Part V: Bankable Machines

What makes this a schedule is that every durable machine class has joined the balance sheet the same way, and the history is unusually well documented.

Rail wrote the original template in the 1800s. The equipment trust certificate, the Philadelphia Plan, put the locomotives in a title-holding trust, advanced about 80% against them, retired the notes serially, and earned a reputation as one of the safest securities on the rails, because the trust survived the railroad's bankruptcy. Aircraft inherited the structure: a modern EETC is a rated pass-through trust with a liquidity facility covering 18–24 months of interest, senior tranches rated above the airline itself, and Section 1110 of the Bankruptcy Code giving a 60-day affirm-or-surrender right on the collateral. No Class A EETC investor lost principal through American Airlines' 2011 bankruptcy, and the market runs $15–25 billion a year. Containers show the endpoint of full bankability: Triton, the largest lessor at over seven million TEU, was bought whole by Brookfield Infrastructure for $13.3 billion in 2023, Textainer went to Stonepeak for $7.4 billion, CAI to Mitsubishi HC. When the records are good enough, infrastructure funds simply buy the lessors.

$54.43Msingle tranche rated BBB+ by S&PSolar is the modern rehearsal, and it started small. SolarCity 2013-1, priced November 21, 2013: $54.43 million, a single tranche rated BBB+ by S&P, 4.8% coupon, overcollateralized, backed by pooled systems and leases. It was the only solar securitization that year, roughly three years after standardized production data existed to underwrite it. By 2024 the average solar deal ran $298 million and the largest, Sunrun's, $886 million; Sunrun raised about $2.8 billion non-recourse in 2025 alone. The structure passed its stress test the same year: Sunnova went bankrupt in 2025 and its roughly $6 billion of rated securitizations kept paying: bankruptcy-remoteness, proven in the worst case.

Compute ran the same movie at speed. Datacenter securitization set a record above $8 billion in 2024, with Morgan Stanley forecasting ~$25 billion by 2028. CoreWeave went from a $2.3 billion exotic GPU-collateralized facility in 2023 to the $8.5 billion DDTL that closed in March 2026 as the first investment-grade GPU-backed financing, rated A3, with a publicly syndicated $3.1 billion facility behind it: machine collateral to investment grade in thirty months. Lambda issued the first GPU ABS in April 2024, $500 million led by Macquarie. The pattern, measured twice in a decade: standardized performance data first, securitization about three years later.

~$1.3TUS equipment finance a yearThe balance sheet waiting for robots is not hypothetical either. US equipment finance runs about $1.3 trillion a year, with the machinery and the staff already in place. Robot fleets sit before the clock has started, and the missing input is the record.

$46,000robot costs to build, Chinese suppliersThe hard half is what the machine is worth in year four, and mid-2026 pricing shows the problem in one line. Unitree sells its G1 from $13,500, its H2 from $29,900, its R1 from under $6,000, while Western-built platforms price around $150,000, and Morgan Stanley's teardown says the same robot costs about $46,000 to build with Chinese suppliers. Western-priced hardware reprices toward the Chinese floor over its own service life, and whoever holds the residual eats the slope.

The GPU market just demonstrated what happens when the residual is guessed. Lambda's first-of-its-kind ABS was reportedly underwritten around 50% residual value at three years (a figure from trade commentary; the deal documents disclose no curve), and Nvidia's Blackwell generation promptly cratered H100 resale values.

The classes that survive fast generational turnover survive by decomposition. A jet engine gets superseded too (the CFM56-3 was parked in droves when the -7B arrived) and engines stayed among the most reliably financed assets on earth, because 60–80% of a used engine's value sits in the remaining life of its life-limited parts: the asset is priced as a bundle of serviceable components, with half-life values published and power-by-the-hour contracts covering a third of the fleet. Cars re-model annually, and the entire auto-finance complex prices off a published residual curve plus a deep auction market. A transparent curve and a liquid resale channel make even a fast-modeled asset bankable. Robots have neither today, and the resale desk would create both. The desk takes returned units, re-tasks and re-leases them, publishes what they actually cleared, and turns generational churn into a parts and re-lease market instead of a write-off; used-industrial-robot marketplaces and certified pre-owned programs already exist to plug into. Residual risk therefore belongs at the financing trust that owns the machines and runs the desk. On the customer it stalls the sale. On a thin operator it eventually kills the operator. A robot that can be re-tasked is collateral; a stranded robot is e-waste with a loan against it.

90%of a single shift to break even todayThe lease rate has a ceiling (the ~$30-an-hour fully loaded human the robot displaces), and debt service does not care about utilization. At today's operating cost (Agility's cited $10–12 an hour) with the collateral marked near retail, a financed robot must bill roughly 90% of a full single shift just to cover debt service and operating cost. The model does not clear on today's numbers. It clears when the operating-cost curve bends toward the $2–3 target or the collateral marks near the Chinese BOM floor; at that corner, breakeven falls to roughly 15–31% of a single shift, with real margin left for the operator. One scope note governs that comparison: the sub-$17,000 BOM is BofA's China-built figure, and Western pilot-stage units run $90,000–100,000 by the same bank's count, so the floor-marked scenario assumes Chinese-built hardware. Bankability depends on the operating-cost curve bending before the pilot cohort matures.

The mechanics: a local owner-operator enters with modest capital (Amazon's own figure says startup costs begin around $10,000; third-party consultants put the practical ceiling near $30,000) and runs 20–40 vans leased through Element Fleet, against a scorecard graded Fantastic to Poor that gates route allocation. The vans are financeable because an investment-grade counterparty guarantees the demand. The returns are published: $1–4.5 million gross per DSP at 7–10% net margin, $75,000–300,000 annual profit, across 3,500-plus operators, roughly 275,000 jobs, and some $45 billion of operator revenue since 2018. Amazon got a national fleet with no fleet capex and no employment liability.

$44.6Mjury verdict on Amazon's controlThe failures are equally specific, and each one maps to a design-out. A class action on behalf of roughly 2,500 DSPs (Fli-Lo Falcon v. Amazon) alleges the company controlled operations while overstating operator profits. A King County settlement paid $8.2 million on driver misclassification in the DSP chain. Amazon terminated the Battle Tested Strategies contract in 2023 and the NLRB found it a joint employer of that DSP's drivers anyway. Rural DSPs collapsed in 2024 when routing-algorithm changes overloaded them, and operators rank scorecard threshold creep their top financial concern, above insurance, above fleet cost. A $44.6 million jury verdict in a case about Amazon's control over its DSPs made the general point: control equals liability, whatever the contract says.

So the robot version, call them Robot Service Partners, gets built with the fixes in the founding documents. The residual and the hardware risk stay at the trust, with guaranteed buybacks; the thin operator never holds the depreciation. The scorecard rubric is published as a contractual right, because operators sued over opaque metrics gating their income and they were right to. Operator downside is capped by uptime guarantees backed by spare-parts pools, which is also the only structure that holds robot-parts inventory without losing money: pools already spoken for under service contracts, per the inventory trap in the parts arithmetic. The relationship is legally a franchise, which is boring, known work: an FTC disclosure document, registration in about thirteen states. RSP equity is the operator's reputation, priced by the scorecard, and vetted operators bid for financed fleets in one marketplace.

214%Chinese government humanoid procurement, 2024The demand guarantor is the load-bearing piece, and no government anywhere has signed a humanoid-fleet procurement contract yet. Congress is already writing rules for federal ground-robot procurement, DARPA funds humanoid logistics programs, defense robotics spending is projected past $30 billion a year by 2027, and Chinese government humanoid procurement grew 214% in 2024. Even so, the first postal route, base-maintenance contract, or municipal fleet deal remains unsigned. Whoever signs it first hands its RSPs the most creditworthy demand guarantee available and starts the loss history everyone else will have to buy later. No robot owner-operator franchise has launched anywhere as of this writing.

~$707per robot per year, ¥500,000 capThe insurance half of that spine already exists on one side of the Pacific. Since September 2025, PICC, China Pacific, and Ping An write dedicated humanoid policies; a documented Wuhan case insured two 60-kilogram humanoids at about ¥5,000 (roughly $707) per robot per year with a ¥500,000 cap, covering physical damage, third-party liability, and algorithmic malfunction. Western coverage so far is AI-liability wraps (Relm's PONTAAI, Munich Re's products, Mosaic's aiSure), with no physical-damage humanoid-fleet book anywhere in the Western market. Every insurer entering the category builds its risk view from scratch, and the trade press quotes them saying exactly that.

Telematics supplies the precedent, with observed discount schedules. Progressive's Snapshot repriced personal auto risk off driving data, and about one customer in five sees a premium increase, proving the data actually reprices risk. Its fleet program pays 5% up to 18% for sharing telematics from preferred vendors; Samsara's direct integration carries 3%; the observed band is 5–20%, partner-specific. Samsara itself supplies the operating model: a neutral platform ingesting 25 trillion-plus data points a year, at $1.6 billion of revenue growing 30%, whose feed fleets hand to insurers to cut premiums. Vehicles have their reference record layer. Robots do not; the closest things are the fleet-ops tools from the deployment chapter, none of them a standard a lender prices against.

The sequence, once the record exists, is the one every asset class above followed: originate fleets one at a time on contracted revenue; rate them as loss history accrues; roll them into a program; scale. The first standardized robot-fleet feed becomes the reference the way the first credible mortality table did: every counterparty prices against it because there is nothing else to price against, and nobody owns that seat today.

Then the compounding: fleets financed one at a time roll into a master program the way credit-card trusts work: one venue, one ratings framework, pooled diversification, cheaper capital each vintage. The platform's take is deliberately boring: one to two points of origination, 50–100 basis points of servicing, ratings and telemetry priced like a small Moody's, a share of the insurance program built on the same records.

$10.8Bannual originations at 30% financed share and 75% advanceThe origination bracket uses two explicit assumptions: 30% of BofA's 1.2 million humanoids a year by 2030 are fleet-financed, and lenders advance 75% of collateral value. At a ~$40,000 installed mark, the arithmetic is 1.2 million × $40,000 × 30% × 75%, or about $10.8 billion a year of originations.

$4.6Bannual originations at China BOM mark; same assumptionsAt BofA's own sub-$17,000 China-built BOM mark, the same assumptions produce about $4.6 billion a year. The result scales linearly with the assumed fleet-financed share and advance rate. The collateral mark dominates the bracket, the low end assumes Chinese-built hardware, and both numbers inherit every uncertainty in the unit forecast. Even the conservative corner is a real securitization category, and the platform economics above are tolls on whichever number arrives.

The comparison set for what the toll-taker is worth runs in both directions. Moody's is an ~$85 billion company because underwriting data compounds; every rated deal makes the next rating more authoritative. The cautionary comp is the crypto launchpad this design borrows its allocation mechanic from: Virtuals peaked near $5 billion of market value and gave back roughly 90%, having generated about $66.7 million of cumulative protocol revenue, with measured correlation between its tokens' prices and actual agent commerce of approximately zero. On venue: the securities-law path for tokenized notes has clarified (an SEC no-action letter, a joint SEC-CFTC taxonomy), but the CLARITY Act stalled in the Senate after passing the House, so the conservative structure leads: plain private securitization through clean, KYC'd entities, with tokenization as optional plumbing.

If fleet utilization can't service debt at achievable lease rates once pilots mature, there is no asset class, only a services business wearing a securitization costume. The observable is mature-pilot lease rates against debt service, with the $10–12-an-hour operating cost bending toward $2–3. The model is pre-viable at today's cost; the tripwire is the curve failing to bend by the time the pilots season.

If every robot generation obsoletes the last and the resale desk can't re-lease, the collateral math fails. The GPU market has already shown this failure mode is real. The observable is the first cohort of returned units: what the desk actually clears, published. Scale issuance only after the desk has cleared real returns at real prices.

53%GM captive financed US retail salesIf the platform incumbents captive-finance their own fleets before an independent standard exists, the independent version shrinks to the long tail. This is the liveliest of the three, because captives have taken exactly this seat before: GM's captive financed 53% of its US retail sales at the pandemic peak and low-40s steady-state, John Deere's captive is the model franchise, and Tesla already reprices insurance monthly off its own vehicle telemetry. The aircraft market argues the other way: independent lessors, with residual diversification across operators, ended up placing half the world fleet against the manufacturers' captives. Which pattern robots follow is genuinely open and turns on timing: whether a dominant OEM locks the financing-plus-telemetry seat before a neutral record standard exists. The observable is public and unambiguous: an OEM launching a captive finance arm and a proprietary telemetry standard together.

None of the three has triggered. The flywheel runs until one does: deployments produce the records, the records produce cheap capital, and cheap capital fields more deployments.