Humanoid Robots: Deploy in 2026 or Wait Until 2028? The Cost-of-Delay Decision

Physical AI Journal Research Team — Independent analysis by Sekason Research Limited, United Kingdom

Published: August 2026  ·  physicalaijournal.org

Global humanoid shipments reached 19,100 units in H1 2026, yet Gartner forecasts fewer than 20 companies will reach production-stage deployment by 2028 — a gap that defines the decision environment for procurement buyers in 2026. This report delivers the Humanoid Wait-or-Deploy Threshold: the minimum 2028 TCO reduction required before waiting becomes economically superior to deploying now. Deploying generates measurable value only where an organisation has a quantifiable cost of delay, a bounded high-utilisation task, and operational readiness; waiting is the correct path where those conditions are absent and the expected 2028 improvement in cost and capability would outweigh two years of forgone automation value.

---

1. Executive Intelligence Synthesis

Should industrial companies deploy humanoid robots in 2026 or wait until 2028? There is no universal answer. Deploying in 2026 makes the strongest economic case where an organisation has a measurable labour or capacity constraint, a bounded task with high utilisation potential, and a tangible cost of delay. Waiting until 2028 is more rational where current humanoid TCO is economically marginal, the task can be automated by conventional means, and the expected improvement in platform capability and cost by 2028 would outweigh two years of forgone automation value. The critical calculation is the Humanoid Wait-or-Deploy Threshold: the minimum TCO reduction and capability improvement that must arrive by 2028 to make waiting economically superior.

Humanoid robot commercial deployment moved from demonstration to limited production reality in 2026 — but the evidence is unevenly distributed, the economics remain buyer-specific, and the decision to deploy now or wait is not a question any single market forecast can answer.

Five strategic signals define the current decision environment.

Signal 1 — Commercial deployment is real, but still narrow. 

BMW Group's deployment of Figure 02 at its Spartanburg, South Carolina plant ran for ten months and contributed to the production of more than 30,000 BMW X3 vehicles through sheet-metal handling for welding. GXO Logistics entered a multi-year Robots-as-a-Service (RaaS) agreement with Agility Robotics for Digit at the SPANX fulfillment facility in Georgia. Mercedes-Benz is trialling Apptronik's Apollo at its Berlin-Marienfelde Digital Factory Campus. These are verified deployments, not press releases. However, none of these operators has publicly disclosed a complete return on investment calculation covering acquisition cost, integration, labour saving, downtime, and payback period. Deployment is real; proven ROI is not yet publicly established.

Signal 2 — Manufacturing scale is accelerating rapidly. 

Global humanoid shipments reached 19,100 units in the first half of 2026 alone, nearly quadrupling year-on-year, according to Smart Analytics Global as reported by Reuters. Figure AI (company-claimed) reports scaling its Figure 03 production rate from one robot per day to one robot per hour — a claimed 24× throughput improvement — with more than 350 Figure 03 units delivered as of mid-2026. Faster production ramp is one of the mechanisms most likely to reduce future humanoid costs, which matters directly to the 2028 cost-improvement assumption.

Signal 3 — The price landscape is fragmenting, not converging on a single curve. 

Unitree lists its G1 from $13,500 and R1 from $4,900 (both company-claimed, manufacturer-listed prices). Industrial-grade platforms from Figure AI, Agility Robotics, and Apptronik carry no published industrial purchase price. Reuters reported in August 2026 that Chinese industrial humanoids commonly trade at around RMB 300,000–500,000 in the commercial market. These figures are not comparable. A buyer treating Unitree's listed price as evidence that all humanoids are approaching consumer-electronics pricing is making a material analytical error.

Signal 4 — Production-scale adoption remains constrained through 2028. 

Gartner forecasts that fewer than 20 companies will reach production-stage humanoid deployments in manufacturing and supply chain by 2028, and fewer than 100 companies will move beyond proof-of-concept.

Abdil Tunca, Senior Principal Analyst at Gartner, stated in January 2026:

“The promise of humanoid robots is compelling, but the reality is that the technology remains immature and far from meeting expectations for versatility and cost-effectiveness.”

That is the analyst consensus baseline against which the deploy-or-wait decision must be made.

Signal 5 — The economic question is shifting from capability to cost-of-deployment. 

The relevant question in 2026 is no longer whether a humanoid robot can perform an industrial task. Some clearly can. The question is whether the total cost of deploying one today — hardware, software, integration, training, maintenance, downtime, supervision, workflow redesign, support, fleet management, and upgrade risk — produces a better economic outcome than capturing expected 2028 improvements at the cost of two years of forgone automation value. That is the decision this report is structured to answer.

Signal	Evidence	Buyer implication
Deployment maturity	BMW Figure 02 — 10 months, 30,000+ vehicles	2026 pilots are credible entry points
Market scale constraint	Gartner: fewer than 20 production-stage companies by 2028	Scale remains limited; early movers face real risk
Price fragmentation	Unitree vs industrial platforms — not comparable	Headline price is not deployment economics
Supply growth	19,100 H1 2026 shipments	Learning curve may accelerate cost reduction
Evidence gap	Commercial evidence exists; public ROI does not	Capital decisions require buyer-specific modelling

---

2. Platform / Market Landscape

The industrial humanoid robot market in 2026 consists of a small number of commercially active enterprise-supported platforms and a larger number of lower-cost hardware entrants that are not comparable on a price or capability basis. The key distinction for procurement buyers is that public price is not industrial purchase price, and industrial purchase price is not TCO. No enterprise-supported industrial humanoid platform — Figure 03, Digit, Apollo — has published a definitive industrial unit price. Unitree's listed hardware prices reflect a different product category and should not be used as proxies for industrial deployment economics.

The 2026 commercial humanoid market divides into three distinct tiers: enterprise-supported industrial platforms with verified deployment activity, enterprise-supported platforms with confirmed agreements but limited public outcome data, and hardware-accessible platforms positioned for research and development with published consumer-facing prices.

Figure AI has the most publicly documented industrial deployment record in the current market. Figure 02 operated at BMW Group's Spartanburg plant for ten months, contributing to the production of more than 30,000 BMW X3 vehicles. BMW subsequently introduced Figure 03 into a new sequencing application at Spartanburg and announced a Figure 03 pilot at its Leipzig facility. Figure AI (company-claimed) reports the Figure 03 stands 5'8" tall, weighs 61 kg, carries a payload of 20 kg, and operates for 5 hours per charge at 1.2 m/s. No public industrial purchase price has been identified.

Agility Robotics' Digit is the most commercially active logistics-sector humanoid by number of verified agreements. GXO Logistics signed a multi-year RaaS agreement for Digit at the SPANX fulfillment facility in 2024. Toyota Motor Manufacturing Canada (TMMC) signed a commercial agreement for Digit in February 2026 following a pilot that Agility described as successful. Peggy Johnson, CEO of Agility Robotics, stated in August 2026 to The Wall Street Journal that humanoid robotics “have actually moved into real paid deployments,” citing Amazon, Toyota, Schaeffler, GXO, and Mercado Libre as customers (company-claimed). Digit's published specifications include a carrying capacity of 35 lb and a battery life of 4 hours, both company-claimed. No definitive public industrial purchase price has been identified.

Apptronik's Apollo is being tested at Mercedes-Benz's Berlin-Marienfelde Digital Factory Campus in production-support applications. Apptronik raised $520 million in a Series A extension in February 2026, backed by Google and Mercedes-Benz, at a reported valuation of approximately $5 billion. The company launched Apollo 2 and a robot-training facility developed with Google DeepMind in June 2026. Apollo's published specifications include a payload of 55 lb and a swappable 4-hour battery, both company-claimed. No public industrial purchase price has been identified.

Unitree's G1 and R1 represent a materially different product category. Unitree lists the G1 from $13,500 and the R1 from $4,900, with an R1 variant at $5,900 (all company-claimed, manufacturer-listed prices, excluding tax and shipping). These platforms are positioned for research, development, and general-purpose humanoid applications. Unitree raised approximately $900 million through its Shanghai STAR Market IPO in July 2026. Unitree's hardware prices cannot be treated as equivalent to the fully integrated, enterprise-supported deployment cost of Figure, Digit, or Apollo. The products differ materially in autonomy, application support, software stack, integration tooling, intended deployment environment, and operational context.

Humanoid — the Chinese manufacturer — announced a planned deployment of 1,000–2,000 robots across Schaeffler's global manufacturing sites by 2032, with an initial rollout planned for December 2026 – June 2027 at two German locations (company-claimed for unit volumes; VERIFIED for the existence of the commercial agreement per Reuters, May 2026). Contract value and precise unit numbers were not disclosed.

The pricing architecture buyers must internalise: public price ≠ industrial purchase price ≠ TCO. No credible apples-to-apples industrial humanoid price series covering 2024–2026–2028 currently exists in the public domain. Any report claiming to show a clean industrial humanoid price-decline curve should be treated with significant scepticism.

Platform	Manufacturer	Target sector	Public price	Deployment evidence	Price class.
Figure 03	Figure AI	Automotive mfg.	Not published	BMW (VERIFIED)	N/A
Digit	Agility Robotics	Logistics / mfg.	Not published	GXO, Toyota (VERIFIED)	N/A
Apollo	Apptronik	Mfg. / logistics	Not published	Mercedes-Benz trial (VERIFIED)	N/A
G1	Unitree	Research / dev.	From $13,500	Not established	COMPANY-CLAIMED
R1	Unitree	General purpose	From $4,900	Not established	COMPANY-CLAIMED

---

3. Deployment Evidence: What Has Actually Happened?

Operational evidence that humanoid robots can perform bounded industrial tasks in commercial settings is increasing in 2026. BMW Group has the clearest publicly documented production deployment, with Figure 02 contributing to the output of more than 30,000 BMW X3 vehicles over ten months. GXO Logistics and Toyota Motor Manufacturing Canada hold verified commercial agreements with Agility Robotics. However, verified public ROI data — covering acquisition cost, integration, labour saving, downtime, and payback — does not exist for any named humanoid deployment as of August 2026. Operational evidence and financial evidence are at different points on the maturity curve.

Five verified humanoid deployments are documented in the public record as of August 2026 — and none has publicly disclosed a complete ROI calculation covering acquisition cost, integration, labour saving, and payback. The distinction matters because a capital decision based on deployment announcements — rather than economically relevant outcomes — carries material risk.

The evidence hierarchy for the current market runs from announcement to verified economic outcome across five levels:

● Level 1 — Announcement: press release stating intent to deploy or evaluate

● Level 2 — Pilot: confirmed operational trial in a real facility

● Level 3 — Production deployment: robot operating in a live production environment

● Level 4 — Measured operational outcome: verified production contribution, throughput, or operating hours

● Level 5 — Verified economic outcome: independently confirmed ROI covering full TCO and labour savings

The majority of publicly available humanoid evidence sits at levels 2–4. Level 5 evidence — verified economic outcome — does not exist in the public domain for any humanoid deployment covered by this report.

BMW × Figure AI reaches level 4: the most concrete publicly available evidence in the current market. Figure 02 ran for ten months at Spartanburg, operating on 10-hour shifts Monday through Friday (company-claimed work schedule), accumulating more than 1.2 million robot steps (company-claimed) and contributing to the production of more than 30,000 BMW X3 vehicles. BMW Group's move from Figure 02 to Figure 03 within the same facility signals operational continuity rather than evaluation: the programme expanded because the first deployment generated sufficient value to extend — a stronger signal than any deployment announcement. What BMW has not publicly disclosed is the acquisition cost, integration investment, labour hours replaced, or payback period. The deployment is VERIFIED; the ROI is not publicly established.

GXO × Agility Robotics reaches level 3–4. A verified commercial multi-year RaaS agreement was signed in June 2024. Agility subsequently reported that Digit has moved more than 100,000 totes in commercial deployment (company-claimed throughput metric). The RaaS structure meant GXO avoided upfront capital expenditure, transferring technology obsolescence risk to Agility — a rational response to platform uncertainty. No independently verified productivity rate, ROI, or payback metric has been identified.

Mercedes-Benz × Apptronik reaches level 2. The deployment at Berlin-Marienfelde is a VERIFIED trial in a live industrial environment. No quantified operational outcome — no throughput figure, no labour-hour reduction, no cost-per-task comparison — has been publicly disclosed.

Toyota Motor Manufacturing Canada × Agility reaches level 3. The commercial agreement is VERIFIED. The “successful pilot” description originates from Agility Robotics (company-claimed); no independent productivity or financial metric for the pilot has been identified in public sources.

The buyer-relevant conclusion from this evidence map: the case for deploying in 2026 rests on operational plausibility — the BMW deployment demonstrates that a humanoid can perform a bounded industrial task over sustained periods — but not on publicly verified economics. Any organisation making a capital commitment in 2026 is doing so on the basis of task-level logic and its own pilot data, not on publicly benchmarked ROI comparables.

---

4. Economics & ROI Analysis — The Cost-of-Delay Decision

Waiting until 2028 becomes financially superior to deploying a humanoid robot in 2026 only when the combined reduction in TCO and improvement in platform capability by 2028 exceeds the economic value that would be generated by two years of earlier automation. That value includes forgone labour savings, forgone productivity gain, forgone capacity improvement, lost deployment learning, and delayed competitive positioning. Where the cost of delay is high — measured by a persistent labour constraint, high overtime exposure, or a task that conventional automation cannot address — the economic case for deploying in 2026 is stronger. Where the cost of delay is low, waiting rationally captures expected 2028 cost reductions and capability improvements at minimal economic sacrifice.

4.1  The Wrong Question: Will Humanoids Be Cheaper?

A lower robot purchase price in 2028 does not automatically produce a better economic outcome for a buyer who waits. The cost reduction must be large enough to offset what the buyer would have earned by deploying earlier — and the robot’s purchase price is only one input into deployment economics. Whether humanoid robots will be cheaper by 2028 is a necessary question, but it is insufficient on its own.

TCO for an industrial humanoid deployment includes: hardware acquisition or lease cost; software licensing and updates; integration with existing systems and workflows; facility and infrastructure adaptation; operator training and supervision; planned and unplanned maintenance; downtime and productivity loss; fleet management overhead; upgrade and obsolescence risk; and support and service contract.

A scenario in which robot hardware prices fall by 30% by 2028 but integration costs, software maturity, and utilisation rates remain similar to 2026 produces a much smaller net TCO improvement than a 30% headline figure implies. The deploy-or-wait decision requires modelling the full TCO picture, not the hardware price alone.

4.2  The 2026 Deploy Case

The economic value of deploying in 2026 is the sum of benefits generated over the 2026–2028 period, net of full deployment costs.

2026 deployment value = Labour savings generated 2026–2028 + Productivity gain from automation + Capacity value captured + Avoided overtime and shift premium costs + Deployment learning (operational data, process knowledge, integration experience) + Strategic option value (first-mover positioning, vendor relationship depth)

Minus: Robot TCO (hardware, software, lease or acquisition) + Integration and facility adaptation + Training and supervision + Maintenance and downtime + Deployment management overhead + Technology obsolescence risk

The 2026 case is strongest where the labour or capacity constraint is measurable and current, the task is repetitive and bounded with high utilisation potential, conventional automation is less suitable or has been evaluated and rejected, and the cost of delay — the economic loss from continuing with current arrangements — is quantifiable.

4.3  The 2028 Wait Case

The economic value of waiting until 2028 is the expected improvement in deployment economics, net of the value forgone during the two-year wait.

2028 wait value = Lower expected TCO (hardware price reduction + software maturity improvement) + Higher expected platform capability (better autonomy, task range, utilisation) + Greater vendor choice and market competition + Lower technology risk + More mature integration tooling and data infrastructure

Minus: Two years of forgone labour savings + Two years of forgone productivity gain + Two years of forgone capacity value + Lost deployment learning + Continued exposure to the labour constraint + Potential competitive disadvantage if peers deploy first

The 2028 case is strongest where the cost of delay is low, current humanoid TCO is economically marginal for the specific task, conventional automation can address the need in the interim, and the organisation lacks readiness to deploy effectively in 2026.

4.4  The Humanoid Wait-or-Deploy Threshold

Waiting wins economically only when the combined 2028 TCO reduction and capability improvement exceed the value that two years of earlier deployment would have generated — the calculation the Humanoid Wait-or-Deploy Threshold quantifies. The threshold is defined as: the minimum required reduction in total humanoid deployment cost, combined with the minimum required improvement in platform capability and utilisation, that must materialise by 2028 before waiting becomes economically superior to deploying in 2026.

Required 2028 TCO improvement = Value of two years of forgone benefits ÷ 2026 deployment cost base

Editorial Estimate — illustrative inputs below; substitute buyer-specific figures for actual analysis.

Applied to an illustrative warehouse scenario: assume an annual labour cost for the automated task of $120,000, a 2026 humanoid deployment TCO (all-in, including integration, software, support, and maintenance) of $200,000 over two years, a productivity premium of 15% over baseline labour, and a deployment learning value estimated at $20,000 over the period.

The 2026 deploy net value over the two-year window:

($120,000 × 2) + ($18,000 productivity) + ($20,000 learning) − $200,000 TCO = $78,000.

For waiting to be economically superior, the 2028 deployment must offer a TCO low enough to generate a present value exceeding $78,000 — requiring the 2028 TCO reduction to exceed roughly 39% of the 2026 baseline in this illustrative scenario.

This is not a universal benchmark. Every buyer’s cost of delay, labour economics, task utilisation, and TCO baseline is different. The framework is a buyer-configurable model, not an industry forecast.

4.5  Scenario Analysis

No authoritative, apples-to-apples industrial humanoid price series covering 2024–2028 exists in the public domain. The report therefore models price decline and capability improvement as three analytical scenarios rather than presenting a single 2028 price forecast.

Editorial Estimate — the percentage ranges below are analytical scenarios derived from the cost model, not sourced price forecasts.

Scenario A — Conservative (Outcome: Wait)

Expected 2028 TCO reduction: modest (10–20% of 2026 baseline). Capability improvement: limited incremental gain. Integration costs remain high. Utilisation constrained by autonomy limitations. Cost of delay: low.

Scenario B — Base Case (Outcome: Pilot/Staged Deployment in 2026–27)

Expected 2028 TCO reduction: moderate (25–40% of 2026 baseline). Capability improvement: meaningful improvement in autonomy and task range. Integration tooling maturing. Cost of delay: moderate. The rational path is a 2026 or 2027 pilot that generates deployment learning without committing full capital.

Scenario C — Accelerated (Outcome: Deploy in 2026)

Expected 2028 TCO reduction: substantial (40% or more of 2026 baseline). Capability improvement: significant, including materially higher autonomy and task coverage. Cost of delay: high, driven by acute labour constraint or competitive pressure. Deploying in 2026 generates two years of automation value, deployment learning, and competitive positioning that more than offset the premium paid for earlier technology.

Variable	Deploy 2026	Wait 2028
Robot TCO	Higher (current market pricing)	Lower expected (scenario-dependent)
Labour savings	Captured from 2026	Delayed two years
Capacity value	Captured earlier	Forgone during wait
Deployment learning	Accumulated from 2026	Delayed
Technology risk	Higher (earlier platform)	Lower (later adoption)
Vendor choice	Current market	Potentially broader
Platform capability	Current	Expected higher
Obsolescence risk	Higher	Lower
Integration readiness required	Immediate	Can be built during wait

---

5. Named Deployment Case Studies

The humanoid robot deployments offering the strongest evidence base for industrial buyers in 2026 are BMW × Figure AI at Spartanburg — the clearest publicly documented production deployment — and GXO × Agility Robotics at the SPANX facility in Georgia, which provides verified commercial agreement evidence under a RaaS structure. Mercedes-Benz × Apptronik at Berlin-Marienfelde confirms that a major automotive manufacturer is trialling the technology in a live production environment. Toyota Motor Manufacturing Canada × Agility confirms that automotive buyers are transitioning from pilots to commercial agreements. No deployment covered here has publicly disclosed a verified ROI calculation.

Case Study 1: BMW Group × Figure AI — Spartanburg, South Carolina

Platform: Figure 02 → Figure 03  |  Sector: Automotive manufacturing  |  Evidence level: Measured operational outcome (Level 4)

BMW Group's deployment of Figure 02 at its Spartanburg assembly plant is the most substantiated publicly documented humanoid industrial deployment currently available. The robot handled sheet-metal components for welding and operated over ten months on 10-hour shifts Monday through Friday (company-claimed work schedule). The deployment contributed to the production of more than 30,000 BMW X3 vehicles — a figure sourced from BMW Group's own disclosures and classified as VERIFIED for the production contribution.

BMW's move to Figure 03 in a new sequencing application at Spartanburg, and a subsequent pilot at Leipzig, signals operational continuity: the programme expanded because the first deployment generated sufficient value to extend — stronger evidence than any deployment announcement.

Buyer lesson: A repeat deployment is stronger evidence than a single announcement. However, the absence of a publicly disclosed ROI — covering acquisition cost, integration, labour hours replaced, downtime, and payback — means no buyer can benchmark their own economics against BMW's experience. The production contribution is VERIFIED; the financial return is not established in public sources.

Case Study 2: GXO Logistics × Agility Robotics — SPANX Facility, Georgia

Platform: Digit  |  Sector: Logistics and fulfillment  |  Evidence level: Production deployment with company-claimed throughput (Level 3–4)

GXO Logistics signed an industry-first multi-year RaaS agreement with Agility Robotics in June 2024, covering Digit deployment at the SPANX fulfillment operation in Georgia. Agility subsequently reported that Digit has moved more than 100,000 totes in commercial deployment (company-claimed throughput metric — not independently audited).

The RaaS structure is commercially relevant to the deploy-or-wait decision. By contracting under a service model, GXO transferred technology obsolescence risk and capital exposure to Agility. A buyer who might hesitate to commit capital to a 2026 robot purchase may find the economics more tractable under a service contract — provided the per-task or per-unit economics are modelled correctly against the full service cost.

Buyer lesson: The RaaS model changes the capital structure of the deploy-or-wait decision. The fact that GXO chose a service contract rather than a capital purchase is itself a signal about how an informed logistics buyer assessed platform risk at the time of the agreement.

Case Study 3: Mercedes-Benz × Apptronik — Berlin-Marienfelde, Germany

Platform: Apollo  |  Sector: Automotive manufacturing  |  Evidence level: Verified trial (Level 2)

Mercedes-Benz confirmed in March 2025 that it was testing Apptronik's Apollo at its Berlin-Marienfelde Digital Factory Campus, focusing on production-support applications including component movement and assembly-adjacent tasks. Apptronik subsequently raised $520 million — with Mercedes-Benz participating as an investor — and launched Apollo 2 in June 2026. The public record contains no quantified labour-hour reduction, throughput rate, cost-per-task figure, or payback period for the Berlin-Marienfelde trial.

Buyer lesson: A named industrial trial by a major manufacturer establishes commercial relevance and technical plausibility. It does not establish economic viability. When an automotive manufacturer also becomes an investor in the platform provider, a buyer should note that the relationship carries commercial interests beyond a standard operator-vendor evaluation.

Case Study 4: Toyota Motor Manufacturing Canada × Agility Robotics — Canada

Platform: Digit  |  Sector: Automotive manufacturing, supply chain, and logistics  |  Evidence level: Verified commercial agreement; company-claimed pilot success (Level 3)

Agility Robotics announced in February 2026 that Toyota Motor Manufacturing Canada had signed a commercial agreement for Digit deployment in manufacturing, supply-chain, and logistics operations, following what Agility described as a successful pilot (company-claimed). No independent assessment of the pilot outcome, productivity metrics, or financial performance has been identified in public sources.

Buyer lesson: The transition from a pilot to a commercial agreement is evidence of commercial intent, not of verified ROI. Buyers conducting their own evaluations should require task-level performance data — not deployment-level announcements — before scaling a pilot to a commercial agreement.

Operator	Platform	Location	Stage	ROI disclosed?	Evidence
BMW Group	Figure 02/03	Spartanburg, SC	Production deployment + expansion	No	VERIFIED
GXO Logistics	Digit	Georgia, USA	Commercial RaaS agreement	No	VERIFIED agreement / COMPANY-CLAIMED throughput
Mercedes-Benz	Apollo	Berlin-Marienfelde	Trial	No	VERIFIED trial
Toyota Mfg. Canada	Digit	Canada	Commercial agreement	No	VERIFIED agreement / COMPANY-CLAIMED pilot success

---

6. Friction, Risk & Unresolved Issues

The largest risks associated with deploying a humanoid robot in 2026 are economic, not mechanical. Gartner forecasts fewer than 20 companies will reach production-stage humanoid deployments by 2028, reflecting widespread immaturity in versatility and cost-effectiveness. The specific risks buyers must quantify before committing capital are: utilisation shortfall, integration cost overruns, vendor concentration risk, and the possibility that waiting until 2028 captures a significantly better software and data layer at lower risk. The risk of deploying too early and the risk of waiting too long are not symmetrical — they depend on the buyer’s cost of delay.

Risk 1: Economic immaturity is the primary constraint. 

Gartner's January 2026 forecast — fewer than 20 production-stage humanoid deployments in manufacturing and supply chain by 2028, and fewer than 100 companies progressing beyond proof-of-concept — reflects a structural assessment of where the market actually sits. Abdil Tunca of Gartner characterised the technology as “immature and far from meeting expectations for versatility and cost-effectiveness.” For a procurement buyer, the implication is that they are likely to be among the earliest adopters in their sector, with all the associated learning cost, integration friction, and vendor dependence that early adoption implies.

Risk 2: Units shipped are not productive deployments. 

Reuters reported in August 2026 that despite 19,100 global humanoid shipments in H1 2026, many of those units remain connected to training and data collection rather than productive commercial work. The industrial pricing reality — approximately RMB 300,000–500,000 per unit in the Chinese market per Reuters reporting — further constrains broad commercial deployment. Shipment growth is not equivalent evidence to deployment productivity.

Risk 3: Utilisation shortfall destroys the ROI case faster than any other variable. 

A humanoid robot achieving 60% utilisation in a task generates materially different economics from the same robot at 90% utilisation. Autonomy limitations, task-switching friction, supervision requirements, and downtime accumulate in ways that static ROI models typically understate. Buyers should model conservative utilisation rates — and then stress-test the economics at 40% and 50% — before committing capital.

Risk 4: Integration cost is a systematic underestimation risk. 

Digit's Arc platform is designed to interface with warehouse management and execution systems — but the integration of any humanoid into an existing operational environment involves facility adaptation, workflow redesign, operator training, and software configuration that does not appear in a hardware price. No publicly available benchmark for industrial humanoid integration cost as a percentage of hardware cost was identified in the research for this report. Buyers should budget integration as a material additional cost and verify assumptions with an independent integrator.

Risk 5: Vendor concentration risk is real and under-discussed. 

The current industrial humanoid commercial market is served by a small number of vendors, several of whom raised large capital rounds in 2025–2026 but have not yet demonstrated production-scale operational resilience. A buyer who deploys at scale with a single vendor and that vendor changes its commercial strategy, pricing model, or product roadmap materially faces a support and upgrade risk with no obvious near-term mitigation.

Risk 6: The software and data layer may improve substantially by 2028. 

Apptronik launched a robot-training facility with Google DeepMind in June 2026 specifically to accelerate the move from pilot to production by collecting real-world operational data at scale. This is evidence that the data and software infrastructure underpinning humanoid autonomy is still being built. A buyer who waits until 2028 may access a materially more capable software layer — one that reduces supervision requirements and extends task range — without paying the early-adopter learning cost.

Risk 7: Autonomy demonstrations are not evidence of productive industrial deployment.  Reuters reported in April 2026 that nearly 40% of robots in China’s humanoid half-marathon were navigating autonomously, compared to remotely controlled operation the previous year. Mobility and coordination improvements are real. They should not be translated directly into evidence of economic readiness for high-repetition industrial tasks at production scale. The research for this report found no basis for concluding that humanoid deployments have systematically experienced cost overruns or schedule failures.

Risk 8: The deploy-or-wait risk is not symmetrical. 

For a buyer with a high cost of delay — an acute labour shortage in a bounded task, high overtime exposure, a task that cannot be addressed by conventional automation — the risk of waiting is higher than the risk of early deployment. For a buyer with a low cost of delay, the asymmetry reverses. The risk analysis cannot be conducted in the abstract; it requires a buyer-specific cost-of-delay calculation.

Risk	Deploy 2026	Wait 2028	Severity
Utilisation shortfall	High	Lower (better software layer)	High
Integration cost overrun	High	Moderate (better tooling)	High
Vendor concentration	High	Lower (more vendors by 2028)	Medium
Technology obsolescence	Higher	Lower	Medium
Cost of delay	Avoided	Incurred	Context-dependent
Software immaturity	Higher	Lower	Medium
Capital commitment risk	Higher	Lower	Context-dependent

---

7. Competitive / Platform Comparison

No single humanoid robot platform is the correct choice for all 2026 industrial deployments. Platform selection for a deploy-or-wait decision should focus on five dimensions: deployment evidence depth, support model maturity, commercial pricing structure, integration readiness, and 2028 capability trajectory. Figure 03 has the deepest publicly documented automotive manufacturing evidence. Digit has the most commercial agreements across logistics and automotive. Apollo is in active industrial trials with significant capital backing. Unitree platforms are not comparable to enterprise-supported industrial deployments and should not be used as benchmarks in industrial procurement economics.

Four platforms have commercially relevant evidence for a 2026 industrial deployment decision: Figure 03, Digit, Apollo, and Unitree G1/R1, where the last occupies a distinct, non-comparable product category. This section compares commercially relevant dimensions; it makes no assessment of technical superiority, reliability, or platform safety.

Figure 03 (Figure AI)

Industrial target: automotive manufacturing and assembly-adjacent tasks. Commercial evidence: VERIFIED deployment at BMW Spartanburg (two generations) and a new Leipzig pilot. Production status: in commercial production (company-claimed: more than 350 units delivered, production rate 1/hour). Specifications — height 5'8", weight 61 kg, payload 20 kg, runtime 5 hours, speed 1.2 m/s — all company-claimed. No public industrial purchase price. Evidence grade: B (customer deployment evidence, production contribution verified by customer).

Digit (Agility Robotics)

Industrial target: logistics, fulfillment, and automotive supply chain. Commercial evidence: VERIFIED agreements with GXO, Toyota Motor Manufacturing Canada, and others cited by Agility Robotics CEO. Carrying capacity 35 lb, battery life 4 hours — both company-claimed. RaaS commercial structure available, reducing capital commitment. No public industrial purchase price. Evidence grade: B (multiple verified commercial agreements; throughput metrics company-claimed).

Apollo (Apptronik)

Industrial target: manufacturing and logistics. Commercial evidence: VERIFIED trial at Mercedes-Benz Berlin-Marienfelde; $520 million capital backing from Google and Mercedes-Benz (February 2026). Payload 55 lb, swappable 4-hour battery — company-claimed. Apollo 2 launched June 2026 alongside a Google DeepMind-backed training facility. No public industrial purchase price. Evidence grade: B/C (verified trial; no quantified outcome data disclosed).

Unitree G1 / R1

Industrial target: research, development, and general-purpose humanoid applications. Listed prices: G1 from $13,500, R1 from $4,900–$5,900 (all company-claimed, manufacturer-listed). Unitree raised approximately $900 million via Shanghai STAR Market IPO in July 2026. No verified industrial production deployment identified. Evidence grade: D (product available; no comparable industrial deployment evidence). Unitree's listed hardware prices reflect a product category not comparable to enterprise-supported industrial platforms on application, autonomy, integration support, or operational context.

Platform	Evidence grade	Commercial structure	Key 2026 evidence	2028 trajectory signal
Figure 03	B	Purchase	BMW VERIFIED deployment	Production scale increasing
Digit	B	Purchase / RaaS	GXO, Toyota VERIFIED	Training data investment
Apollo	B/C	Purchase	Mercedes-Benz trial VERIFIED	Apollo 2 + DeepMind facility
Unitree G1/R1	D	Purchase	No industrial deployment identified	Capital from IPO for mfg. scale

Evidence grades: A = independently/externally verified; B = customer/company deployment evidence; C = company claim; D = projection or no comparable evidence

---

8. Strategic Recommendations for Automation/Procurement Buyers

An industrial company should deploy a humanoid robot in 2026 — rather than waiting until 2028 — when three conditions are simultaneously present: a measurable and current labour or capacity constraint that cannot be addressed by conventional automation, a task that is repetitive, bounded, and capable of generating high utilisation, and organisational readiness to manage the integration, training, and operational demands of an early-stage platform. Where these conditions are absent or uncertain, a staged pilot in 2026–27 or a disciplined wait until 2028 is the more defensible capital decision.

The research, economic model, and evidence analysis in this report produce three actionable paths. Buyers should select based on their specific cost-of-delay calculation, not on market sentiment or vendor-provided timelines.

Deploy in 2026 when:

● A labour or capacity constraint is measurable, current, and generating a quantifiable cost — not projected or hypothetical

● The target task is repetitive, physically bounded, and capable of sustaining high robot utilisation (target minimum 70–80% productive uptime) (Editorial guidance — no published industry benchmark available)

● Conventional automation has been evaluated and rejected as less suitable for the specific task

● The organisation has integration capability — in-house or contracted — to deploy without excessive friction

● Two years of forgone automation value represents a meaningful economic cost

● Technology risk is manageable given the organisation’s scale, financial resilience, and tolerance for early-adopter learning

Pilot in 2026–27 when:

● The economic case for humanoid automation is directionally positive but not yet quantified with confidence

● The task can generate deployment data and operational learning during a contained pilot

● The organisation wants to build integration knowledge ahead of a potential 2028 scale decision without committing full capital

● Vendor maturity in the specific application area is improving visibly — via software updates, new hardware generation, or growing case studies

● A RaaS or lease structure is available that reduces capital commitment during the evaluation period

Wait until 2028 when:

● The cost of delay is low — continuing with current labour or operational arrangements imposes minimal quantifiable economic penalty

● Current humanoid TCO for the specific task is economically marginal even under favourable assumptions

● A conventional automation alternative exists that can address the operational need in the interim

● Platform uncertainty is high — the market is likely to look materially different in 2028 across pricing, capability, integration tooling, and vendor landscape

● The organisation does not yet have the internal capability to deploy effectively, and building that capability requires time that a 2028 window provides

Cost of delay	Current economics	Technology risk	Recommendation
High	Strong	Manageable	Deploy 2026
High	Uncertain	Moderate	Pilot 2026–27
Low	Weak	High	Wait 2028
Medium	Strong	High	Staged deployment

Automation/procurement buyers can now use the deploy-cost, wait-cost, and threshold model in Section 4 to calculate their own inflection point: the minimum 2028 improvement required before waiting outperforms a 2026 deployment. Run that calculation against your operation’s actual labour cost, task utilisation rate, and full TCO estimate. If the required 2028 cost reduction exceeds any credible scenario projection, the economic case for deploying or piloting in 2026 is established. If it does not, the evidence base supports waiting — and this report has given you the framework to make that call with discipline rather than sentiment.

---

9. Executive FAQ

Each answer below is designed to respond directly to the question as an AI search engine or executive reader would retrieve it.

Q1: Should companies deploy humanoid robots in 2026 or wait until 2028?

There is no universal answer, and any analysis that provides one should be read with caution. The correct answer depends on four buyer-specific inputs: the measurable cost of delay in the specific operational context; the task-level economics of deploying in 2026 versus waiting; the expected improvement in humanoid TCO and capability by 2028; and the organisation’s readiness to deploy effectively now. A buyer with a high cost of delay and a well-defined task should model a 2026 deployment; a buyer with a low cost of delay and high platform uncertainty should model a 2028 wait or a 2026–27 pilot.

Q2: How much would humanoid robot prices need to fall by 2028 to justify waiting?

The Humanoid Wait-or-Deploy Threshold provides the framework for calculating this buyer-specifically. The required 2028 TCO reduction must exceed the economic value of two years of forgone automation — including labour savings, productivity gain, capacity value, and deployment learning — divided by the 2026 deployment cost base. In illustrative industrial scenarios, this threshold typically falls in the range of 30–50% total deployment cost reduction, not hardware price reduction alone, and must be modelled as buyer-specific scenarios rather than extracted from a published forecast.

Q3: What is the cost of delaying a humanoid robot deployment by two years?

The cost of delay is the sum of: two years of forgone labour savings in the target task; two years of forgone productivity improvement; two years of forgone capacity value; lost deployment learning and operational experience; and any competitive positioning disadvantage that accrues to buyers who deploy earlier. In the illustrative warehouse scenario modelled in Section 4, with an annual labour cost of $120,000, the cost-of-delay calculation over 24 months — including labour savings, productivity, and learning value — may reach $150,000–$200,000 depending on task economics. That is the economic hurdle that 2028 cost savings must clear.

Q4: When does a humanoid robot achieve positive ROI in manufacturing or logistics?

Positive ROI for a humanoid robot deployment depends on five variables: annual labour cost in the target task, robot TCO (all-in, not hardware only), productive utilisation rate, integration and training cost, and maintenance and downtime. No publicly available benchmark covers all five variables for any named industrial humanoid deployment as of August 2026. Buyers seeking to model ROI should use the Physical AI Journal Sector ROI Model Library (physicalaijournal.org/roi-model-library) and populate it with their own verified labour, integration, and utilisation figures.

Q5: Which humanoid robots have the strongest commercial deployment evidence in 2026?

Ranked by publicly documented operational depth: Figure AI (Figure 02/03 at BMW Spartanburg) provides the clearest production-contribution evidence — more than 30,000 BMW X3 vehicles contributed over ten months. Agility Robotics (Digit at GXO and Toyota Motor Manufacturing Canada) has the most verified commercial agreements across logistics and automotive. Apptronik (Apollo at Mercedes-Benz Berlin-Marienfelde) has a confirmed industrial trial. No deployment has publicly disclosed a verified ROI calculation.

Q6: Is it better to pilot a humanoid robot now or wait for cheaper, more capable robots in 2028?

A 2026–27 pilot occupies the optimal middle position for buyers who cannot yet confirm the three deployment conditions — measurable cost of delay, strong task economics, and organisational readiness — but want to build deployment knowledge and reduce integration risk ahead of a 2028 scale decision. A pilot under RaaS or a contained purchase minimises capital commitment while generating the operational data needed to make a credible 2028 decision. The worst outcome is a buyer who neither deploys nor pilots and arrives at 2028 with no operational experience.

---

Scope & Disclaimer

This report is an independent economic decision-intelligence analysis produced by Physical AI Journal Research Team at physicalaijournal.org, operated by Sekason Research Limited, London, United Kingdom (Company No. 14339910).

This report does not constitute financial, investment, legal, or engineering advice of any kind. It does not certify the safety, technical conformity, fitness for purpose, or regulatory compliance of any humanoid robot platform or deployment. It does not rank platforms as technically superior or more reliable. It does not predict a definitive 2028 industrial humanoid robot price. It does not guarantee the commercial performance, operational productivity, or return on investment of any humanoid robot deployment. All company-claimed specifications, prices, throughput metrics, and production targets are labelled as such and should not be treated as independently verified facts. Company-claimed figures should be verified directly with the relevant manufacturer or vendor before being used as the basis for a capital decision.

Full disclaimer: physicalaijournal.org/disclaimer

---

References & Strategic Sources:

• Reuters — Humanoid startup Apptronik raises $520 million with backing from Google and Mercedes-Benz — 11 Feb 2026 — Source URL — Verified — Tier 1

• Reuters — Apptronik raises $350 million to scale production of humanoid robots — 13 Feb 2025 — Source URL — Verified — Tier 1

• Reuters — Humanoid to deploy up to 2,000 robots at Schaeffler plants — 13 May 2026 — Source URL — Verified — Tier 1

• Reuters — Beyond marathons and backflips, China's robots face a commercial test — 18 Aug 2026 — Source URL — Verified — Tier 1

• Reuters — China robot makers flock to Beijing show, seek path to mass adoption — 19 Aug 2026 — Source URL — Verified — Tier 1

• Reuters — China humanoid robot half-marathon to showcase technical leaps — 18 Apr 2026 — Source URL — Verified — Tier 1

• Wall Street Journal — Peggy Johnson (CEO, Agility Robotics) interview on humanoid deployments — Aug 2026 — URL not confirmed in research data — verify separately at wsj.com before publication — Verified quote — Tier 1 (URL unconfirmed)

• Gartner — Humanoid Robots Will Stall at Pilot Scale: Gartner predicts fewer than 20 companies will scale humanoid robots to production-stage by 2028 — 21 Jan 2026 — Source URL — Verified — Tier 2

• BMW Group — BMW Group advances the use of Physical AI in production with Figure 03 project in Spartanburg — 25 Jun 2026 — Source URL — Verified — Tier 2

• BMW Group — BMW Group to deploy humanoid robots in production in Germany — 27 Feb 2026 — Source URL — Verified — Tier 2

• Figure AI — F.02 Contributed to the Production of 30,000 Cars at BMW — 19 Nov 2025 — Source URL — Company-claimed — Tier 2

• Figure AI — Ramping Figure 03 Production — 2026 — Source URL — Company-claimed — Tier 2

• Figure AI — Figure 03 specifications — 2026 — Source URL — Company-claimed — Tier 2

• Mercedes-Benz — AI and humanoid robots at Berlin-Marienfelde Digital Factory Campus — 18 Mar 2025 — Source URL — Verified — Tier 2

• GXO Logistics — GXO signs industry-first multi-year agreement with Agility Robotics — 27 Jun 2024 — Source URL — Verified — Tier 2

• Agility Robotics — Toyota Motor Manufacturing Canada commercial agreement — 19 Feb 2026 — Source URL — Verified — Tier 2

• Agility Robotics — Digit platform and solutions — 2026 — Source URL — Company-claimed — Tier 2

• Apptronik — Apptronik Unveils Apollo — 23 Aug 2023 — Source URL — Company-claimed — Tier 2

• Apptronik — Apptronik Closes Over $935 Million Series A — 11 Feb 2026 — Source URL — Company-claimed — Tier 2

  Note: total cumulative Series A; the report body uses the Reuters-sourced $520 million February 2026 figure.

• Unitree — G1 Humanoid Robot — 2026 — Source URL — Company-claimed — Tier 2

• Unitree — R1 Humanoid Robot — 2026 — Source URL — Company-claimed — Tier 2

• IEEE Spectrum — Humanoid Robots Are Getting to Work — 2024 — Source URL — Verified — Tier 2

Note: This report is backed by authoritative research, independent verification, and structured analytical methodology.

© 2026 Sekason Research Limited. All rights reserved.

Physical AI Journal is a publication of Sekason Research Limited, London, United Kingdom (Company No. 14339910).