Capital markets have made up their mind about humanoids. Equity investors poured roughly $56 billion into robotics companies in 2026, nearly double the prior year's total, with the overwhelming majority landing in the US and China, according to Forbes' John Koetsier. London-based Humanoid just closed $152 million at a $1.35 billion valuation, pushing its total raised to $270 million. That is conviction capital betting on the makers of humanoid robots.
What doesn't exist yet is conviction capital for the robots as an asset class — the purchase-finance infrastructure a fleet buyer would need to acquire and hold a humanoid on a balance sheet rather than rent it. No securitization market, no standardized residual curves, no resale comp set. The equity story and the credit story are running on entirely different timelines, and the gap between them is the central underwriting problem this publication exists to track.
A collateral class with no floor
Start with price dispersion. Per RoboZaps' 2026 production-economics analysis, humanoid unit pricing already spans an enormous range — Unitree's G1 near $13,500 at one end, Figure's 02/03 units running $30,000 to $150,000, Agility Robotics' Digit around $250,000 in pilot deployments, and Fourier Intelligence's GR-2 priced between $150,000 and $170,000. That is not a market with an emerging comp set; it's five different asset classes wearing the same category label. An underwriter trying to set a loan-to-value ratio or a residual assumption has no consistent reference price to anchor to, which is the first precondition equipment finance needs before purchase lending scales.
The second precondition — a stable cost curve — is also missing. RoboZaps, citing DIGITIMES Research, notes that actuators and motion systems alone account for 40-50% of a humanoid's bill of materials, and per-unit costs are projected to fall 50-70% after 2029 as actuator technology matures and scales. A robot financed today is being underwritten against a foreseeable step-down in replacement cost. That is not a hypothetical technology risk lenders might someday face; it's a scheduled repricing event already visible in the supply-chain data.
The GPU-server preview
Equipment finance doesn't have to imagine how this plays out — it's living the analog case in real time. In the GPU-server market, residual value assumptions after a standard three-year lease term range from as low as 10% to as high as 60% of original cost, according to Monitor Daily's Bernie Margulies, writing on American Compute's GPU Residual Value Report, which draws on nearly 77,000 resale transactions. A 50-point spread on the central assumption in a lease structure is not a modeling nuance; it's a sign the asset class hasn't settled on what it's worth used. Margulies frames the risk starkly: obsolescence in fast-moving compute hardware is 'sudden, not gradual.'
He reaches back to 1979 for the cautionary precedent — IBM's mainframe price cut of nearly 70% in a single announcement, which repriced the entire used mainframe market overnight and pushed residual-value lessors into bankruptcy within months. That is the scenario equipment financiers invoke whenever a new fast-depreciating tech asset shows up asking to be collateral. Humanoids, with a scheduled cost collapse already flagged for 2029 and no track record of resale liquidity to fall back on, sit squarely inside that pattern.
Why the RaaS default makes sense — for now
This is precisely why humanoid deployment today runs almost entirely through robotics-as-a-service and pilot-lease structures rather than purchase financing: the manufacturer, not a third-party lender, is best positioned to absorb residual-value risk it alone can forecast. That default won't hold indefinitely. Industrial robots, cobots, and AMRs crossed this same bridge years ago — real resale markets and depreciation schedules now exist for those categories, which is why purchase financing for them is unremarkable today. Humanoids will eventually follow, but the signal to watch isn't a headline funding round. It's the first point at which someone — a lender, a marketplace, a rating agency — publishes a resale comp set for a used humanoid unit. Until then, the $56 billion flowing into equity and the near-total absence of purchase-finance infrastructure for the hardware itself are two honest measurements of the same immature market.