Neobanks from debit to credit: which products, why, how AI underwrites them and what can go wrong

Why are neobanks expanding from debit to credit?

Because debit interchange is a thin, capped and legally fragile revenue line, and credit is where the money is. A US neobank that runs on a partner bank with under $10 billion in assets (affiliates included) earns uncapped interchange on its debit cards. The Federal Reserve's Regulation II data show what that exemption is worth: in 2024 the average exempt debit transaction paid $0.51 of interchange against $0.23 for a transaction on a covered issuer's card; on Visa dual-message debit the fee was 1.46% of the transaction value for exempt issuers and 0.44% for covered ones. That gap is the business model of US neobanks that run on small partner banks, Chime first among them. But it caps out. Interchange grows only with spend, it is paid in cents, and it depends on the partner bank staying small. Chime's S-1 says so in its risk factors: the company depends “in part on the maintenance of our bank partners' qualification for the small issuer exemption”, and its own growth “could itself put pressure on our bank partners' ability to qualify”.

Credit monetises the same customer several times over. Chime's 2025 10-K shows the shift inside one company: payments revenue, which is interchange, grew 18% to $1.50 billion, while platform-related revenue, the line that holds MyPay, SpotMe and Instant Loans, grew 73% to $686 million. In the second quarter of 2026 the gap widened again, with payments up 17% and platform-related revenue up 48%. At Dave the shift is complete: transaction-based revenue, mostly interchange, was $10.8 million in Q2 2026 against $160.0 million of service-based revenue, nearly all of it fees on ExtraCash advances. At Nubank, which started with a credit card, credit produced 41% of gross profit in Q2 2026 and fees only 25%.

The rate cycle explains the timing. Neobanks earned large float income on deposits while policy rates were high, and that income falls when rates fall; lending income is stickier because the spread is priced to risk. Monzo's FY2026 annual report shows the squeeze: its net interest margin fell to 3.7% from 4.1% as savings balances grew and the UK base rate came down 75 basis points during the year, while its lending balances grew 42% to £2.6 billion. In the US the Federal Reserve cut by 75 basis points over three meetings from September to December 2025, held for nine months and then raised the federal funds target by a quarter point to 3.75–4% on 16 September 2026, as Governor Waller recounted on 8 October 2026. A rate path that turns back up makes deposit spread less predictable and puts a premium on owning an asset that reprices. Revolut's 2025 accounts show the mix this produces at a fee-led neobank: card payments were its largest revenue stream at 22.2%, interest income 21.6%, and its credit book more than doubled to £2.2 billion.

Regulation pushed in the same direction. The CFPB's $8 credit-card late-fee rule, which would have stripped issuers of an estimated $10 billion a year, was vacated on 15 April 2025, and the CFPB's December 2025 advisory opinion put ‘covered’ earned-wage access outside Regulation Z. Meanwhile the debit cap itself became uncertain: a North Dakota federal court vacated Regulation II in August 2025 and stayed its own order while the Fed appeals, so the rule that makes exempt interchange valuable is now being argued in the Eighth Circuit. A neobank that lives on interchange is exposed to that case in both directions; one that earns interest is less exposed.

Which credit products have neobanks launched, and when?

The products climb a ladder. Each step moves more credit risk onto someone's balance sheet and more of the revenue from interchange to fees and interest. US neobanks on partner banks started with the lowest rungs, where the exposure is a few hundred dollars and days long; licensed neobanks in Brazil and Europe started higher or went straight to cards and loans.

The table lists the credit products on the record for the neobanks this site profiles, plus SoFi as the chartered precedent. Dates and figures are as the companies state them in their filings, results and releases; where a launch year is not in a primary source the cell says so.

United States: from fee-free overdraft to instalment loans

Chime is the clearest case of the ladder. SpotMe, fee-free overdraft, reached full launch in 2019 and members had drawn $43.3 billion through it by 31 March 2025. Credit Builder followed in 2020: a secured Visa card with no annual fee, no interest, no late fee and no minimum deposit, which Chime's S-1 calls “the first secured credit card of its kind”. MyPay, earned-wage access of up to $500, reached full launch in July 2024 and had paid out $8.8 billion by March 2025; in Q2 2026 alone it originated $4.5 billion at a 0.9% loss rate. Instant Loans came in 2025: up to $500 repaid in equal instalments over three months at $5 for every $100 borrowed. Q2 2026 originations were $300 million, and Chime expects the product to exit Q3 2026 at a run rate above $100 million of revenue. The 10-K names “unsecured credit cards” first among the products it may add. On 8 September 2026 Chime agreed to buy its partner, Stride Bank, for $590 million, saying ownership will let it “more efficiently expand its fast-growing lending business”.

Varo, the one US consumer neobank with its own national charter, climbed faster: Varo Advance (October 2020, up to $100 at launch), the Varo Believe secured card with no deposit and no APR (February 2021) and the Varo Line of Credit (March 2024). Advance and the line produced $547 million of lending volume in 2025. Dave never left the second rung and made it the business: ExtraCash is a discretionary overdraft of up to $500 issued by a partner bank, $2.3 billion of it originated in Q2 2026, with Dave Flex, a credit card, in test cohorts. MoneyLion pairs Instacash advances with Credit Builder Plus, a membership around secured Credit Builder Loans of $500 to $1,000 at 5.99% to 29.99%; total originations were $3.1 billion in 2024, its last year as a listed company.

Brazil and Europe: credit first, or credit on a licence

Nubank inverted the ladder. Its CFO's team said in May 2026 that Nubank “launched with a credit card in Brazil, unlike most digital banks that started with payments or savings and added credit later”, and that this forced it to manage credit cycles from the start. By 30 June 2026 its credit portfolio was $39.4 billion, up 37% in a year: $26 billion of cards, $10.3 billion of unsecured loans and $3.1 billion of secured loans. Revolut launched personal loans in Ireland, Romania and the US in 2022 and credit cards in Spain and Ireland in 2023; by the end of 2025 loans, cards and buy-now-pay-later were live in 13 countries and the book was £2.2 billion. Its UK bank, unrestricted since 11 March 2026, “paves the way for a wider range of services in the future, including lending”, and its US charter application aims at “full-scale lending, credit cards”. Monzo launched Flex, a pay-later credit line at 29% APR representative, on 16 September 2021 and now runs a £2.6 billion lending book.

The other European neobanks took narrower paths. Starling lent through the government's Bounce Back Loan Scheme and then into buy-to-let mortgages through Fleet, so that mortgages were 90% of gross lending by March 2025; it took a £28.2 million provision after removing the government guarantee from Bounce Back loans that may not have complied with it. N26 offers N26 Credit personal loans of €1,000 to €25,000 and since July 2017 has routed some of them to the German marketplace auxmoney, which lends where N26 does not. bunq offers deposit accounts and a secured credit card in the Netherlands; the OCC turned down its plan to make unsecured credit cards the core US product. OakNorth is the exception: a lender to lower mid-market businesses for its whole decade, more than £15.1 billion of it, with £2.8 billion of originations in 2025 and cumulative principal losses of 0.045% over ten years.

How do interchange and credit economics compare at neobanks?

Side by side, the disclosed numbers show interchange as the base of the model and credit as the growth line, at a higher loss cost. The table puts the two halves of each company's revenue next to each other for the latest period it reports.

Three things follow. First, short liquidity products are high-margin when the model works: Chime's MyPay made $73 million of transaction profit in Q2 2026 on $4.5 billion of originations, and Dave's ExtraCash earned a monetisation rate net of losses of 4.8% while its 28-day past-due rate fell to 2.12%. Second, losses arrive with the product, not after it: Chime's transaction and risk losses nearly doubled from $219.7 million to $407.3 million in the year MyPay scaled and Instant Loans launched, and its allowance on MyPay receivables it holds rose from $30.6 million to $76.6 million. Third, the margin on balance-sheet credit is a function of funding. SoFi, with a charter, funds over 90% of its liabilities with deposits and puts the saving against warehouse lines at about $712.6 million a year; its net interest margin was 5.98% in Q2 2026. Nubank's NIM was 22.9%, because its book is mostly cards and unsecured loans priced for Brazilian risk, and its cost of credit was $1.7 billion in the quarter.

What is not on the record: none of the US partner-bank neobanks discloses the interchange rate it earns per transaction, the share of interchange it pays its partner bank, or the loss rate by product beyond the headline figures above. Chime's release on the Stride purchase says ownership will bring “sponsor bank fee savings” and “a significantly lower cost of funds” worth more than $100 million of net synergies, but it does not say how owning a bank will affect the small-issuer interchange exemption its S-1 relies on.

How do neobanks underwrite credit with AI?

They underwrite on the account itself. A neobank sees every paycheck, bill and card swipe that runs through its app, and the credit products above are sized to that data rather than to a credit-bureau score. Dave's 10-K is the most explicit: CashAI analyses “cash flow data from Members' primary bank accounts”, hundreds of data points on income, spending and transaction history, and sets the ExtraCash limit in a process that “requires no credit check and does not rely on FICO or credit bureau data”. It has learned from more than 180 million ExtraCash originations, and because ExtraCash is repaid in about 11 days the model gets outcomes back fast. Version 6, with more than 700 features, began rolling out in Q2 2026. Chime's MyPay limit is pay-based, its Instant Loans pre-approve direct-deposit members, and MoneyLion's 10-K says the Instacash limit is “primarily based on a percentage of income or other recurring deposit amounts detected through the linked bank account”, without a credit check.

The larger lenders have moved to transformer models that read the sequence of a customer's events rather than hand-built features. Nubank's nuFormer is in production for credit cards in Brazil and Mexico and unsecured loans in Brazil, with SME and Colombian cards in testing (Q2 2026); Nubank says it delivered a 70% reduction in risk for an equivalent population compared with earlier models, and that write-offs stayed at 2.8% to 2.9% of the book in Q4 2025. Among the lending fintechs, Affirm put a transformer underwriting model live at US checkout on 17 September 2026 and reported 3.4% more completed purchases against a control group, including approvals for applicants with no FICO score. Affirm, like the lenders profiled under /fintech, shows where neobank underwriting is heading: the model is the product, and the bank is a funding and compliance layer.

Open-banking data widens the input beyond the app's own account. MoneyLion's model works from linked external accounts by design: its 10-K counts about 27.1 million bank accounts linked since inception and “billions of transactional data points each day”, and Dave's CashAI reads a member's primary bank account wherever it is held. What is not on the record is how much of any neobank's credit decision rests on data from other banks rather than its own ledger; none of the filings reviewed splits it out. The table below is computed from this site's neobank profiles: every profiled neobank whose record names an AI use in credit, underwriting, lending or loans, with the status recorded on its profile. It updates when a profile does.

The governance layer depends on the licence. Where a partner bank is the legal lender, as at Chime, Dave and MoneyLion, the bank must manage the fintech's model as a third-party risk under the interagency guidance the OCC, Fed and FDIC issued on 6 June 2023 (OCC Bulletin 2023-17), and model validation runs under the bank's own model risk framework, now the April 2026 interagency guidance (OCC Bulletin 2026-13, SR 26-2; see our SR 11-7 guide). Where the neobank is the bank, as at Varo, Nubank in Mexico and SoFi, the examiner reads the model directly. Either way the lender must give specific reasons for a decline under Regulation B section 1002.9, which a 700-feature or transformer model has to be able to produce, and must not discriminate; the CFPB's April 2026 Regulation B rule removed disparate-impact liability from the federal regulation from 21 July 2026 but did not touch that notice duty. The AI credit underwriting hub covers the bank side of the same rules.

Which rules shape the move from debit to credit?

Six items decide the economics and the compliance cost of the shift. The table gives each with its status as of October 2026; the linked rows are tracked documents on this site.

Earned-wage access is the clearest example of rules pulling in opposite directions. Federally, the CFPB's advisory opinion effective 23 December 2025 says covered EWA does not involve the offering of credit under Regulation Z and withdrew the 2024 proposal that would have treated many of these products as loans. At state level, American Banker counted six new EWA laws in 2025, doubling the number of states with one, and Connecticut's, in force since 1 October 2025, classifies EWA as a small-dollar loan, caps advances at $750 and charges at $4 per advance or $30 a month. A national product such as MyPay or Instacash therefore runs under several regimes at once, and Chime's 10-K notes that MyPay and Instant Loans “are not yet available in all 50 states”.

What are the risks of neobanks moving into credit?

Four risks, each visible in the filings: losses that rise with the product, funding that depends on a bank, partner banks that can be lost or constrained, and capital that regulators judge against the credit plan. The table gives the evidence.

The bunq decision is the sharpest statement of what a regulator expects from a neobank that wants to lend. In Corporate Decision #1384 of 4 August 2026 the OCC noted that bunq's Dutch bank offers deposit accounts and secured credit cards, that the proposed US bank's core credit product was unsecured cards in which its management had no experience, and that “the proposed allowance for credit losses is below other credit card banks”. An AI underwriting claim does not substitute for loss assumptions that a supervisor finds credible against peers. Nubank is the counter-example of a neobank that has priced risk through a cycle: its CFO's colleagues describe 17 generations of limit-increase models and a rule to “expect the future to be worse than the past”, and its Q2 2026 release attributes a 35-basis-point rise in 90+ NPLs to seasonal migration and deliberate moves into riskier segments.

The partner-bank model has its own failure modes. When the partner is the legal lender, the neobank's growth can push the bank over the $10 billion line that keeps its interchange uncapped, and the bank's supervisor, not the neobank, decides how much model risk is acceptable. That is why the 2026 licensing wave matters: Chime agreed to buy Stride, with closing expected in the first half of 2027, and the lending fintechs are following the same route. A charter brings deposit funding of the kind that lets SoFi fund itself 156 basis points cheaper than on warehouse lines, and it brings capital requirements and direct examination of every model that sets a limit.

What should you watch next in neobank credit?

• The Eighth Circuit's ruling on Regulation II: a restored cap is the status quo; a rewritten, lower cap would narrow the exempt-issuer premium shown by the Fed's $0.51 vs $0.23 data.
• Chime's Stride closing, expected in the first half of 2027 subject to OCC and Federal Reserve approval, and whether Chime launches the unsecured card its 10-K lists first among new products.
• Loss rates as liquidity products scale: Chime's MyPay at 0.9% and Instant Loans, Dave's 28-day past-due rate at 2.12% under CashAI v6, Monzo's 9.0% total loan loss rate.
• Revolut's first UK credit products on its own licence and its US bank, which its 2025 report says will unlock “full-scale lending, credit cards”.
• State EWA laws after Connecticut, and whether the CFPB's December 2025 opinion survives a change of leadership.
• Disclosure of model performance: few neobanks publish approval or loss rates by model version; Dave and Nubank are the exceptions on this record.

The neobank profiles linked above carry the dated record for each company, and the neobank hub lists the other pages in this series.

1. Chime Financial, Inc. Registration Statement on Form S-1 — Chime Financial via SEC EDGAR, May 13, 2025 (sec.gov)
2. Chime Financial, Inc. Annual Report on Form 10-K for 2025 — Chime Financial via SEC EDGAR, Mar 6, 2026 (sec.gov)
3. Chime Reports Second Quarter 2026 Financial Results (Exhibit 99.1) — Chime Financial via SEC EDGAR, Aug 5, 2026 (sec.gov)
4. Chime Announces Agreement to Acquire Stride Bank — Chime, Sep 8, 2026 (investors.chime.com)
5. Nu Holdings Ltd. Reports Second Quarter 2026 Financial Results — Nu, Aug 13, 2026 (nu.com)
6. Nubank's AI-powered credit model extends access while maintaining portfolio quality — Nu, May 7, 2026 (nu.com)
7. Revolut Group Holdings Ltd, Annual Report 2025 — Revolut, Mar 12, 2026 (assets.revolut.com)
8. Revolut Launches UK Bank — Revolut, Mar 11, 2026 (revolut.com)
9. Monzo Annual Report and Accounts 2026 — Monzo, May 19, 2026 (monzo.com)
10. Introducing Monzo Flex – a better way to pay later — Monzo, Sep 16, 2021 (monzo.com)
11. Starling Reports Strong Revenue and Invests for Growth (FY25 results) — Starling, May 28, 2025 (starlingbank.com)
12. Introducing Varo Advance: Instant Cash Access up to $100 — Varo Bank via PR Newswire, Oct 16, 2020 (prnewswire.com)
13. Varo Bank Introduces Varo Believe, a New and Better Way for Millions of Americans to Build Credit — Varo Bank via PR Newswire, Feb 25, 2021 (prnewswire.com)
14. Varo Bank Raises $123.9 Million to Fuel New Stage of Growth — PYMNTS, Feb 2, 2026 (pymnts.com)
15. Dave Inc. Annual Report on Form 10-K for the year ended December 31, 2025 — Dave Inc. via SEC EDGAR, Mar 2, 2026 (sec.gov)
16. Dave Reports Second Quarter 2026 Financial Results — Dave Inc. via PR Newswire, Aug 5, 2026 (prnewswire.com)
17. MoneyLion Inc. Annual Report on Form 10-K for 2024 — MoneyLion Inc. via SEC EDGAR, Feb 25, 2025 (sec.gov)
18. N26 Credit partners with auxmoney to provide more loans — N26, Jul 31, 2017 (n26.com)
19. Corporate Decision #1384: Denial of application to charter bunq US Bank, National Association — Office of the Comptroller of the Currency, Aug 4, 2026 (occ.gov)
20. OakNorth 2025 Results: £223M Profit, 33% Origination Growth — OakNorth, Mar 17, 2026 (oaknorth.co.uk)
21. SoFi second-quarter 2026 earnings release (8-K Exhibit 99.1) — SoFi Technologies via SEC EDGAR, Jul 29, 2026 (sec.gov)
22. Affirm launches transformer-based machine learning model for real-time underwriting — Affirm, Sep 17, 2026 (investors.affirm.com)
23. Regulation II: Average Debit Card Interchange Fee by Payment Card Network (2024 data, retrieved 2026-10-16) — Board of Governors of the Federal Reserve System, Oct 16, 2026 (federalreserve.gov)
24. Federal Reserve files reply brief in Reg. II appeal — ABA Banking Journal, Mar 23, 2026 (bankingjournal.aba.com)
25. Texas judge throws out CFPB's credit card late fee rule — American Banker, Apr 15, 2025 (americanbanker.com)
26. Truth in Lending (Regulation Z); Non-application to Earned Wage Access Products (advisory opinion) — Consumer Financial Protection Bureau, Dec 23, 2025 (files.consumerfinance.gov)
27. How Connecticut became a battleground for earned wage access — American Banker, Oct 20, 2025 (americanbanker.com)
28. Third-Party Relationships: Interagency Guidance on Risk Management (OCC Bulletin 2023-17) — Office of the Comptroller of the Currency, Jun 6, 2023 (occ.gov) · our page on OCC Bulletin 2023-17
29. Equal Credit Opportunity Act (Regulation B) final rule — Consumer Financial Protection Bureau via Federal Register, Apr 22, 2026 (federalregister.gov) · our page on Regulation B final rule on disparate impact (April 2026)
30. Speech by Governor Waller on the economic outlook — Board of Governors of the Federal Reserve System, Oct 8, 2026 (federalreserve.gov)