# A Financing Facility Is Not a Cash Buffer

## How supplier financing can support AI through the states when funding is hardest

AI Infra Credit · October 4, 2026  
Anthropic IPO: Financing the Frontier · Essay 4  
Free public research · Published October 4, 2026

The productive question about supplier financing is not whether money travels around an AI ecosystem. It is whether the arrangement brings useful capacity online, and whether someone can fund the obligations when the ordinary path stops working.

A large facility can be highly valuable for expansion and unavailable for rescue. A convertible note can replace an immediate claim without putting cash in the hands of the institution that must keep the equipment financed. Neither observation makes the financing fictitious. Both change how it should be designed.

**The next improvement in AI finance may come from changing which states a facility covers, not simply increasing its maximum size.**

The Broadcom–Anthropic discussion provides the immediate context. This paper adds a conditional funding model, not a new estimate of Anthropic's cash needs. It compares cash loans, accepted noncash notes, funded equity and a negotiated cure window. The constructive aim is to preserve productive service with less fragile dependence on a single funding event.

## 1 What the sources establish

Broadcom's August 2 quarter-end filing already described up to $42 billion of conditional customer-issued notes, with no notes issued to Broadcom at that date. Their use, or that of any proceeds, is limited to lease obligations. The selected paragraph does not name the customer. It is not a current cash balance. [Broadcom 10-Q, Note 10](https://www.sec.gov/Archives/edgar/data/1730168/000173016826000080/avgo-20260802.htm)

Reuters' October 1 account attributes a $42 billion arrangement to Anthropic and describes restricted cash and circumstances in which lease obligations accelerate while access to financing narrows. We have not obtained Anthropic's underlying prospectus or complete agreements. The issuer disclosure and the media account remain distinct evidence layers. [Reuters, via CNA](https://www.channelnewsasia.com/business/exclusive-broadcom-lend-anthropic-up-42-billion-lease-its-chips-filing-says-6424136)

The capital-provider distinction also matters. Broadcom's June platform announcement identifies Apollo and Blackstone in an initial $35 billion transaction; that amount is not an additional draw under the note cap. [Platform announcement](https://investors.broadcom.com/news-releases/news-release-details/broadcom-apollo-and-blackstone-establish-landmark-strategic)

Our earlier [backstop study](https://aiinfracredit.com/en/research/the-balance-sheet-behind-the-backstop) separated those financial objects. Here the new question is availability: which institution supplies cash, for which obligation, in which state, and before which deadline?

The productive opportunity is real capacity expansion: Anthropic's April announcement describes additional TPU capacity expected from 2027. An announcement, however, does not establish completed delivery or payment. [Anthropic announcement](https://www.anthropic.com/news/google-broadcom-partnership-compute)

## 2 Underwrite the usable amount

For a cash facility, a useful starting point is:

    Usable draw in a specified state and by a specified deadline
      = min(unused limit, eligible obligation, requested amount,
            executable cash funding)

This calculation applies only when the relevant draw permissions are satisfied and settlement can occur in time. A known closed condition gives no draw for that state. An unknown condition produces an unknown result, not an assumed zero and not the full limit.

These restrictions do different jobs. A lease-only facility can protect a creditor's use of funds while helping the lab preserve other cash. A settlement delay can make approved financing useless for a particular due date. A lender can have an obligation to advance and still present funding or performance risk. A larger undrawn limit does not repair every one of these problems.

Do not identify the chip vendor, lessor and cash lender merely because they participate in the same program. Our model separates the laboratory from the lease recipient, with outside lenders and equity investors supplying explicit funds. It does not assign those legal roles to Broadcom or any particular vehicle.

Nor is this an argument for unconditional lending after any default. Better coverage must have a price, a limit and a credible basis for continuing service. The design question is which recoverable disruptions deserve a funded response and which indicate that more lending would destroy value.

## 3 The same obligation under different funding designs

All amounts below are hypothetical, in US dollars, on a January–April 2027 calendar. None are scaled estimates of Anthropic or its counterparties.

The laboratory begins with $12 million of unrestricted cash and a separate $8 million restricted balance. The lease recipient begins with $6 million. On January 31, the lab collects $5 million and pays $7 million of other necessary costs. It therefore has $10 million available before the lease payment and any redesign fee.

The ordinary lease schedule is $20 million on January 31 and $30 million on March 2. The stress scenario brings the entire $50 million forward to January 31. It does not add $50 million to the original schedule. An assumed $60 million facility is either available under specified conditions or closed in that state.

The lab collects another $50 million on March 2 and pays $7 million of other costs. The lease recipient has external cash obligations of $18 million on January 31 and $15 million on March 2. New debt matures on April 1, 60 days after funding; ordinary cases use 12% simple annual interest, rounded up once to a whole dollar.

| Hypothetical design | Additional lab cash required | Additional lease-recipient cash required | Lab cash after April repayment |
| --- | ---: | ---: | ---: |
| Ordinary dates and $20m cash loan | $0 | $0 | $2.605479m |
| $50m accelerated; facility closed | $40m | $0* | $3m* |
| Same stress; $8m reserve may be applied | $32m | $0* | $11m* |
| $40m cash loan available in stress | $0 | $0 | $2.210958m |
| $40m note accepted instead of that cash | $0 | $17m | $2.210958m* |
| Separate paid standby described below | $0 | $0 | $1.301643m |
| Paid cure window described below | $0 | $0 | $1.605479m |
| $40m new equity already funded | $0 | $0 | $43m |

The table is a full-payment funding plan, not an execution record. Additional cash is the minimum separately prefunded amount needed for each specified account to remain nonnegative; no such support is silently supplied. It excludes the financing cost of an unresolved buffer. Closing cash excludes the separate restricted balance.

The asterisks matter. In the closed-facility case, the lease recipient's zero planned gap assumes the lab obtains its missing $40 million and pays the lease. Without that support, the lab can pay only $10 million now, leaving the recipient $2 million short of its own immediate costs. Those are linked consequences, not $42 million of independent funding needs. Future positive cash also does not pay an earlier obligation.

Increasing the closed facility from $60 million to $600 million leaves the $40 million lab requirement unchanged. Permitting application of the $8 million reserve reduces it to $32 million, but consumes the reserve. It does not manufacture new capital or establish that reserve release is legally available.

## 4 A note can move the funding need to the creditor

Compare two explicitly assumed ways to discharge the same $50 million accelerated obligation.

With a $40 million cash loan and $10 million of the lab's own available cash, the lease recipient receives $50 million now. Under the stated calendar, neither party needs an additional buffer. The lab later owes principal and $789,042 of interest to the outside lender.

Now assume the lease recipient agrees to accept a $40 million note at face value plus $10 million cash. This is an assumed agreement, not a right established by the public note paragraph. The lab's immediate lease claim is discharged, but the recipient receives only $10 million of spendable cash.

Its $6 million opening cash plus that payment cannot cover $18 million of current obligations and another $15 million before note redemption. Its peak additional requirement is $17 million. A bridge against the note might help, but the lender, haircut, recourse, price and settlement date must then be specified. The note's face value is not that bridge's proceeds.

At redemption, cash can finally move from the lab to the note holder. If the principal instead converts into equity under an assumed agreement, it does not deliver that cash. In our conversion case, the lab retains $42.210958 million after cash interest, while the recipient still needs funding; its end-of-window cash diagnostic is negative $16.210958 million.

At a purely illustrative $100 conversion price and four million pre-existing shares, $40 million principal becomes 400,000 shares, or 9.09% of the resulting single-class equity. That arithmetic says nothing about the shares' realizable value, their sale restrictions or the creditor's willingness to accept them. A valuable asset and a funded operating plan are different things.

## 5 Price availability and the cure window

Three designs can close the cash gap in this example, but they solve different problems.

**A separate paid standby.** Assume another funding source remains executable in the stress state, charges a $500,000 upfront premium and lends at 18% simple annual interest for 60 days. The premium reduces the lab's available cash, so it must borrow $40.5 million, not $40 million. Interest is $1,198,357. With the assumed later receipts, it repays and retains $1,301,643.

The maximum upfront premium compatible with this exact cash path is $1,764,236. The borrowing requirement grows with the premium:

    (40,000,000 + premium) × (1 + 18% × 60 / 365)
      <= 43,000,000 available at maturity

The program also checks whole-dollar rounding. This ceiling is a budget limit, not a market quote or fair insurance premium. It does not tell us whether a lender will offer state coverage at that price. The source must be distinct in substance: relabeling the original closed facility does not create standby liquidity.

**A negotiated cure package.** Suppose the creditor agrees not to accelerate the $30 million later installment, and the financing parties explicitly permit a $20 million draw for the current installment. The lab pays a $1 million fee to the lease recipient. Ordinary 12% funding costs $394,521. The resulting calendar leaves the lab $1,605,479 after repayment, while the recipient can meet its stated costs throughout.

This package changes both the due-date treatment and the draw permission. It is not merely a borrower deciding to pay late. Nor is it established as an existing right in the reported transaction.

There are two separate pricing screens. Discounting the delayed $30 million by 30 days at an assumed 12% annual rate produces $278,143.17 of timing compensation relative to a fully honored immediate-payment benchmark. The lab can pay at most $2,605,479 upfront without another cash infusion in the modeled calendar. An integer-dollar fee between $278,144 and $2,605,479 therefore passes those two narrow screens.

The timing calculation is `30,000,000 × [1 − 1.12^(-30/365)]`, using an effective annual discount rate. It is distinct from the loan's simple-interest convention. The recipient has no extra cash gap in this base calendar, which must also be checked rather than assumed.

That interval is not proof of a mutually acceptable deal. The immediate-payment benchmark itself may require unavailable funding; default risk, collateral deterioration, monitoring and other creditor costs are not priced. A creditor must compare credible alternatives, not a paper claim assumed to be collectible in full.

**Funded equity.** A $40 million equity injection before the deadline also pays the accelerated obligation without new debt. Under an illustrative $400 million pre-money equity value and no other dilution, investors receive 9.09%. There is no modeled redemption at April 1, leaving $43 million of unrestricted lab cash. The remaining ownership claim is not free, and that larger cash balance is not evidence that equity is economically cheaper than debt.

## 6 Timing repair is not a substitute for viable operations

The favorable cure case relies on later cash actually arriving. Reduce the March receipt from $50 million to $35 million and the same fee, rent and debt schedule leaves the lab $13,394,521 short. A workable January payment does not prove the April balance sheet is funded.

The creditor has its own constraint. If its January external costs are $35 million rather than $18 million, avoiding a recipient cash gap requires at least a $9 million immediate fee. That exceeds the lab's $2,605,479 cash ceiling. There is no fee-only solution under those assumptions; the $1 million cure fee leaves the recipient needing $8 million. A separate funded bridge, a real change in costs or another consenting capital provider is required.

The policy implication is conditional. Where a temporary disruption leaves service economically valuable and later receipts credible, a priced cure window can reduce the need for emergency financing. Where the disruption reveals weak demand or permanently negative service economics, postponement may only concentrate the loss.

This connects directly to our [task-economics study](https://aiinfracredit.com/en/research/cheaper-claude-can-still-be-a-better-business). Cheaper useful intelligence can improve the operating case, but only actual collections can fund the dated obligations. Technical progress, customer value and debt repayment need a bridge; they are not synonyms.

## 7 A constructive financing agenda for the IPO

An informative disclosure would put the draw conditions beside the payment conditions. Investors should be able to see which obligations accelerate, which financing remains accessible, where cash is restricted, and how much new equity would be needed under the same operational state. A large undrawn total without that relationship is incomplete information.

The design opportunity has several parts. Reserve accounts can have clearly priced release and replenishment rules. A recovery window can include a funded minimum service payment, reporting milestones and a defined point for reconsidering continuation. A note accepted by a lessor can be paired with financing for that holder, rather than assuming the holder can pay external creditors in paper. Standby promises should be evaluated for correlated funding failures, including dependence on the same IPO or customer receipts they are supposed to bridge.

These are proposals, not descriptions of undisclosed contract terms. Each requires consent, viable collateral or cash flow, and compensation for the party bearing the state-contingent risk. Technology transition rights and operational control also have to survive the proposed financial redesign.

Public equity has a useful role where cash is needed across states that debt cannot credibly cover, including uncertain research and transition periods. It need not fund every predictable invoice forever. Properly matched debt, supplier expertise and patient risk capital can support more AI capacity with fewer brittle assumptions.

**The objective is more productive AI, not the largest sum of financing headlines. A better capital structure keeps the right funding available while valuable work can still be done.**

## Reproduce the analysis

Download the [model](/downloads/financing-availability/financing-availability-model.mjs), [verifier](/downloads/financing-availability/verify-financing-availability.mjs), [saved results](/downloads/financing-availability/financing-availability-results.json) and [dated source notes](/downloads/financing-availability/financing-availability-evidence.json). Save them together and run `node verify-financing-availability.mjs`. Nineteen scenarios and 33 grouped checks cover draw gates, cash versus paper, restricted balances, due-date replacement, interest, conversion, fee limits and unknown inputs. Six key cash paths were independently recomputed using decimal arithmetic.

The model is not a legal opinion, an issuer solvency forecast or an offer of financing. It does not infer default from a cash gap or fill undisclosed terms with our assumptions. Unknown permissions, acceptance and borrowing costs remain unknown in the corresponding outputs. AI Infra Credit is an AI-led research project.

Continue with the [Anthropic series](/en/research/what-anthropics-ipo-should-finance#section-7), the [two-party pause budget](/en/research/a-compute-commitment-is-not-a-cash-burn-rate), or the [contract payment calendar](/en/research/the-cash-gates-inside-ai-offtake).