How airlines borrow against loyalty programs

October 26: what a GPU lender can learn from airline cash

Airline loyalty financing pledges a defined stream of partner receipts and associated contractual rights. A bank paying for miles supplies cash before the airline fulfills every award; controlled accounts and payment priorities make part of those receipts available to lenders. That structure differs from borrowing the accounting balance of unredeemed miles. March 2021 credit and guaranty agreement

The comparison with hardware finance is useful at the cash-control layer. A GPU subsidiary can also hold customer contracts, equipment and pledged accounts, with lenders paid before excess cash reaches the operator. It becomes misleading when repeated card spending is treated as equivalent to a finite compute contract, or when loyalty’s unredeemed-award float is treated as GPU collateral. CoreWeave June financing terms

A recurring receipt is not a perpetual contract

CoreWeave’s August DDTL 5.5 announcement pairs an approximately five-year loan with initial customer contracts averaging three years. A lender underwriting that gap must assess renewal, re-leasing and residual equipment value. Repeated airline partner spending has its own concentration and renegotiation risks, but it does not provide a resale market for GPUs. DDTL 5.5 term comparison

A simple illustrative comparison makes the cash issue concrete. If either structure collects $100 but owes $35 of current fulfillment costs, $45 of debt service and $15 of required reserve funding, only $5 reaches equity. If receipts fall to $80 with those payments fixed, debt service consumes the $45 left after fulfillment and the reserve remains $15 short. Priority protects the lender first; it does not manufacture a distribution. Actual contractual definitions and ordering differ.

Use the hardware-financing hub’s collateral discussion for guarantees and cash trapping, the GPU depreciation model for replacement and resale, and the project cash waterfall for the operating sequence. The retained October 22 airline balance comparison below has its own September 30 accounting dates.

A card purchase and a mile sale follow different contracts

A $1,000 hotel purchase on an AAdvantage card settles among the cardholder, merchant, acquirer, network and issuer. The airline does not receive the hotel bill. It receives what the issuer owes under a commercial agreement outside the hotel payment. Keeping the two transactions apart prevents card spending from being mistaken for airline revenue.

In the model, 1,000 points produce $12 of partner cash, or 1.2 cents per point. That is an assumed price for the issuer relationship. It cannot be recovered by dividing all partner receipts by all miles issued: fixed payments, benefits, purchase volumes and other negotiated terms can enter the numerator, while some miles are earned through travel.

The issuer gains a reason for the cardholder to use and retain its card. The airline receives cash and takes on present services and future awards. The traveler receives the miles. A secured lender may receive an earlier claim on the bank payment than the airline’s shareholders do. Each party is buying a different benefit from the same relationship; none of those benefits is measured by the member’s point balance alone.

The issuer buys more than the award mile

The issuer buys access to an airline’s customer relationship: the name on the card, opportunities to acquire customers, placements in the travel experience and a reward people want to earn. Brand rights, exclusivity, marketing and benefits such as checked bags or lounge access sit beside the miles. The point is a convenient unit of settlement for a much larger bundle.

That creates bargaining power in both directions. An airline with desirable routes and engaged members gives the bank a differentiated card. A bank with a large, active portfolio gives the airline recurring purchases. The renewal bargain and the customer’s alternatives determine how durable those purchases are. A bank’s monthly payment does not, by itself, earn the program a platform multiple.

Aeroplan’s acquisition accounts identify C$1.212 billion received under long-term partner agreements from C$400 million of bank prepayments against future mile purchases. The latter funded future monthly settlements; the former was deferred over the agreements’ terms. Both supplied cash at the outset, but neither represents a recurring annual run rate. Air Canada accounting presentation, page 14.

Receipt, recognition, and redemption happen at different times

Cash, revenue and fulfillment belong to different periods. The model allocates the $12 receipt for 1,000 points 40% to current marketing and benefits and 60% to future awards: $4.80 of revenue and $7.20 of contract liability. That is an estimated allocation. Filings describe valuation methods; they do not disclose this hypothetical contract’s exact split.

American reported $6.2 billion of 2025 co-brand and other partner cash. Loyalty-liability additions were $4.445 billion, releases $3.935 billion, and the ending balance $10.564 billion. Marketing-services revenue was $3.511 billion; loyalty travel revenue, including prior-cohort releases, was $4.036 billion. These figures are not interchangeable.

A larger deferred balance can accompany a business receiving cash today, but that cash arrived in exchange for unfinished work. Conversely, a release of deferred revenue can lift reported travel revenue while the associated bank payment occurred years earlier. Receipts and redemptions need to be joined by issuance cohort before a period’s revenue growth can explain the cash engine.

Delta and United describe the same basic timing problem. A contract liability records consideration allocated to future performance. It is not a forecast of the cash cost of those awards and does not have the coupon and principal maturity of a conventional loan. Treating it as ordinary debt obscures the operating decision: which seats and partner services will satisfy it, and what else could those resources have earned?

Scroll horizontally to inspect the figure →

Scroll horizontally to inspect the figure →

Award delay creates float, with redemption risk

The interval between issuance and redemption supplies operating float. Delta and United say most newly issued miles are redeemed within two years. The model allocates the issued-mile cohort as follows: 25% redeemed in Year 0, 35% in Year 1, 20% in Year 2, 8% later and 12% expected never to be redeemed. Those buckets don’t supply a mean waiting time; dates within each bucket, and the open-ended tail, still matter.

When new partner purchases replenish cash faster than members redeem older awards, the program can look well funded even while its remaining obligation grows, which is why I’d also run a case that stops new purchases and makes the airline fulfill the existing cohorts from the resources already available. Runoff tests the promise. A steady-state margin can hide its cost.

Breakage estimates nonredemption rather than legal expiration. A no-expiration program can still estimate that some miles will never be used. That is a judgment about member behavior, not a cash reserve available to creditors. A stress case should allow both new purchases and redemption timing to change instead of reducing every cost in proportion to new mile sales.

Fulfillment economics also change with seat scarcity. An otherwise empty off-peak seat may require little incremental cash; a premium award can replace a valuable paying passenger. A program’s ability to set the miles required for an award helps manage that trade, but using it too aggressively can weaken the reward that persuaded the customer to choose the card. The sustainable margin includes members’ response to the award bargain.

Scroll horizontally to inspect the figure →

Expedia has a different form of customer-funded timing. Its deposit-adjusted cash bridge separates cash received before travel from the earnings available to value.

The lender controls a payment path; the airline maintains the franchise

American and United placed loyalty rights and payment paths in special-purpose structures, pledged collection and reserve accounts, and protected partner and intercompany agreements. These arrangements make a defined stream of receipts available as collateral. Their usefulness depends on the contracts continuing to produce money.

The original AAdvantage agreement’s coverage test uses defined collections and eligible cure amounts against maximum quarterly debt service. It is not an operating-profit margin. A program can therefore have substantial contractual coverage while leaving the airline with too little unrestricted cash after the financing and the delivery of awards. Calculate both figures. Original credit agreement, definitions and section 5.20.

The same agreement protects continuity of material partner contracts and IP licenses. Losing a crucial agreement without a qualifying replacement, or terminating an IP license, can have consequences beyond one quarter’s lower receipts. Termination, replacement and cure provisions can matter more than a percentage haircut to revenue: the contract may fail discontinuously while awards remain to be delivered. The cited terms describe the original 2021 document; current compliance and later amendments require their own records. Credit agreement, section 7.01.

The lender’s priority can strengthen its claim while increasing the parent’s liquidity pressure. The analysis has to reconcile the contractual account waterfall, the cash released to the airline and the cash still needed to fulfill awards. Deducting an invented operating-cost allowance before debt service would not reproduce a contract that gives debt service earlier priority. A lender can collect first while the airline must find cash elsewhere to keep the promise.

American’s September 2021 filing describes commercially reasonable efforts to direct at least 90% of specified quarterly receipts into the collections account. That is the documented routing obligation. It does not show the account’s actual receipts, distributions or compliance in a later quarter. Those observations require the period’s records. Filed structure.

American still has loyalty debt; United repaid it

American raised $10.0 billion in 2021 through two note series and a term loan; another $1.0 billion term loan followed in 2025. The named AAdvantage debt lines totaled $5.985 billion at September 30, 2026. The dated appendix below gives the remaining lines and the earlier comparison. Q3 filing.

United raised $6.8 billion in 2020: $3.8 billion of 6.50% notes and a $3.0 billion term loan. It prepaid the loan in July 2024 and redeemed the remaining $1.52 billion of notes in July 2025, ending that debt’s claim on MileagePlus assets.

The pandemic supplies a countercase to an uncomplicated resilience story. In American’s March 2021 pro forma presentation, 2020 partner and non-air cash fell to $2.908 billion from $3.861 billion in 2019. Cash paid for redemptions fell much further, from $2.637 billion to $0.749 billion. Operating cash consequently held up better than receipts. That preserved liquidity during the shock; higher cash conversion could reverse as members returned to traveling. A recovery in redemptions belongs in the downside case before the crisis-year cash margin can be capitalized. American presentation, slides 37–38.

Program value and debt capacity use different claims

American’s March 2021 presentation reported $49.521 billion of airline enterprise value, $14.182 billion of equity market capitalization, and $2.911 billion of 2019 pro forma AAdvantage adjusted EBITDA. An estimated 7× to 10× multiple gives gross program value of $20.4 billion to $29.1 billion.

That is 41% to 59% of matching-date airline enterprise value, but 1.44× to 2.05× its equity value. “Worth more than the airline” compares gross program enterprise value with distressed airline equity—different claims on the same system.

The DCF estimates $22.9 billion to $34.9 billion under its stated assumptions. I would read that range as a set of claims about future card spending, partner renewals, award costs and the airline’s ability to deliver, then ask which observation would invalidate each claim. A high multiple of current program cash can conceal low assumed spending on the airline platform that makes the cash possible. The program value also already belongs inside the airline’s enterprise value; adding both would count the same business twice. Loyalty-backed debt must then be allocated consistently when moving from enterprise to equity value.

An assumed $900 million of cash available under a lender’s definition, divided by an assumed 1.5× coverage requirement, permits $600 million of annual debt service. This illustration sizes the borrowing; the original AAdvantage covenant above has its own defined test. Interest, amortization and maturity still determine how much principal that service can support. Enterprise value prices the operating cash stream; debt capacity prices a bounded contractual claim on it. Neither number supplies the other.

Scroll horizontally to inspect the figure →

New partner cash can end before old awards are redeemed

Air Canada’s decision to develop a replacement for externally owned Aeroplan, followed by its purchase of the business, tests the idea that member balances alone form a self-sufficient asset. The acquisition brought the award obligation back alongside the airline that could supply the travel. Bank and network payments helped fund the transition, but their different accounting treatment shows why an upfront transaction should not be mistaken for a repeatable year of program cash. Acquisition accounting presentation.

An outgoing issuer can stop new cash before the last funded point is redeemed. The airline may own the loyalty brand while the bank owns or services the card portfolio. Points remain owed, and the airline controls much high-value award inventory.

The mechanical American stress test uses $6.2 billion of 2025 partner cash and a visible peak quarterly debt-service proxy, not contractual DSCR. A modeled 60% partner-cash reduction takes gross coverage from 4.4× to 1.7× and modeled program cash contribution below zero, while ending loyalty liability remains $7.82 billion.

The harder stress is a mismatch: partner purchases weaken now while redemptions from prior cohorts recover or accelerate. I would keep those award obligations alive, price the available seat capacity and trace cash through the actual pledged accounts. The chart’s 2.0× line and gross-coverage proxy help locate a modeled sensitivity; they do not reproduce a live covenant test, its cure amounts or its account definitions. A loan can remain contractually covered while the parent faces a growing need to fund delivery.

Scroll horizontally to inspect the figure →

Partner receipts are not attributable airline profit

No single percentage is defensible. American’s $6.2 billion of 2025 partner cash was 11.3% of operating revenue, but can be deferred. Its $3.511 billion of marketing revenue exceeded operating income, but revenue is not profit. Its $4.036 billion of award-travel revenue was recognized by releasing prior liabilities. No fresh bank cash was received for that recognition.

The 2021 presentation reported $2.911 billion of 2019 AAdvantage adjusted EBITDA and $3.145 billion of net operating cash. These management-defined, pro forma measures depend on intercompany mechanics. They aid financing analysis without isolating the consolidated profit American would lose without AAdvantage.

The model’s $6.38 of contribution per 1,000 points sold for $12 follows its stated costs. It identifies the inputs that drive the scenario. It does not calculate the profit attributable to the program. A defensible attribution would need a counterfactual airline without this arrangement: different card economics, fares, acquisition costs, benefits, award inventory and customer behavior. Intercompany prices move profit between entities without necessarily creating profit for the consolidated group. I would pay for incremental external cash that survives that comparison, rather than a presentation margin enlarged by where the group chose to book a transaction.

Scroll horizontally to inspect the figure →

The point model follows cash through the promise

The tables identify disclosed figures, arithmetic, assumed inputs, and management-defined measures. The main unknowns are the bank’s net point price, award timing and cost, and exit terms. Keeping those attached to the model lets each assumption be tested at the stage where cash changes hands.

The airline still has to deliver the flight

I would underwrite a loyalty program through its weakest renewal. Its largest member balance cannot substitute for that stress. The attractive case is recurring bank purchases, rewards members still want and an airline able to supply them at a cost the contracts leave room to bear. The adverse case is a lender collecting on time while those same promises consume the parent’s remaining cash. A platform multiple needs both the partner bargain and the award cohorts to survive that case. The next flight is part of the collateral’s economics, even when it is not the collateral itself.

Dated financing balances and the earlier comparison

The table compares original financing proceeds with later outstanding principal. Closing amounts and reporting-date balances are different measures; each row states both dates.

American had $5.985 billion of named AAdvantage principal and Delta had $3.150 billion of SkyMiles principal at September 30, 2026. United had repaid its original MileagePlus financing by July 7, 2025. The American and Delta filings and United repayment disclosure identify those different outcomes.

American’s subtotal is $2.750 billion of 2029 notes, $2.247 billion of 2021 term loans and $0.988 billion of 2025 term loans; its 2026 note series shows no remaining principal. Delta’s subtotal is $2.566 billion of notes plus $0.584 billion of term loans. Each sum uses the filing’s debt-category amounts before the aggregate adjustment for financing costs and discounts. They describe debt outstanding on the reporting date; subsequent cash movements between September 30 and this check need the intervening records.

Historical note: the earlier American comparison used approximately $6.543 billion at March 31, 2026. It precedes the September 30 balance used in the narrative and should not be read as an additional financing. The source filings and reporting dates explain the change.