The Humanoid Robot Exit Clause: What It Costs to Leave a Platform, and the Contract Architecture That Prevents You from Being Trapped in One
Physical AI Journal | physicalaijournal.org
Physical AI Journal Research Team — Independent analysis by Sekason Research Limited, United Kingdom
Research cutoff: 21 September 2026
Section 01 — Executive Intelligence Synthesis
Direct Answer: The five highest-priority risks in a multi-year humanoid robot contract are: vendor insolvency or acquisition before contract term ends; platform discontinuation mid-contract without successor hardware rights; loss of access to fleet software or AI models if the Robot-as-a-Service (RaaS) relationship terminates; absence of data portability provisions leaving the buyer unable to migrate operational logs or trained workflows; and pricing lock-in at today’s hardware costs while successor platforms become cheaper. All five risks are addressable through contract provisions. None are currently disclosed as standard in any publicly available humanoid commercial agreement reviewed for this report.
The humanoid robot contract has become a procurement-critical instrument. The transition from pilots to multi-year commercial commitments in 2026 is structurally complete: Agility Robotics has disclosed more than $300 million in Digit 5 orders covering 1,000 robots under 3-year RaaS agreements (company-claimed / illustrative per SEC filing), and Humanoid has announced a potential 1,000–2,000 unit planned deployment with Schaeffler extending through 2032 (company-claimed, Reuters reporting of announcement verified). These are no longer exploratory relationships. They are multi-year commercial commitments signed against a vendor landscape that remains financially unsettled, technically accelerating, and contractually opaque.
Signal 1:
The software stack is the true lock-in — not the hardware.
A buyer who replaces one humanoid platform with another does not simply swap metal. The operational value accumulated in a deployment — trained AI models, customised pick-and-place logic, fleet management configuration, warehouse management system (WMS) integration, and accumulated sensor data — resides entirely within the vendor’s software environment. If the RaaS relationship ends, that operational value does not automatically transfer. No publicly disclosed humanoid commercial agreement reviewed for this report contains a defined data portability or software export provision. The hardware is replaceable. The embedded software layer is not — unless the contract explicitly says otherwise.
Signal 2:
Vendor financial structures are changing faster than the contracts signed against them.
Agility Robotics disclosed $138.1 million in net losses against $1.78 million in net revenue for the period covered by its June 2026 SEC merger filing (S-4/A, 24 Jun 2026, VERIFIED). That is not a measure of Agility’s long-term prospects — it is a procurement variable. A buyer signing a 3-year RaaS agreement with a vendor carrying that financial structure must ask what contract provisions survive an insolvency filing, an acquisition, or a strategic pivot. Currently, the answer in most commercial agreements is: unknown.
Signal 3:
RaaS does not eliminate risk — it relocates it from asset depreciation to counterparty dependency.
An outright hardware purchase creates hardware obsolescence risk. A RaaS agreement transfers that risk to the vendor — but creates a new dependency: if the vendor’s service is interrupted, the fleet may go offline. A buyer who has not negotiated source-code escrow, fleet software continuity provisions, or the right to appoint a third-party systems integrator is operationally dependent on the vendor’s continued willingness and financial capacity to deliver the service. That dependency is structurally invisible in the contract documents reviewed.
Signal 4:
The gap between announced order values and operating deployment evidence is systematic.
The International Federation of Robotics (IFR) places 2025 global humanoid sales at approximately ~7,000 units — many of which were purchased for AI training data collection rather than productive factory deployment (Reuters/IFR, 21 Sep 2026, VERIFIED). Bank of America estimates ~20,000 units for 2025 using broader methodology (analyst estimate, Bank of America Global Research, cited in Reuters Aug 2026; methodology not independently verified). These two figures are not reconcilable and must not be averaged. The procurement implication is that headline deployment announcements overstate the number of productively operating humanoid robots in commercial environments. Buyers should model contract provisions against actual vendor operating maturity, not announced order totals.
Signal 5:
Hardware price declines are eroding the economic logic of today’s fixed-cost RaaS structures.
Morgan Stanley estimated an average ~15% hardware cost decline across humanoid platforms in 2026 (Reuters, 27 Aug 2026, VERIFIED). Leju and Kuavo reported approximately ~26% year-on-year price reductions in the same period (Reuters, 27 Aug 2026, VERIFIED). A buyer locked into a $8,500/month illustrative RaaS structure (Agility company-claimed / illustrative per SEC filing) for 3 years faces a compounding gap between their contracted cost and replacement hardware pricing. Without a repricing right or a successor-platform pricing option in the contract, that gap represents pure stranded-asset exposure.
Section 02 — Platform & Market Landscape
Direct Answer: As of September 2026, platforms with confirmed or announced commercial deployments include Agility Digit 5 (RaaS, logistics and automotive), Apptronik Apollo 2 (purchase and RaaS, manufacturing), Figure 02 (automotive manufacturing at BMW Spartanburg, via pilot; Figure 03 production ramp underway for future deployments), and Humanoid HMND 01 (planned Schaeffler deployment from December 2026). IFR places the global humanoid installed base at approximately ~7,000 industrial units as of end-2025 (Reuters/IFR, 21 Sep 2026, VERIFIED). Bank of America estimates ~20,000 units using different methodology (analyst estimate, Bank of America Global Research, cited in Reuters Aug 2026; methodology not independently verified). These figures reflect incompatible counting methodologies and must not be averaged. All platform specifications cited below are company-claimed unless explicitly stated otherwise.
The commercial humanoid market in 2026 presents procurement teams with a fundamental structural problem: eight platforms at varying stages of commercial readiness, operating under four distinct commercial models, each carrying a different contractual exposure profile.
Commercial model determines exit exposure more directly than technical specification. A buyer evaluating Agility Digit 5 under a 3-year RaaS agreement faces counterparty dependency risk that a buyer purchasing Unitree G1 outright does not face — but the outright buyer faces platform obsolescence risk that the RaaS buyer is partially insulated from. The table below maps each currently available commercial platform against the dimensions that govern exit risk.
V1 — Platform Contractual Exposure Matrix
Platform | Commercial Model | RaaS Available | Software Dependency Level | Publicly Disclosed Exit Provisions | Independent Deployment Evidence | Financial Continuity Signal | Classification |
Agility Digit 5 | RaaS / Purchase | Yes | High (fleet management layer) | Not publicly disclosed | VERIFIED (GXO, TMMC agreements) | SEC filing: $138.1M net loss / $1.78M revenue (VERIFIED) | COMPANY-CLAIMED specs; VERIFIED SEC financials |
Apptronik Apollo 2 | RaaS / Purchase | Yes | High (Fleet Connect platform) | Not publicly disclosed | Limited — announced partnerships | $520M raised Feb 2026 (Reuters, VERIFIED) | COMPANY-CLAIMED specs |
Figure 02 | Not publicly disclosed | Not publicly disclosed | High (proprietary OS) | Not publicly disclosed | BMW Spartanburg pilot — BMW states Figure 02 supported production of 30,000+ X3 vehicles over 10 months (company-claimed outcome) | Not publicly disclosed | COMPANY-CLAIMED |
Figure 03 | Not publicly disclosed | Not publicly disclosed | High (proprietary OS) | Not publicly disclosed | No confirmed external customer deployment at research cutoff; production ramp underway (Figure AI, Apr 2026, COMPANY-CLAIMED) | Not publicly disclosed | COMPANY-CLAIMED |
Humanoid HMND 01 | Not publicly disclosed | Not publicly disclosed | Not publicly disclosed | Not publicly disclosed | Not yet deployed (Dec 2026 earliest) | Not publicly disclosed | COMPANY-CLAIMED |
Tesla Optimus Gen 3 | Internal / future commercial | Not publicly disclosed | High (Tesla AI stack) | Not publicly disclosed | Internal Tesla factories only | Tesla 2025 Annual Report (VERIFIED filing; production targets COMPANY-CLAIMED) | COMPANY-CLAIMED |
Unitree G1 | Purchase | No | Low-Medium (open SDK) | Not publicly disclosed | Limited commercial deployment evidence | Not publicly disclosed | COMPANY-CLAIMED specs |
UBTech Walker S2 | RaaS / Purchase | Yes | High (proprietary) | Not publicly disclosed | Limited commercial deployment in China | Not publicly disclosed | COMPANY-CLAIMED |
Xpeng IRON | Not publicly disclosed | Not publicly disclosed | Not publicly disclosed | Not publicly disclosed | Not publicly deployed | Not publicly disclosed | COMPANY-CLAIMED |
Three findings from this matrix carry direct procurement implications. First, not a single platform in the 2026 commercial market publicly discloses its exit provisions. This is not a regulatory gap — it is a negotiating posture. Every buyer who accepts undisclosed terms is accepting terms the vendor has not written in the buyer’s favour. Second, software dependency level is uniformly High for any platform operating a proprietary fleet management layer — meaning the operational continuity risk upon vendor exit is structurally identical across competing RaaS offerings despite capability differences. Third, the financial continuity signals available for the two most commercially active platforms (Agility and Apptronik) reflect businesses in rapid capital consumption, which is a normal early-market condition but a procurement variable nonetheless.
Apptronik raised $520 million in February 2026 (Reuters, 11 Feb 2026, VERIFIED), placing it in a stronger near-term continuity position than its closest RaaS competitor at time of writing. Figure AI’s commercial and financial position is not publicly disclosed to the degree required for a full continuity assessment.
Section 03 — Deployment Evidence: Commitment vs Operating Reality
Direct Answer: As of September 2026, the gap between announced humanoid commitments and verified operating deployments is systematic, not incidental. Agility Robotics disclosed $1.78 million in net revenue for the period covered by its June 2026 SEC merger filing against $300 million-plus in disclosed multi-year orders — a ratio that reflects the early-stage gap between commercial pipeline and operating revenue (SEC filing, 24 Jun 2026, VERIFIED). IFR notes that many 2025 humanoid unit sales were for AI training data collection rather than productive deployment (Reuters/IFR, 21 Sep 2026, VERIFIED). Procurement buyers should calibrate contract protections to actual vendor operating maturity, not announced order totals.
The evidence review for this report applied a five-stage deployment pipeline to each announced commercial deployment:
announced commitment → contracted units → delivered units → operating units → measured outcome with disclosed metrics.
The result across all four deployments reviewed is that independent evidence thins rapidly after the announcement stage. This is not a criticism of any individual vendor’s commercial ambitions — it is a procurement reality that directly affects the risk weighting a buyer should place on financial continuity provisions.
“Humanoid robots remain a fraction of the overall robot population globally. Many of the units counted in 2025 were purchased for research and AI training data generation, not for productive industrial deployment.”
— Susanne Bieller, Secretary General, International Federation of Robotics
(21 September 2026, VERIFIED)
V2 — Deployment Outcome Tracker
Deployment | Announced Commitment | Contracted Units | Delivered Units | Operational Status | Measured Outcome |
GXO + Agility Digit (USA) | Industry’s first commercial humanoid RaaS deployment | VERIFIED: RaaS agreement confirmed | Not publicly disclosed | VERIFIED: GXO confirmed Digit is most advanced of its humanoid experiments; integrated with warehouse automation | Productivity/ROI data: Not publicly disclosed |
TMMC + Agility Digit (Ontario, Canada) | RaaS agreement following successful pilot | VERIFIED: RaaS agreement confirmed (Feb 2026) | Not publicly disclosed | Not publicly disclosed | Performance metrics: Not publicly disclosed |
BMW + Figure AI (Spartanburg, USA) | BMW Smart Robotics integration; announced Feb 2026 | COMPANY-CLAIMED: BMW states Figure 02 supported production | COMPANY-CLAIMED | COMPANY-CLAIMED: BMW states more than 30,000 X3 vehicles supported over 10 months | Independent audit: Not publicly disclosed |
Schaeffler + Humanoid (Germany) | 1,000–2,000 robots through 2032; 1 million+ actuator units (company-claimed) | COMPANY-CLAIMED announcement only | Not yet delivered — first install Dec 2026–Jun 2027 | Not operational at research cutoff | No outcome data available — deployment has not begun |
The Schaeffler entry illustrates the most consequential form of announcement-to-evidence gap: a 6-year commercial horizon announced with 2026 signing and 2027 delivery, carrying contract terms that span multiple hardware generations and are entirely undisclosed. The $1.78 million net revenue figure from Agility’s SEC filing is the starkest single data point in this review: a vendor with a disclosed $300 million-plus order pipeline that had converted $1.78 million of that pipeline into recognised net revenue at time of filing. The filing does not represent a terminal condition, but it does represent the actual operational-revenue gap that a three-year RaaS commitment is secured against.
Section 04 — Economics & ROI Analysis: The Migration Cost Model
Direct Answer: Fleet migration costs include eight components: integration redevelopment (new APIs, WMS/MES connections), operator retraining, AI model and workflow conversion, planned and unplanned downtime, replacement hardware at new platform pricing, tooling and end-effector replacement, workforce retraining, and decommissioning. For a 50-robot fleet transitioning between platforms in 2026, modelling based on Agility’s illustrative SEC economics suggests total migration costs could equal or exceed the original annual RaaS commitment when integration redevelopment and production downtime are included. The exact figure depends on integration complexity, software dependency depth, and whether third-party servicing rights were secured in the original contract.
Migration Cost = integration redevelopment + operator retraining + data and AI model conversion + planned and unplanned downtime + replacement hardware + tooling and end-effector replacement + workforce retraining + decommissioning
This formula is the report’s primary analytical instrument. It converts an abstract contractual risk into a figure procurement teams can weigh against the cost of securing better contract terms. The reference case throughout this section is a 50-robot fleet operating under an illustrative RaaS structure at $8,500/month per robot (Agility company-claimed / illustrative per SEC filing). The annual RaaS commitment for this fleet would be $5.1 million (Agility company-claimed / illustrative per SEC filing). All scenario figures below are illustrative and derived from publicly disclosed source materials as indicated.
Three Migration Scenarios
Scenario A — Voluntary Platform Upgrade
The buyer decides to migrate to a successor-generation platform at the end of a contract term or during a major hardware generation change. This is the lowest-risk scenario because timing is within the buyer’s control and the vendor relationship remains intact.
Scenario B — Vendor Platform Discontinuation
The vendor discontinues the contracted robot model mid-term — a commercially realistic scenario given the rapid hardware generation cycles observed across all eight platforms reviewed. The buyer must migrate without the original vendor’s full cooperation, potentially without access to successor hardware at contractually preferential pricing, and potentially with limited migration assistance.
Scenario C — Vendor Insolvency or Acquisition
The vendor files for insolvency or is acquired by a strategic buyer who restructures the commercial model. Service continuity for the existing fleet is at risk. Software support may be withdrawn. The buyer may need to migrate under time pressure without vendor assistance.
V3 — Migration Cost Model: 50-Robot Fleet Reference Case
Cost Component | Scenario A: Voluntary Upgrade | Scenario B: Vendor Discontinuation | Scenario C: Insolvency / Acquisition |
Integration redevelopment (new APIs, WMS/MES) | $150,000–$300,000 (illustrative) | $300,000–$600,000 (illustrative; no vendor cooperation) | $400,000–$800,000 (illustrative; no vendor cooperation, time pressure) |
Operator retraining | $50,000–$100,000 (illustrative) | $75,000–$150,000 (illustrative) | $100,000–$200,000 (illustrative) |
AI model and workflow conversion | $75,000–$200,000 (illustrative; model portability assumed partial) | $150,000–$400,000 (illustrative; no export provision) | $200,000–$600,000 (illustrative; data may be inaccessible) |
Planned and unplanned downtime (50-robot fleet; illustrative) | $100,000–$250,000 (illustrative) | $250,000–$750,000 (illustrative; unplanned transition) | $500,000–$1.5 million (illustrative; emergency transition) |
Replacement hardware (successor platform) | At market pricing — offset by hardware price decline | At market pricing — no contractual discount | At market pricing — no contractual discount; availability not guaranteed |
Tooling and end-effector replacement | $50,000–$150,000 (illustrative; varies by task specificity) | $100,000–$300,000 (illustrative) | $100,000–$300,000 (illustrative) |
Workforce retraining | $25,000–$75,000 (illustrative) | $50,000–$100,000 (illustrative) | $75,000–$150,000 (illustrative) |
Decommissioning | $25,000–$50,000 (illustrative) | $50,000–$100,000 (illustrative) | $75,000–$150,000 (illustrative) |
Illustrative total range | $475,000–$1.1 million | $975,000–$2.4 million | $1.5 million–$3.7 million |
All figures are illustrative, derived from publicly disclosed source materials and standard enterprise integration cost structures. They do not constitute financial projections for any specific deployment. Source: Agility SEC merger filing (Jun 2026) for RaaS reference pricing; Morgan Stanley and Reuters for hardware price trajectory data. See Section 10 — Scope & Disclaimer.
Against a $5.1 million annual RaaS commitment for a 50-robot fleet (Agility company-claimed / illustrative per SEC filing), the Scenario C migration range of $1.5 million–$3.7 million represents 29%–73% of one year’s RaaS cost absorbed entirely in a single migration event — with no new operational output during the transition period. Stated differently: a buyer who failed to negotiate source-code escrow, data portability, and migration assistance provisions may absorb the equivalent of approximately 3–9 months of RaaS payments in a single involuntary migration event (calculated as $1.5 million–$3.7 million ÷ $425,000/month fleet cost = 3.5–8.7 months).
The hardware price decline compresses this exposure further. Morgan Stanley estimated an average ~15% hardware cost decline across humanoid platforms in 2026 (Reuters, 27 Aug 2026, VERIFIED). Specific platforms exhibited steeper declines: Leju and Kuavo both reported approximately ~26% year-on-year reductions (Reuters, 27 Aug 2026, VERIFIED). A buyer locked into a fixed-price 3-year RaaS structure signed in early 2026 at peak hardware costs faces a compounding gap between their contracted cost-per-unit and the market replacement cost of equivalent capability. By month 36, replacement hardware at the same task capability could cost materially less than the contracted RaaS rate — but the contract carries no repricing right unless one was explicitly negotiated.
Section 05 — Named Deployment Case Studies
Direct Answer: Four disclosed deployments provide partial procurement precedents. BMW’s Spartanburg deployment used standardised interfaces — a replicable model for demanding interoperability rather than accepting proprietary integration. GXO’s RaaS agreement with Agility was the industry’s first commercial humanoid RaaS deployment, establishing a reference structure for service-level terms. Toyota Motor Manufacturing Canada progressed from pilot to RaaS agreement with Agility following a confirmed evaluation phase. The Schaeffler–Humanoid planned deployment through 2032 is the longest disclosed horizon reviewed and the most consequential test case for successor-platform and continuity provisions.
The standardised interface precedent: BMW and Figure AI at Spartanburg
The most immediately replicable procurement precedent from any reviewed deployment comes not from the contract terms — which BMW and Figure AI have not publicly disclosed — but from the integration architecture BMW chose. BMW Group announced in February 2026 that its Smart Robotics ecosystem would integrate Figure 02 using standardised interfaces (BMW Group, 27 Feb 2026, COMPANY-CLAIMED). BMW states Figure 02 supported production of more than 30,000 X3 vehicles at its Spartanburg plant over 10 months (company-claimed). The procurement implication is structural: a buyer who demands standardised interfaces — ISO-compliant where applicable, or open-protocol API connections to WMS and MES systems — is not building a proprietary integration layer that becomes worthless if the vendor exits. It is building an integration layer that the next platform can connect to with predictable redevelopment cost.
“Pilot projects help us to test and further develop the use of Physical AI under real-world industrial conditions.”
— Michael Nikolaides, Senior Vice President, Production Network, Supply Chain Management, BMW Group
(27 February 2026, COMPANY-CLAIMED)
No other publicly disclosed deployment in this review has matched that structural choice as explicitly. The absence of an independently verified audit of Figure 02’s Spartanburg performance data does not diminish the interface architecture as a contract negotiation precedent.
The multi-vendor hedging strategy: GXO Logistics and Agility Digit
GXO Logistics has been explicit about its strategy: it is testing Agility Digit, Apptronik Apollo, and Reflex robots in parallel rather than committing to a single platform (GXO corporate communications, COMPANY-CLAIMED). GXO confirmed that Digit has progressed furthest among its humanoid experiments and is integrated with warehouse automation (VERIFIED: GXO corporate statement). The contract structure governing GXO’s RaaS agreement with Agility has not been publicly disclosed. The procurement lesson is not the specific terms — it is the strategy. A buyer evaluating a single-platform RaaS commitment should ask whether the integration architecture, data portability provisions, and vendor exit terms would permit them to replicate GXO’s multi-vendor approach if the primary vendor relationship changed. If the answer is no, the contract is structurally preventing competitive leverage on renewal.
The pilot-to-commitment transition: Toyota Motor Manufacturing Canada and Agility
Toyota Motor Manufacturing Canada (TMMC) confirmed a RaaS agreement with Agility Robotics in February 2026 following a successful evaluation phase (Agility Robotics, 19 Feb 2026, COMPANY-CLAIMED for pilot success characterisation; VERIFIED for existence of RaaS agreement). This case illustrates the procurement risk point that most organisations encounter after a successful pilot: the transition from a limited-scope evaluation — where exit costs are minimal — to a multi-year commercial agreement — where exit costs are material. The terms that govern a pilot and the terms that should govern a commercial RaaS agreement are structurally different documents. A pilot’s terms typically contain no multi-year software dependency, no large-scale integration investment, and no accumulated AI model asset. A commercial agreement does. TMMC’s agreement terms are not publicly disclosed.
The duration risk case: Schaeffler and Humanoid through 2032
The highest-stakes case in this report has produced no deployment outcome, because the deployment has not yet begun. Schaeffler announced a planned deployment of 1,000–2,000 Humanoid HMND 01 robots through 2032, with an actuator supply commitment of at least 1 million units (Reuters, 13 May 2026, VERIFIED for announcement; all commercial terms and outcomes COMPANY-CLAIMED).
The first installation is scheduled for December 2026 to June 2027 at Schaeffler’s German facilities, with broader global rollout planned thereafter. The exit-risk calculation for this relationship is: 6 years from initial agreement to contract horizon; multiple hardware and software generations across that period (Agility, Figure and Apptronik have each cycled major platform versions within approximately 18–24 months, suggesting the 6-year Schaeffler horizon may span 3 or more hardware generations — Physical AI Journal Research Team inference; not a forecast); 0 publicly disclosed provisions covering what happens when Humanoid releases a successor platform, changes its commercial model, or is acquired. Schaeffler is a sophisticated industrial corporation with experienced procurement teams. The absence of any public statement about exit provisions does not mean those provisions were not negotiated. It means buyers evaluating comparable long-horizon deployments have no reference point from this case.
Section 06 — Friction, Risk & Unresolved Issues
Direct Answer: Six procurement risks remain materially unresolved as of September 2026: general-purpose task reliability below the 99.99% industrial threshold; vendor ability to alter robot behaviour through remote software updates without buyer consent; the gap between announced order values and recognised revenue (Agility: $1.78 million net sales against $300 million-plus in orders, SEC filing, 24 Jun 2026, VERIFIED); price decline exposure in fixed-cost RaaS commitments; the absence of publicly disclosed exit provisions in any commercial agreement reviewed; and operational dependency on vendor cloud services that may not survive a change in commercial relationship.
Risk 1 — Remote software update without buyer consent
No publicly disclosed humanoid commercial agreement reviewed for this report contains a provision requiring buyer consent before a software update is deployed to a live commercial fleet, a mandatory testing period before production deployment, or a buyer right to reject or delay an update. This is the least-discussed procurement risk in current humanoid commercial analysis. Every platform operating a proprietary fleet management layer — which includes all 8 platforms in the V1 matrix where software dependency is rated High — retains the technical ability to push software updates to deployed robots that alter their behaviour. In a conventional enterprise software contract, change management provisions are standard. In humanoid commercial agreements, they appear to be absent. A buyer whose warehouse logic has been tuned over 12 months of deployment can have that logic disrupted by a vendor software update. The procurement remedy is a clause requiring vendor notification, buyer consent for production-affecting updates, and a rollback right.
Risk 2 — Software-as-infrastructure dependency
Apptronik Fleet Connect and Agility’s fleet management layer are not software add-ons to a robot — they are the operational infrastructure through which the robot delivers value. Task assignment, telemetry, safety coordination, performance monitoring, and WMS integration all run through the vendor’s cloud layer. If the vendor’s cloud services are discontinued, degraded, or subject to pricing restructuring, the operational continuity of the buyer’s entire fleet is immediately at risk. No publicly disclosed humanoid RaaS agreement addresses what happens to fleet management access after the RaaS relationship ends. Source-code escrow for fleet management software — the provision that deposits the software source code with an independent escrow agent who releases it to the buyer upon a defined triggering event — is the contractual remedy. It is not present in any publicly disclosed agreement reviewed.
Risk 3 — Vendor financial continuity as a procurement variable
Agility Robotics disclosed $138.1 million in net losses against $1.78 million in net revenue in its June 2026 SEC merger filing (24 Jun 2026, VERIFIED). This is cited here not as an editorial commentary on Agility’s commercial prospects — early-stage capital consumption is structurally characteristic of this market — but as a procurement variable that a risk-adjusted contract model must incorporate. A three-year RaaS commitment with a vendor carrying this financial profile without fleet continuity obligations, insolvency protections, or source-code escrow is a procurement decision made without quantifying the counterparty risk dimension. The same financial scrutiny applies to any vendor in this market whose balance sheet is not publicly disclosed.
Risk 4 — Performance reliability at the industrial threshold
IEEE Spectrum analysis notes that humanoid platforms have demonstrated reliable performance for specific, constrained tasks but have not demonstrated the 99.99% availability threshold required for general-purpose industrial deployment (IEEE Spectrum, 2025, VERIFIED analysis). This is not a capability criticism — it is a scope definition. Buyers must define the task perimeter within which reliability is adequate for their production environment and ensure that the contract’s service-level agreement is scoped to that task perimeter, not to a general-purpose performance claim.
Risk 5 — Long-term fixed-price exposure as hardware costs decline
With ~15% average hardware price decline observed in 2026 (Morgan Stanley via Reuters, VERIFIED) and selected platforms showing ~26% year-on-year reductions (Reuters, 27 Aug 2026, VERIFIED), a buyer in a fixed-price RaaS contract without a repricing mechanism is exposed to a compounding gap between contracted cost and market replacement value across a multi-year term.
Risk 6 — Contractual opacity as a structural finding
Not a single commercial humanoid deployment agreement reviewed for this report contains a publicly disclosed exit provision. This is not a data availability problem — several of these agreements were announced through press releases that could have referenced exit terms without disclosing their substance. The absence of any reference to exit terms in publicly disclosed commercial humanoid agreements is itself decision-relevant intelligence: the default position of every vendor in this market, at time of writing, is to enter commercial agreements without surfacing exit provision terms to the market. Buyers who do not raise exit provisions in negotiation are accepting terms written entirely in the vendor’s negotiating interest.
Section 07 — Competitive Platform Comparison: Contractual Exposure by Commercial Model
Direct Answer: Each commercial model carries different exit exposure. Outright purchase maximises operational independence but leaves buyers holding hardware with no upgrade path if the vendor discontinues the model. RaaS transfers hardware risk to the vendor but creates counterparty and software-continuity dependency. Long-term leases typically carry termination fees. Outcome-based RaaS adds service-provider performance dependency. No single commercial model eliminates exit exposure. Buyers should negotiate exit provisions independently of the commercial model chosen — regardless of model, exit provisions must be negotiated as a separate exercise from the commercial terms.
Structuring platform comparison around commercial model rather than capability reflects the core finding of this report: contractual exposure type is determined by commercial structure, and capability differences between platforms are secondary to the procurement decision about which structural risks a buyer is willing to accept under which contract provisions.
V4 — Commercial Model × Exit Exposure Matrix
Commercial Model | Primary Exit Exposure | Software Continuity Risk | Data Portability Risk | Financial Continuity Risk | Termination Penalty Risk |
Outright purchase | Hardware obsolescence — no upgrade path if model discontinued | Medium — software updates may still require vendor connectivity | Medium — operational data may reside in vendor cloud regardless | Low — no ongoing service dependency | Low — no RaaS termination penalties |
RaaS (Robot-as-a-Service) | Counterparty and service-continuity dependency | High — fleet management layer depends on vendor uptime | High — AI models and operational data reside in vendor’s environment | High — vendor financial failure = service interruption | Medium — early termination fees typically apply |
Long-term lease | Termination penalty and residual value risk | Medium — depends on software licence terms | Medium — varies by lease structure | Medium — leasing entity may be separate from vendor | High — lease break penalties typically significant |
Outcome-based RaaS | Service-provider performance dependency and renegotiation risk | High — identical to standard RaaS | High — identical to standard RaaS | High — identical to standard RaaS | Medium-High — outcome definition disputes complicate exit |
The most consequential finding in the matrix above is that RaaS — the commercial model that has become the dominant structure in publicly disclosed humanoid commercial agreements — carries the highest software continuity risk and the highest data portability risk simultaneously. These are the two risks for which contractual remedies are most clearly available (source-code escrow, data export provisions) and most consistently absent from publicly disclosed agreements.
The “not publicly disclosed” position is a finding, not a data gap. When a vendor does not disclose its exit provisions, it is presenting the buyer with a negotiation that begins from a position of information asymmetry. The vendor knows what its standard terms say about vendor insolvency, platform discontinuation, and data portability. The buyer does not. The procurement standard should be to treat undisclosed terms as absent terms until the vendor demonstrates otherwise through the negotiation.
Section 08 — Strategic Recommendations: The Humanoid Exit Clause Matrix
Direct Answer: A multi-year humanoid commercial agreement requires ten exit provisions at minimum. Five are Mandatory regardless of commercial model: a vendor insolvency clause with fleet continuity obligations; platform discontinuation provisions with successor hardware pricing rights; source-code escrow for fleet management software; data and AI model portability with defined export format standards; and migration assistance obligations when the vendor triggers a platform transition. The remaining five are Mandatory or Negotiable by context. No publicly disclosed humanoid commercial agreement reviewed for this report contains any of these ten provisions.
Each of the ten provisions below is defined, contextualised against the migration scenarios in Section 04, and classified Mandatory or Negotiable under current market conditions. A buyer who secures all ten provisions has materially different stranded-asset exposure from a buyer who signs without them.
Provision 1 — Termination Rights and Notice Periods
What it is: A defined right for the buyer to terminate the agreement before its natural end date, with specified notice periods, cure periods for material breach, and termination-for-convenience rights. Why it is necessary: Without a defined termination right, the buyer’s ability to exit depends entirely on what the contract’s default terms say — which, where not publicly disclosed, should be assumed to favour the vendor. For a 50-robot fleet on a 3-year RaaS agreement, an undefined termination right could mean the buyer must continue paying $8,500/month per robot (Agility company-claimed / illustrative per SEC filing) regardless of whether the vendor is meeting service obligations.
Classification: Mandatory.
Provision 2 — Vendor Insolvency and Restructuring Protections
What it is: A clause specifying what happens to the fleet, its software, and its service obligations if the vendor enters insolvency, administration, or a restructuring process. Minimum provision: an obligation on the vendor — or its insolvency practitioner — to maintain fleet software access for a defined period and to cooperate with buyer data export. Why it is necessary: Agility’s SEC-disclosed financial position ($138.1 million net loss against $1.78 million net revenue, VERIFIED) is a quantified illustration of why this provision must be negotiated before signing, not after a trigger event. In Scenario C of the Section 04 model, migration costs of $1.5 million–$3.7 million (illustrative) are incurred at the worst possible time — when the vendor is least able to provide migration support.
Classification: Mandatory.
Provision 3 — Change-of-Control Notification and Right to Terminate on Acquisition
What it is: An obligation on the vendor to notify the buyer of any acquisition, merger, or majority ownership change within a defined period, combined with a buyer right to terminate the agreement without penalty if the acquiring entity is a direct competitor or if the commercial terms change as a result. Why it is necessary: In a consolidating market, the vendor the buyer contracted with and the entity that controls the buyer’s fleet post-acquisition may be different organisations with different interests. A buyer without a change-of-control clause may find their operational data and fleet management access transferred to a competitor.
Classification: Mandatory.
Provision 4 — Platform Discontinuation Provisions with Successor Hardware Pricing Rights
What it is: An obligation on the vendor to provide advance notice (minimum 18 months recommended) of any decision to discontinue the contracted robot model, combined with a buyer right to purchase successor hardware at a preferential pricing arrangement. Why it is necessary: Hardware generation cycles in the humanoid market are currently operating at under 24 months across the reviewed platforms. A 3-year RaaS agreement almost certainly spans at least one major hardware generation change. Without a successor pricing right, the buyer’s migration from a discontinued model to its successor is executed at whatever commercial terms the vendor chooses to offer. Classification: Mandatory.
Provision 5 — Source-Code and Software Escrow for Fleet Management Systems
What it is: An agreement with an independent escrow provider to hold the source code for the vendor’s fleet management software, with defined release triggers — typically vendor insolvency, withdrawal of software support, or failure to maintain agreed service levels — that give the buyer access to the code to maintain operations independently. Why it is necessary: This is the single most important provision for addressing the software-as-infrastructure risk identified in Section 06. Without it, Apptronik Fleet Connect, Agility’s fleet management layer, or equivalent proprietary systems become inaccessible upon any disruption to the vendor relationship — and the buyer’s fleet goes offline. Classification: Mandatory for any RaaS agreement with a fleet management dependency. Negotiable in purchase agreements where the buyer takes local software ownership.
Provision 6 — Data and AI Model Portability with Defined Export Format Standards
What it is: A contractual right to export all operational data — sensor logs, task performance data, workflow configurations, trained AI models — in a defined, non-proprietary format, upon request or upon contract termination, within a specified delivery window. Why it is necessary: The operational value accumulated in a humanoid deployment is a buyer asset, not a vendor asset — but without an explicit portability provision, the buyer has no contractual basis to demand it. AI model conversion costs in the Section 04 migration model range from $75,000 (Scenario A, illustrative) to $600,000 (Scenario C, illustrative) and are the single most difficult cost to compress without portability provisions.
Classification: Mandatory.
Provision 7 — Fleet Software and API Continuity Guarantee if RaaS Relationship Ends
What it is: A commitment from the vendor to maintain robot operational software — the software that runs on the robot itself, distinct from the cloud fleet management layer — for a defined period following any termination of the RaaS agreement, allowing the buyer to continue operating the fleet while arranging migration. Why it is necessary: Without this provision, a RaaS termination event could render the physical robots non-operational immediately — converting a migration situation into an emergency shutdown situation.
Classification: Mandatory.
Provision 8 — Right to Appoint a Third-Party Systems Integrator
What it is: An explicit right for the buyer to engage a third-party systems integrator — not the vendor — for programming, maintenance, optimisation, and integration work on the deployed fleet. Why it is necessary: Vendor lock-in through exclusive servicing rights is the oldest mechanism in enterprise hardware contracting. A buyer without the right to engage independent technical support is dependent on the vendor’s servicing capacity, pricing, and availability for every operational change. Classification: Mandatory. This provision also underpins the GXO multi-vendor hedging model — a buyer cannot run parallel platform evaluations if integration architecture requires exclusive vendor involvement.
Provision 9 — Spare Parts Availability Commitment for a Defined Post-Discontinuation Period
What it is: A commitment to maintain spare parts availability — or an escrow of critical spare parts at the buyer’s facility — for a specified period (minimum 5 years recommended) following any discontinuation of the contracted robot model. Why it is necessary: A buyer whose platform is discontinued mid-contract may continue operating the remaining fleet if software continuity is secured but will need physical components to maintain uptime.
Classification: Negotiable — the specific period and mechanism are negotiable, but some form of this provision should be present in every contract.
Provision 10 — Migration Assistance Obligations if the Vendor Triggers a Platform Transition
What it is: An obligation on the vendor to provide defined migration assistance — technical support, data export, integration documentation, and training — at no additional cost, if the vendor triggers a platform transition through discontinuation, insolvency, or fundamental service change. Why it is necessary: Scenario B and Scenario C in Section 04 both assume zero vendor migration assistance by default. The illustrative cost gap between Scenario A (voluntary upgrade with vendor cooperation: $475,000–$1.1 million) and Scenario C (insolvency/acquisition with no vendor cooperation: $1.5 million–$3.7 million) is the value of this provision in isolation.
Classification: Mandatory when the vendor triggers the transition; Negotiable for buyer-initiated transitions.
V5 — Humanoid Exit Clause Matrix
Provision | Classification | Present in Publicly Disclosed Agreements | Negotiation Priority |
1. Termination rights and notice periods | Mandatory | Not publicly disclosed | High |
2. Vendor insolvency and fleet continuity obligations | Mandatory | Not publicly disclosed | High |
3. Change-of-control notification and termination right | Mandatory | Not publicly disclosed | High |
4. Platform discontinuation with successor pricing | Mandatory | Not publicly disclosed | High |
5. Source-code and software escrow | Mandatory (RaaS) | Not publicly disclosed | High |
6. Data and AI model portability | Mandatory | Not publicly disclosed | High |
7. Fleet software continuity post-RaaS termination | Mandatory | Not publicly disclosed | High |
8. Third-party systems integrator right | Mandatory | Not publicly disclosed | Medium-High |
9. Spare parts post-discontinuation | Negotiable | Not publicly disclosed | Medium |
10. Migration assistance (vendor-triggered) | Mandatory (vendor-triggered) | Not publicly disclosed | High |
V6 — The “Can I Leave?” 7-Question Procurement Checklist
Before signing any multi-year humanoid commercial agreement, a procurement team should be able to answer Yes to each of the following questions:
1. Does the contract specify our right to terminate with defined notice and cure periods?
☐ Yes ☐ No
2. Does the contract specify what happens to our fleet, software access, and data if the vendor enters insolvency?
☐ Yes ☐ No
3. Does the contract require the vendor to notify us of any change-of-control event and give us a right to terminate on acquisition?
☐ Yes ☐ No
4. Does the contract give us the right to continue operating our fleet using software held in escrow if the vendor’s services are withdrawn?
☐ Yes ☐ No
5. Does the contract give us the right to export all operational data and trained AI models in a non-proprietary format?
☐ Yes ☐ No
6. Does the contract give us the right to appoint a third-party systems integrator for maintenance and programming?
☐ Yes ☐ No
7. Does the contract require the vendor to provide migration assistance at no additional cost if the vendor discontinues our platform?
☐ Yes ☐ No
A buyer who can answer Yes to all 7 questions has materially reduced their stranded-asset exposure across Scenarios A, B, and C from the Section 04 migration cost model. A buyer who cannot answer Yes to provisions 2, 4, 5, 6, and 7 simultaneously is exposed to the upper range of the Scenario C cost model — $1.5 million–$3.7 million (illustrative) for a 50-robot fleet — as a worst-case outcome with no contractual mitigation. The decision this report supports: engage qualified legal counsel to negotiate all ten provisions before signature, or use the Section 04 migration cost model to quantify the financial exposure of not doing so.
Section 09 — Executive FAQ
What contract clauses should companies require when buying or leasing a humanoid robot?
Ten provisions govern buyer exit rights in a multi-year humanoid commercial agreement, as set out in Section 08. The 5 highest-priority provisions — mandatory across all commercial models — are: vendor insolvency clause with fleet continuity obligations; data and AI model portability with defined export format standards; source-code escrow for fleet management software (essential for RaaS); platform discontinuation provisions with successor hardware pricing rights; and migration assistance obligations if the vendor triggers a platform transition. No publicly disclosed humanoid commercial agreement reviewed for this report contains any of these provisions. Engage qualified legal counsel to draft them into any multi-year agreement before signature.
What happens if a humanoid robot manufacturer goes bankrupt after we have deployed their robots?
Without an insolvency provision and software escrow in the original contract, the buyer faces the Section 04 Scenario C outcome: emergency migration with no vendor cooperation, potential immediate loss of fleet management software access, and migration costs in the $1.5 million–$3.7 million illustrative range for a 50-robot fleet. Agility Robotics disclosed $138.1 million in net losses against $1.78 million in net revenue for the period covered by its June 2026 SEC merger filing (VERIFIED) — a financial profile that makes vendor insolvency provisions a non-optional negotiating point for any buyer entering a multi-year RaaS commitment. The contractual remedy is Provision 2 (insolvency and fleet continuity clause) plus Provision 5 (source-code escrow), both detailed in Section 08.
Who owns the data and AI models trained during a humanoid robot deployment?
Ownership depends entirely on the contract. In the absence of an explicit data and AI model portability clause (Provision 6 in the Section 08 matrix), the operational data and trained workflows generated during a deployment may reside in the vendor’s cloud environment with no contractual basis for buyer export. This is the software-as-infrastructure risk documented in Section 06: the operational value accumulated in a deployment — trained AI models, workflow configurations, sensor logs — is a buyer asset, but without a portability provision, the buyer has no contractual mechanism to retrieve it upon relationship end. Buyers should specify both ownership (buyer retains all rights to data generated from buyer operations) and portability (vendor must export in a defined, non-proprietary format within a specified delivery window) as separate contractual clauses.
Can we keep using humanoid robots if we terminate the RaaS agreement?
Only if the contract explicitly provides for it. Without Provision 7 (fleet software continuity post-RaaS termination) and Provision 5 (source-code escrow), a RaaS termination event could render the physical robots non-operational immediately — because the software that runs the robot depends on the vendor’s continued provision of the service. GXO’s multi-vendor approach (Section 05) is the operational hedge against single-vendor software dependency, but it requires that individual vendor agreements permit third-party integration and do not contain exclusive operation provisions.
What does it actually cost to migrate a humanoid robot fleet from one platform to another?
Eight cost components determine total migration cost: integration redevelopment, operator retraining, AI model and workflow conversion, planned and unplanned downtime, replacement hardware, tooling and end-effector replacement, workforce retraining, and decommissioning. For a 50-robot fleet in 2026, illustrative modelling in Section 04 places voluntary migration (Scenario A) at $475,000–$1.1 million and emergency migration under vendor insolvency (Scenario C) at $1.5 million–$3.7 million. The difference between those ranges is the negotiable value of Provisions 2, 5, 6, and 10 in the Section 08 matrix — buyers who secure those provisions convert Scenario C exposure to something closer to Scenario A.
What should a five-year humanoid robot RaaS contract include to protect against vendor failure and platform discontinuation?
A five-year RaaS commitment spanning the 2026–2031 horizon will cover at least 2 major hardware generation changes based on current platform development cadences. The minimum contractual architecture for this horizon must include: insolvency clause with fleet continuity obligations (Provision 2); platform discontinuation provisions with at least 18-month advance notice and successor pricing rights (Provision 4); source-code escrow with defined release triggers (Provision 5); data and AI model portability (Provision 6); and migration assistance obligations for vendor-triggered transitions (Provision 10). The Schaeffler–Humanoid planned commitment through 2032 (company-claimed announcement, Reuters VERIFIED) is the market’s reference case for this duration, and its terms have not been publicly disclosed. Use the “Can I Leave?” checklist in Section 08 as the binary acceptance test for any contract at this horizon.
Section 10 — Scope & Disclaimer
This report is a market intelligence and contract analysis document produced by the Physical AI Journal Research Team, an independent analytical division of Sekason Research Limited, United Kingdom.
(1) No engineering opinion. This report does not assess, evaluate, or opine on the engineering design, construction, performance, suitability, or technical capability of any humanoid robot platform reviewed. No statement in this report constitutes an engineering opinion on any platform.
(2) No safety-certification claim. This report does not make any claim regarding the safety certification, CE marking, regulatory compliance, or safety record of any platform, deployment, or component. References to safety-related IEEE Spectrum analysis are reproduced as analytical context, not as safety assessments.
(3) No technical conformity assessment. This report does not offer any technical conformity assessment, conformity evaluation, or declaration of conformity for any humanoid robot platform with respect to any standard, directive, or regulation.
(4) Sample clause language is illustrative only — not legal advice. All contract provision descriptions, clause structure descriptions, and checklist questions in this report are illustrative only. They do not constitute legal advice, solicitation, or the creation of a lawyer-client relationship. Buyers should engage qualified legal counsel licensed in their jurisdiction before entering any multi-year humanoid commercial agreement.
Additional disclosures: Company-claimed figures are labelled throughout this report and have not been independently audited by Physical AI Journal or Sekason Research Limited. Illustrative migration cost figures in Section 04 are derived from publicly disclosed source materials and standard enterprise integration cost structures; they do not constitute financial projections for any specific deployment. The research cutoff for all data in this report is 21 September 2026. Information may have changed after that date.
Section 11 — References & Strategic Sources
Reuters / IFR: Document: “Humanoid robot sales tally hit 7,000 globally last year” | Date: 21 Sep 2026 | Tier: 1 | URL: https://www.reuters.com/technology/robotics/humanoid-robot-sales-tally-hit-7000-globally-last-year-2026-09-21/ | Classification: VERIFIED
SEC / Agility Robotics: Document: Agility–Churchill Capital merger filing (S-4/A) | Date: 24 Jun 2026 | Tier: 1 | URL: https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=agility | Classification: VERIFIED (financial data); COMPANY-CLAIMED (business projections)
Reuters: Document: “Humanoid to deploy up to 2,000 robots at Schaeffler plants” | Date: 13 May 2026 | Tier: 1 | URL: https://www.reuters.com/technology/robotics/humanoid-deploy-up-2000-robots-schaeffler-plants-2026-05-13/ | Classification: VERIFIED (announcement); COMPANY-CLAIMED (commercial terms)
Reuters: Document: “Humanoid startup Apptronik raises $520 million” | Date: 11 Feb 2026 | Tier: 1 | URL: https://www.reuters.com/technology/humanoid-startup-apptronik-raises-520-million-2026-02-11/ | Classification: VERIFIED
Reuters: Document: “China can build kung fu-fighting robots cheaper than anywhere” | Date: 27 Aug 2026 | Tier: 1 | URL: https://www.reuters.com/technology/robotics/china-can-build-kung-fu-fighting-robots-cheaper-anywhere-2026-08-27/ | Classification: VERIFIED (price decline data)
BMW Group: Document: “BMW Group to deploy humanoid robots in production” | Date: 27 Feb 2026 | Tier: 1 | URL: https://www.bmwgroup.com/en/news/general/2026/humanoid-robots.html | Classification: COMPANY-CLAIMED (performance outcomes)
IFR: Document: Humanoid Robots: Vision and Reality | Date: 14 Aug 2025 | Tier: 1 | URL: https://ifr.org/ifr-press-releases/news/humanoid-robots-vision-and-reality | Classification: VERIFIED
IFR: Document: World Robotics 2025 | Date: 25 Sep 2025 | Tier: 1 | URL: https://ifr.org/worldrobotics/ | Classification: VERIFIED
IEEE Spectrum: Document: “Humanoid Robots: The Scaling Challenge” | Date: 2025 | Tier: 2 | URL: https://spectrum.ieee.org/humanoid-robots | Classification: VERIFIED (analytical)
IEEE Spectrum: Document: “Digit 5 Sets a New Bar for Humanoid Robot Safety” | Date: 15 Sep 2026 | Tier: 2 | URL: https://spectrum.ieee.org/digit-5-humanoid-robot | Classification: VERIFIED (analytical); COMPANY-CLAIMED (Agility specifications)
GXO Logistics: Document: Humanoid technology / operational incubator | Date: 2026 | Tier: 2/3 | URL: GXO corporate communications (not publicly linked) | Classification: VERIFIED (deployment confirmation) / COMPANY-CLAIMED (performance claims)
Agility Robotics: Document: Digit 5 product announcement | Date: 15 Sep 2026 | Tier: 3 | URL: https://agilityrobotics.com/digit | Classification: COMPANY-CLAIMED
Agility Robotics: Document: “Toyota Motor Manufacturing Canada signs RaaS agreement” | Date: 19 Feb 2026 | Tier: 3 | URL: https://agilityrobotics.com/news/toyota-motor-manufacturing-canada | Classification: COMPANY-CLAIMED
Figure AI: Document: “Introducing Figure 03” | Date: 9 Oct 2025 | Tier: 3 | URL: https://www.figure.ai/news/figure-03 | Classification: COMPANY-CLAIMED
Figure AI: Document: “Ramping Figure 03 Production” | Date: 29 Apr 2026 | Tier: 3 | URL: https://www.figure.ai/news/ramping-figure-03-production | Classification: COMPANY-CLAIMED
Apptronik: Document: Apollo 2 product information | Date: 2026 | Tier: 3 | URL: https://apptronik.com/apollo | Classification: COMPANY-CLAIMED
Unitree Robotics: Document: G1 product and pricing | Date: 2026 | Tier: 3 | URL: https://www.unitree.com/g1 | Classification: COMPANY-CLAIMED
Tesla: Document: 2025 Annual Report | Date: Jan 2026 | Tier: 1 | URL: https://ir.tesla.com/sec-filings/annual-reports | Classification: VERIFIED (filing); COMPANY-CLAIMED (Optimus production targets)
This report is backed by authoritative research, independent verification, and structured analytical methodology.
© 2026 Sekason Research Limited. Physical AI Journal is an independent publication. physicalaijournal.org
