commercial deployment / GXO" on 5 Jun 2024, classified as Verified — Tier 1 (URL).
Agility Robotics / SEC: Disclosed a "$2.5B transaction and contracted Digit v5 orders" on 24 Jun 2026, classified as Verified — Tier 1 (URL).
Figure AI: Highlighted the "Figure 03 production ramp" on 29 Apr 2026, classified as Company-claimed — Tier 3 (URL).
Figure AI: Documented the "Figure 02 BMW deployment" on 19 Nov 2025, classified as Company-claimed — Tier 3 (URL).
Apptronik: Published "Apollo specifications" on 23 Aug 2023, classified as Company-claimed — Tier 3 (URL).
1X: Detailed "NEO pricing and specifications" in 2025/current, classified as Company-claimed — Tier 3 (URL).
Unitree: Provided "Official robot pricing" under current status, classified as Company-claimed — Tier 3 (URL).
NEURA Robotics: Outlined "4NE1 Gen 3.5 pricing and reservation" in 2026, classified as Company-claimed — Tier 3 (URL).
NEURA Robotics: Announced a "Series C up to $1.4B" on 10 Jun 2026, classified as Verified — Tier 1 (URL).
Schaeffler: Published a "Humanoid partnership announcement" on 13 Jan 2026, classified as Verified — Tier 1 (URL).
Schaeffler: Reported on the "Hexagon partnership / 1,000 robots" on 22 Apr 2026, classified as Verified — Tier 1 (URL).
This report is backed by authoritative research, independent verification, and structured analytical methodology.
© 2026 Physical AI Journal, Sekason Research Limited.
