Pay Later, Worry Later - Where BNPL’s Losses are Going

Buy now, pay later (BNPL) boomed during the COVID-19 pandemic as a way to split online purchases into a few equal payments, usually four and often interest-free. It has since moved into essentials. Lenders now pitch it for electricity and rent; many dentists, vets, and medical clinics offer it; and Intuit promotes “File Now, Pay Later” loans to TurboTax users who owe tax. Federal Reserve staff estimate that BNPL providers issued $156.7 billion of U.S. credit in 2025. On paper, the biggest lenders look healthy, with Klarna setting aside just 0.52% of its purchase volume for bad loans in the second quarter of 2026, down from 0.56% a year earlier. Yet a growing share of BNPL loans no longer stays with the lender that made them. This article looks at who is really borrowing, why lenders' numbers look so clean, and where the risk goes once the loans are sold.

Who is Really Borrowing?

BNPL use keeps growing. In 2025, 16% of U.S. adults used it, up from 10% in 2021. Clothing and electronics are still the most common purchases. However, a fifth of users bought groceries or food delivery with BNPL, and 8% used it for medical or veterinary procedures.

For many users, BNPL is more than just a convenient payment method. In the Fed's latest household survey, 29% of users said it was the only way they could afford their purchase, rising to 40% for those earning under $25,000. Among those who bought groceries with it, 45% said they had no other way to pay. Usage is also far higher among households with thin savings. Among adults who could not cover a $100 emergency, 31% had used BNPL, against 8% of those who could cover $2,000.

Most users still cite other reasons, such as spreading out payments, avoiding interest or convenience. BNPL, although a single product, can therefore be seen as two, serving different purposes for different groups: a budgeting tool for those in comfortable positions, and a last resort for those who are stretched thin. The second group shows up in lender data too, whereby nearly two-thirds of BNPL loans in 2022 went to subprime or deep subprime borrowers, and their credit card use tended to rise before their first BNPL loan, suggesting many resort to it as card headroom runs low. Their other debts are under strain as well. In Fitch's index of subprime car loans packaged into bonds, 6.13% were at least 60 days late in July 2026, close to January's record of 6.82%. Prime car loans, by contrast, run at around 0.4%. The strain sits with weaker borrowers, not across the board.

Why Do Klarna's Provisions Look So Low?

Provisions are money that a lender sets aside for loans it expects will not be repaid. In the second quarter, Klarna's provisions came to 0.52% of purchase volume, down from 0.56% a year earlier, while Affirm's rose to 1.58%, from 1.51%. Affirm sets aside about three times more per dollar, largely because it lends differently. About 90% of Klarna's 2025 U.S. credit was interest-free, against 29% at Affirm, and Klarna's loans last about 40 days on average. On longer installment loans, which are the closest match, expected losses look similar at around 3-4%.

The trend is actually where Klarna stands out, whereby, in the first half of 2026, its revenue grew 34.8% while its provisions grew 21.9%. Two things narrow that gap, with the first one being loan sales. These lift revenue, thereby Klarna booked $126 million of gains from selling loans in the half, against none a year earlier, and without them its revenue growth falls to 26.5%. The effect shows in its transaction margin, meaning revenue minus the direct costs of lending. Outside the U.S., transaction margin dollars rose from $283 million in Q1 to $358 million in Q2. That quarter, Klarna sold $967 million of German loans and expanded forward-flow deals, in which buyers agree to purchase new loans as they are made.

Second, some credit costs sit outside the provision line, so when loans Klarna holds for sale lose value, the loss is booked as a funding cost, and that line rose to $96 million from $44 million. Counting it, Klarna's credit costs grew by up to 34%, roughly in line with revenue. Two of the four reasons Klarna itself gave for lower provisions involve selling loans.

From the third quarter, the numbers get harder to read. New U.S. and German Fair Financing loans, Klarna's longer installment product, will be held at fair value, their estimated worth today. Expected losses are deducted inside a day-one gain rather than booked as provisions. The total loss is unchanged, but the ratio will fall for accounting reasons alone. Past periods are not restated, so it will no longer compare with Klarna's own history or with Affirm's.

Who is Buying BNPL Loans?

Klarna and Affirm fund themselves differently. Klarna is a bank, and consumer deposits made up 88% of its funding at June 30, 2026, whereas Affirm has no deposits and relies on securitizations, loan sales and bank credit lines. Both now sell large volumes of loans to outside investors, as does PayPal. Klarna has agreed to sell up to $26 billion of U.S. pay-in-4 loans to Nelnet. PayPal agreed to sell about $7 billion of its U.S. pay-in-4 loans to Blue Owl, and Affirm's buyers include PGIM, Sixth Street and Liberty Mutual Investments. In Europe, funds managed by KKR can buy up to €65 billion of PayPal's BNPL loans through March 2028.

The chain often runs further, whereby KKR buys through funds it manages for clients, and Liberty Mutual Investments is an insurer's investment arm. The risk on a BNPL loan can therefore end up several steps away from the shopper who took it out. About 53% of the loans Affirm manages now sit off its balance sheet, against about 28% at Klarna, and most of any loss on those loans falls on the buyers.

Not every sale moves all of the risk, however. Klarna's German forward flow is a clean transfer, since the buyer takes the first losses and Klarna keeps no stake. In its U.S. version, Klarna also lends to the buyer, though the buyer absorbs the first $310 million of losses, so Klarna's money is at risk only if losses exceed that cushion. In a synthetic securitization, the loans stay on Klarna's books and investors take only a middle slice of the risk.

Klarna puts its worst-case exposure across these structures at $1.37 billion at June 30, 2026, up from $840 million six months earlier. About two-thirds of that is loans it still holds while waiting to sell them. The figure is a ceiling, not an expected loss, but it shows the risk has not moved cleanly off Klarna's books. Affirm publishes no comparable figure.

Is BNPL Credit Getting Worse?

So far, nothing has broken, however on longer installment loans, late payments are drifting up at both firms. The share of Affirm's monthly installment loans at least 30 days late, excluding loans made through Peloton, rose 0.19% YoY to 2.5% in June. No June reading in the past five years was higher. The rise sits in loans that are only slightly late, while loans 90 or more days late held at 0.6%.

Klarna's closest product shows the same drift. Its U.S. Fair Financing loans made in Q1 2026 were about 3.0% at least 30 days late, against about 2.6% for loans made a year earlier. Klarna highlights an improvement on the quarter, but Q1 lending is the low point every year, so part of that is seasonal. These values are approximate, since Klarna publishes them only in chart form. Write-offs point the same way. Across both firms, the newest groups of loans sit at the top of past loss ranges, not the bottom, and this is where a turn would show first.

Borrower surveys add a final layer, whereby The Fed found that about one in four BNPL users paid late in 2025, roughly unchanged on the year but up from 15% in 2021. The rate was 40% among users earning under $25,000, against 11% among those earning $100,000 or more. Lenders' write-offs do not yet show this. Short loans may be repaid before strain turns into losses, or the losses may now sit with the buyers of those loans.

What Could Break?

The main risk is that loan buyers pull back. Their commitments are fixed-term. Klarna's German buyer is committed for 24 months from July 2026, its U.S. Fair Financing buyer for one to three years, and KKR's deal with PayPal runs to March 2028. If losses rise as these deals come up for renewal, buyers can demand better prices or walk away. Loans would then return to lenders' books just as they turn bad, and provisions would jump.

Short loans cut both ways. Klarna's whole book turns over about ten times a year, so lenders can tighten quickly, but losses also arrive quickly, and they would come first from stretched households. Affirm is more exposed to buyer demand than Klarna, since it has no deposits to fall back on. For now, that demand is strong; Affirm's latest securitization was more than four times oversubscribed, and its average funding cost fell to 5.8%.

The earliest warning should come from the buyers. The discount they demand on Klarna's Pay Later loans, booked in its funding costs, doubled to $46 million in the second quarter from $23 million a year earlier, while provisions rose just 11%. Part of that jump reflects selling more loans, and the figure includes the buyer's own funding cost. Still, it is the closest thing to a market price for Klarna's credit risk. Steeper discounts or smaller commitments at renewal would flag trouble well before it reaches any lender's provisions.

Regulation adds a separate test. In Europe, revised consumer credit rules apply from 20 November 2026, bringing most BNPL under fuller affordability checks. They will show whether tighter screening lowers losses or mainly slows lending.

Conclusion

BNPL lenders' own books look healthy, and so far the data show no clear break in credit. BNPL is also small next to the wider credit system. Small, however, is not the same as safe. Provisions only measure the risk lenders keep, and BNPL lenders are keeping less of it. A growing share now sits with the private credit funds and insurers that bought the loans, and with stretched households whose BNPL debt other lenders cannot see, since most of these loans never reach a credit report. From the third quarter, Klarna's switch to fair-value accounting will make its provision ratio even less informative. The better signals now are the price buyers pay for these loans, the terms they demand at renewal, and the financial health of the people who took them out.