Chapter 2

How Qifu Turns Volume into Profit

Qifu earns money two ways that look similar on a screen and behave very differently underneath. In 2025, the fee-only side that carries no credit risk shrank, and the growth came from the side where Qifu funds loans on its own balance sheet and keeps the credit risk — financing income alone reached ¥8.6 billion (US$1.2 billion), 44.5% of revenue and the single largest line. That mix shift lifted the reported top line while compressing the profit per unit of volume management steers by, from 5.4% in mid-2025 to below 3% by the fourth quarter. Whether that compression is cyclical or structural is the heart of the earning-power question.

All figures in USD unless noted; historical renminbi (RMB) amounts are converted at the roughly RMB 6.99 = US$1.00 convenience rate implied in the FY2025 20-F, with the RMB original shown alongside headline figures. Take rates, margins, and mix shares are ratios and are currency-independent. Each citation links to the RMB source page.

Two engines, booked very differently

The split established in Business and Repricing — credit-driven (capital-heavy) versus platform (capital-light) — is not just a risk label; it changes how revenue lands on the income statement. Under credit-driven services, Qifu either funds the loan through consolidated trusts and its Fuzhou Microcredit licence (an on-balance-sheet loan) or guarantees an off-balance-sheet loan for a partner bank. On-balance-sheet lending books the borrower's full interest as financing income, gross; the credit losses on those same loans sit lower down as provisions. Guaranteed loans book revenue from releasing of guarantee liabilities. Either way, Qifu holds the credit risk [1].

Under platform services, Qifu passes the borrower to a bank — through its capital-light facilitation model or its Intelligent Credit Engine (ICE) referral channel — and collects a service fee. It takes no credit risk, and the fee is close to pure margin [2]. The consequence matters for anyone reading the growth rate: a renminbi of loan routed on-balance-sheet grosses up revenue far more than the same renminbi routed to a bank for a fee, even though the fee may drop more to the bottom line.

Source: FY2025 20-F, net revenue disaggregation; converted at the FY2025 convenience rate [3].

The mix is intensifying, not lightening

The popular framing of these platforms is a march toward capital-light: more fee income, less balance sheet, less risk. Qifu's 2025 numbers run the other way. Credit-driven revenue rose 19.3% to ¥14.0 billion (US$2.0 billion) and its share of revenue climbed back to 72.7%, while platform revenue fell 4.0% to ¥5.2 billion (US$0.7 billion) [4]. Inside credit-driven, financing income — the purest form of Qifu putting its own balance sheet to work — grew from 38.6% of revenue in 2024 to 44.5% in 2025 [5].

The balance sheet tells the same story from the asset side. Even as the total outstanding book shrank from ¥137.0 billion to ¥126.0 billion, on-balance-sheet loans — the ones Qifu funds and fully owns the losses on — grew from ¥31.9 billion to ¥42.9 billion, lifting their share of the book from 23.3% to 34.0% in a single year, and from 20.1% two years earlier [6].

Financing income as a share of revenue, and on-balance-sheet loans as a share of the outstanding book. Source: FY2025 20-F [7] [8].

By volume the shift is subtler but points the same way: credit-driven services were 55.9% of the ¥327.1 billion facilitated in 2025, up from 47.0% in 2024, while capital-light platform volume fell 15.4% [9]. Some of that is deliberate — management says it dials the mix by cycle, doing more capital-heavy in upturns "generating higher return and higher take rate," and preferring capital-light to offload risk in downturns [10]. Yet 2025 was plainly a downturn, and the mix went heavier, not lighter. The likely reason is defensive: as the April 2025 rules lifted the industry's risk bar, funding for the ICE and referral channels dried up, so volume that once went to partner banks for a fee was pulled onto Qifu's own book to keep it flowing. Referral (ICE) revenue held at ¥2.7 billion for the full year but collapsed to ¥99.7 million in the fourth quarter, down from ¥907.2 million a year earlier [11].

Take rate: the number that is compressing

Gross revenue as a share of facilitation volume — a crude take rate — actually rose over the three years, from 4.4% in 2023 to 5.9% in 2025. But that rise is largely the accounting artifact just described: moving loans on-balance-sheet grosses up revenue with interest income while the matching losses sit in provisions. Credit-driven services carry roughly twice the gross revenue take rate of platform services precisely because they book gross interest, not a net fee.

Segment revenue divided by segment loan-facilitation volume. Source: derived from FY2025 20-F revenue [12] and volume [13].

The take rate management actually steers by — a net measure after funding and risk — tells the opposite, and more important, story. It was 5.4% in the second quarter of 2025, up almost a point year-over-year [14]. By the fourth quarter it had fallen to 3.5% on a headline basis and, excluding one-off items, "slightly below 3%," with management now aiming only to "maintain a take rate of about 3%" if the regulatory environment holds [15]. A business whose per-unit economics move from above 5% toward 3% inside two quarters is repricing its own earning power in real time.

The profit line shows it plainly. Fourth-quarter 2025 net income was ¥1,016.1 million, down 47% from ¥1,912.7 million a year earlier, and the GAAP net income margin fell to 24.8% from 42.7% [16].

Management-stated quarterly take rate and GAAP net income margin. Sources: Q2 2025 call [17]; Q4 2025 call [18]; Q4 2025 results [19].

Where the compression goes: gross-up and provisions

Because more of the book sits on Qifu's balance sheet, both revenue and cost lines inflate together. The provision for loans receivable — the reserve against on-balance-sheet defaults — rose to ¥3.6 billion in 2025, 18.9% of revenue, from 16.2% in 2024, and the provision for contingent liabilities on guaranteed loans jumped to ¥1.7 billion from ¥0.5 billion [20]. This is the mechanical reason, first flagged in Business and Repricing, that record revenue and record operating cash flow in 2025 coincided with net income slightly down: the intensifying balance sheet grosses up the top line and the reserve build at the same time.

The asset quality behind those reserves has turned. The 90-day-plus delinquency rate rose to 2.71% at end-2025 from 2.09% a year earlier, which management attributes to macro uncertainty and the 2025 regulatory changes [21]. The leading edge was sharper still: the fourth-quarter Day-1 delinquency rate was 6.1% and the 30-day collection rate 84.1% [22].

The counter-case, and what would settle it

The bearish read — that a 3% take rate on a shrinking, riskier book is the new normal — is not the only one the evidence supports. Management frames capital-heavy lending as the higher-return option, and funding costs are a tailwind: they fell to 2.9% of revenue in 2025 from 4.0% in 2023 as deposit rates eased and Qifu diversified into asset-backed issuance [23]. The first quarter of 2026 offered a tentative inflection: excluding one-off items, the take rate improved sequentially, and FPD7 — an early-warning indicator for new loans — fell roughly 20% versus the prior quarter as tighter underwriting worked through [24]. And the franchise itself is intact: repeat borrowers still supplied the overwhelming majority of volume — 93.3% for full-year 2025 — so the compression is in pricing and risk, not in customer loss [25].

The read here: the take-rate compression is real and mostly regulatory-and-cycle-driven rather than a permanent impairment, but the shift onto Qifu's own balance sheet raises the stakes on being right about credit. A stable take rate near 3% on ¥327 billion of volume still supports meaningful profit; a take rate that keeps sliding while delinquencies climb would not. What would settle it, over the next few quarters: whether the net take rate holds at or above management's ~3% target, and whether the delinquency and FPD7 improvement seen early in 2026 persists as the tighter-underwritten vintages season. The provisioning question those reserves raise — conservatism or early warning — belongs to a dedicated look at credit and accounting quality.