Who keeps the risk in hardware finance?

The company’s funder may never own the machine

A venture round funds the company. A lease funds an asset. A customer prepayment helps fund delivery. The same deployment can use all three, and each provider receives a different claim on what the business earns or owns.

Those claims matter most together. A fleet operator can owe rent to a lessor, service to a prepaid customer and a return to shareholders; a funding announcement that names only the shareholders leaves most of the operating bargain out of view.

The payment survives the growth plan

Equity absorbs a disappointing deployment through a lower return. Debt and rent introduce a payment calendar. The operator must meet it even when paid use falls, unless the contract gives it a way out.

Customer and public funding have their own clocks. A customer who pays early is buying delivery; a grant may require jobs, spending or milestones before the company can keep the benefit. Cash arrives with a job to do.

I therefore follow the payment from the operating asset to the claimant who can enforce it. Written guarantees decide legal recourse. A parent may also choose to rescue a legally nonrecourse vehicle to protect customers or future funding, which creates a commercial incentive beyond the signed promise.

Waymo buys time with parent equity

Waymo’s 2024 and 2026 rounds total $21.6 billion of visible equity, with Alphabet supplying a significant majority in 2026. In the July comparison, that is a patient parent backstop without an identified large fleet facility setting the refinancing calendar.

The cars still consume capital. Charging and depots have to be built, vehicles maintained, and autonomy hardware refreshed before a growing fleet can pay its way. With no quantified transfer of those obligations in the reviewed package, operating losses and impairment remain with Waymo equity and, economically, Alphabet.

That is flexibility purchased by concentration. Alphabet can tolerate a longer ramp than a lender demanding scheduled repayment, but it also carries the opportunity cost and can change its willingness to fund the next one. The package leaves vehicle titles, depot leases, liens and hardware ownership unresolved.

Serve’s robot has to earn its equipment payment

Serve layers Farnam secured equipment financing and a Magna supplier warrant onto corporate equity. It can deploy a robot with less purchase cash from shareholders, while the lender receives collateral and the supplier receives equity economics.

The July model exposes the early mismatch. FY2024 cash capex was 5.66× revenue. At modeled base utilization of 48.2%, fleet revenue is about 0.28× the Farnam payment, before field operations, maintenance or replacement. That is a revenue-to-payment diagnostic, not a coverage covenant; even on that narrow comparison, the fleet needs more paid work.

A supplier warrant can help secure production and align incentives, but its economic cost includes dilution. Concentration and minimum-volume dependencies also belong in the assessment where the agreement imposes them. Financing the next robot becomes attractive when its service cash grows faster than the fixed claims, rather than merely because another robot can be financed.

CoreWeave borrows against work it still has to perform

CoreWeave’s FY2025 model records $21.615 billion of debt principal, $8.195 billion of operating-lease obligations, $254 million of finance leases and $8.185 billion of deferred revenue. Selected asset financing equals 78.6% of gross property and equipment; capex is 2.01× revenue.

This stack finances capacity before all of its service revenue is earned. Lenders have secured claims, landlords have rent due, and customers who prepaid expect working capacity. Deferred revenue brings cash forward without releasing CoreWeave from the cost of delivery.

The model’s operating-cash-flow debt-service coverage is 0.70×. Removing the deferred-revenue increase leaves negative $1.116 billion of operating cash flow. These analytical adjustments show how much the funding depends on customer cash arriving early; they use no governing covenant definition.

The credit question therefore reaches beyond whether the GPU has a customer today. A delay between equipment delivery and customer acceptance can start financing charges before billing begins, while a later technology change can reduce the price of the next contract before the old debt is repaid.

A five-year loan outlasts a three-year customer contract

DDTL 5.5 makes that timing visible. The August 2026 facility provides $2.6 billion of committed delayed-draw capacity to CoreWeave Financing DDTL V-V, LLC, an indirect subsidiary, primarily for GPU servers and related infrastructure needed to perform customer contracts. Committed capacity is available borrowing, not an amount fully drawn at closing. CW1

The loan matures September 1, 2031, roughly five years after closing; underlying customer contracts average about three years. The initial customer can pay exactly as promised and leave debt outstanding when its contract ends. CoreWeave’s closing release explicitly includes renewal or re-leasing in the repayment story. CW3

The June filing describes DDTL collateral generally as assets underlying contributed customer contracts and pledged contractual cash flows, often involving investment-grade counterparties. That gives a lender a defined customer and asset pool to underwrite. The remaining years require a second judgment about what the equipment can earn after the first contract. CW2

The 8-K summarizes a minimum 1.35× debt-service coverage covenant beginning after its specified commitment-related trigger. Its defined cash-flow numerator, debt-service denominator and start date control the test. Headline revenue divided by interest would answer a different question. CW1

Delayed draws reduce the cost of borrowing before qualifying equipment needs to be paid for. They do not erase it. In the assumed example below, an unused commitment costs money, and a drawn balance costs the benchmark plus the margin before amortization, reserves, hedging and transaction costs.

Cash can be trapped while the parent remains exposed

CoreWeave describes certain financing vehicles as bankruptcy-remote, special-purpose consolidated subsidiaries. Their pledged assets still appear in consolidated accounts. A legal restriction on an asset and accounting consolidation can coexist. CW2

Recourse varies across the same company’s facilities. The June filing classifies DDTL 4.0 as nonrecourse apart from specified carve-outs; DDTL 5.5 carries an unconditional parent guarantee. Calling both “SPV debt” would conceal the difference that matters when collections fall short. CW2 CW1

The cash path matters as much as the asset path. Receipts in a restricted vehicle may have to pay operating costs, replenish reserves and service debt before any remainder reaches the parent’s next build, even while a parent guarantee exposes the wider company to a shortfall inside that vehicle.

A recovered GPU also needs an operating home. Power, networking and a replacement customer affect what it can earn, while CoreWeave supplies the software and operations that make the financed equipment productive. The lender is underwriting that dependency alongside the customer’s payment obligation. A collateral pool organizes the claim; it cannot manufacture a second customer.

Anduril has to earn the right to keep public support

Anduril combines venture capital, customer-linked demand and public support. The July model compares a $310 million JobsOhio grant with $910.5 million of capex, or 34.0%. It estimates $426.1 million of base net public benefit and about 0.88× public support to private capital.

Those figures describe announced or modeled economics. The company benefits when support is funded, collectible and retained after compliance and clawback conditions. A grant, a tax credit and a public guarantee help through different mechanisms and on different dates.

A customer contract can support factory demand without paying the construction bill. A grant can reduce that bill without ensuring a profitable product. Anduril still needs private capital to bridge the dates and qualified output to justify the capacity; a delayed or resized program can weaken the business even after incentives have made the factory cheaper.

Northvolt’s rescue cash could not refill the old stack

Northvolt financed factories with equity, project debt, customer commitments and prepayments, public finance, supplier exposure and bridge capital. When the ramp disappointed, those sources became competing claims on a weaker operating asset.

The July model places a $100 million debtor-in-possession facility beside $5.8 billion of filing debt: 1.7%. Bridge liquidity is 4.22% of debt; modeled base residual exposure is $4.4 billion. Rescue money can preserve the option to operate or sell while leaving a large loss in the existing capital stack.

An offtake agreement promises a buyer. The factory still has to make acceptable batteries at an economic yield and cost. Customer prepayments may become delivery or refund claims, project lenders retain security, and suppliers can have liens or enough operating leverage to shape a rescue.

Recovery then depends on what the unfinished plant can be sold for, what it costs to complete, and which contracts transfer with it. The $4.4 billion is a modeled exposure, not a recovery appraisal. Financial closing bought the chance to execute; it never guaranteed the ramp.

Humanoid equity is financing the experiment and the machine

Apptronik’s February 2026 announcement combined a $520 million Series A-X extension with a $415 million initial Series A, taking the round above $935 million and total capital raised to nearly $1 billion. Figure announced more than $1 billion of Series C commitments in September 2025 at a $39 billion post-money valuation. These are venture and strategic equity rounds carrying technology and deployment uncertainty. HR1 HR2

A mature equipment lender wants a machine whose service life and collections it can estimate. A pre-scale humanoid can change between generations and depend on proprietary controls, repairs and customer integration. Equity can bear that uncertainty because it participates in the upside. Figure’s equity valuation says nothing directly about what a lender could recover from its robots.

The use of funds explains the scale. Figure said the round would expand BotQ manufacturing, deployment into homes and commercial operations, GPU infrastructure for Helix training and simulation, and data collection. Apptronik described production and deployment of Apollo. The same capital is paying for a factory learning curve and an AI learning curve. HR1 HR2

Apptronik’s disclosed investors include Google, Mercedes-Benz, AT&T Ventures and John Deere. A strategic investor can bring manufacturing knowledge or a route to a customer, but an investment is still a different commitment from a paid fleet order. A pilot can end before purchase, and an order can require acceptance before payment.

Eventually, accepted fleets could support debt while equity funds the next generation. I would want a cohort’s installed cost, billable hours, collections and repair history before assigning that debt capacity. Factory capacity and demonstration videos supply neither the collections nor the recovery record.

The loss follows the promise, not just the title

A lessor may hold title while the operator guarantees residual value. A lender may hold a first lien on equipment that has little resale value. Ownership is a starting point; the enforceable promise determines who is asked to make good on the loss.

Lower paid use hits operating contribution before it changes principal. Reserves can absorb the first shortfall, then a sponsor cure or guarantee can move it back to the parent. Foreclosure changes control of the equipment without necessarily repairing the unpaid balance.

The indicators below measure different pressures. They are worth comparing as failure mechanisms, not ranking as though a grant ratio and debt-service coverage were versions of the same accounting measure.

Replacement can arrive before the old debt is gone

Consider an assumed $100 asset financed with $20 of equity and $80 of debt repaid evenly over four years. After two years, $40 of principal remains. If a technology shock makes the asset unusable with no resale value and a replacement still costs $100, the operator needs the new machine while carrying the old $40 claim, before interest and operating costs.

The debt schedule has not extended the machine’s economic life. Insurance, transferable customers, reserves or a restructuring can change the outcome, but each needs an identified contract or funded source.

I would put the replacement date, remaining principal and committed service cash on the same timeline. The strongest structure is the one that can survive a reasonable miss in paid use and asset life without requiring shareholders to keep refinancing the same experiment.

Method and unresolved ownership

The report and workbook classify evidence as confirmed, strongly supported, modeled, plausible but unproven, or unknown. That hierarchy matters here because a disclosed facility can support a precise claim about recourse while leaving title, collateral eligibility or recovery unresolved.

The July comparative inputs were not refreshed for this rewrite. The September-reviewed CoreWeave and humanoid sources add named structures at their own dates. Governing loan documents, asset titles and customer terms would be needed to close the remaining ownership and repayment gaps.

A cheaper deployment can leave the same loss

I prefer financing that follows a proven asset’s collections to financing that asks future growth to rescue old equipment. A good lender can be protected while the operator remains fragile. The attractive hardware business is the one whose capacity pays for service, replacement and its capital claims through a weaker year—not simply the one that deploys the most machines per dollar of new equity.

Sources

1. The Hidden Financing Behind Hardware Startups — supplied comparative report

   Verified Jul 22, 2026

2. Hidden financing comparative model — supplied 29-sheet workbook

   Verified Jul 22, 2026

3. Capital-stack diagrams — supplied PNG and Graphviz source package

   Verified Jul 22, 2026

4. CoreWeave 2025 Form 10-K
5. CoreWeave Q1 2026 Form 10-Q
6. Fitch — CoreWeave DDTL 5.5 rating action
7. Serve Robotics 2024 Form 10-K
8. Serve Robotics FY2025 earnings exhibit
9. Alphabet 2025 fourth-quarter earnings materials
10. CoreWeave August 7, 2026 DDTL 5.5 guarantee and security terms

    Verified Oct 16, 2026

    Dated supplemental primary record: parent unconditionally guarantees this facility. Does not recertify all historical comparative inputs.

11. CoreWeave June 2026 Form 10-Q — DDTL collateral, consolidation and recourse

    Verified Sep 23, 2026

12. CoreWeave DDTL 5.5 closing release — maturity and customer contract duration

    Verified Sep 23, 2026

13. Apptronik — more than $935 million Series A financing

    Verified Sep 23, 2026

14. Figure — more than $1 billion Series C at $39 billion post-money valuation

    Verified Sep 23, 2026

Supporting material

• Full comparative report

  The complete 19-page investigation, including methodology, findings hierarchy, appendices, and source register.

• Editable comparative report

  The supplied editable report used as the source document for this web edition.

• Comparative financing model

  The 29-sheet evidence ledger, assumption book, company comparison, scenarios, waterfalls, and findings workbook.

• Capital-stack diagram package

  The original full-resolution PNG and Graphviz DOT files for all five company capital stacks.

• Waymo capital stack

  Full-resolution parent-funded platform diagram.

• Serve Robotics capital stack

  Full-resolution equity, equipment-financing, and supplier-warrant diagram.

• CoreWeave capital stack

  Full-resolution secured-debt, lease, and customer-prepayment diagram.

• Anduril capital stack

  Full-resolution equity, customer-contract, and public-support diagram.

• Northvolt capital stack

  Full-resolution equity, project-debt, offtake, public-support, and distress diagram.

Read next