Yiren Digital Is Shrinking Its Lender to Buy an AI Cartoon Studio
A micro-cap credit shop reports a 46% revenue collapse, then spends the call explaining why the real story is a warrant to control an AI entertainment platform — with no Q&A.
YRD · Earnings Call · 2026-09-30
The quarter management wants you to look past
Yiren Digital's second-quarter 2026 print is, on its face, a business in retreat. Total net revenue fell to RMB 890 million — down 3% sequentially but a punishing 46% below a year ago — as loans facilitated dropped 29% quarter-over-quarter to RMB 6.3 billion and the performing-loan book shrank 30% to RMB 15.1 billion. The company posted a net loss of RMB 449.6 million, its second straight losing quarter, though it did narrow 9% sequentially. The loss itself is the interesting part, because it is not a credit-loss story. Provisions for contingent liabilities — the guarantee-book cost that dominated the bear case in prior quarters — actually fell 63% sequentially to RMB 233.3 million as the loan book wound down. What turned a near-breakeven operating quarter into a nine-figure loss was a fresh RMB 502.8 million allowance for contract assets, receivables and "certain related party loan receivables," up from RMB 176.4 million in Q1. CFO William Hui was direct that this is an accounting judgment, not cash out the door:So the headline loss conceals two offsetting forces: an improving core credit environment (31–60 day delinquency fell to 2.0% from 2.7%; 61–90 day to 2.4% from 3.2%; repeat borrowers hit a record 82% of volume) and a non-cash write-down on legacy and related-party paper. CEO Ning Tang even said credit conditions recovered "more quickly than we had anticipated." Which raises the question that frames everything else: if credit is healing, why is the book shrinking 30%?Excluding the allowance for the long-aged legacy receivables and related party loan receivable, adjusted operating income turned around to approximately breakeven this quarter.
The deliberate dismantling of a risk-taking lender
The answer is that Yiren is choosing to shrink. The most striking new language in the quarter is the pivot to an asset-light model — a keyword that jumps to the very top of the company's own lexicon this period after barely registering before. Ning framed it as a strategic fork, not a cyclical one:In plain terms: instead of lending its own capital against borrowers, Yiren wants to be a traffic-and-technology middleman — using its AI-driven customer acquisition capabilities to match qualified borrowers to third-party funding partners and collect a technology fee. That shows up in the numbers as a modest technology services line (up 5.9% year-over-year, down 2.3% sequentially) against a collapsing capital-intensive credit solutions line. It's a real business-model shift, and it is being executed while revenue falls. The internal-AI cost story is not new — management has sold it for over a year. Back in March, CFO Ka Hui noted the company had already banked RMB 80 million in direct AI cost savings in 2025. This quarter the framing matured from cost-cutting to operations: automated asset recovery drove the human-handling rate down from 45% to 24.9%, autonomous resolution on text agents climbed from ~60% to nearly 80%, and the fraud-detection systems were credited with avoiding roughly RMB 165 million in potential 2025 losses. That is the part of the AI narrative with hard operating evidence behind it.Rather than returning to the previous scale of our risk-taking business, we are increasingly focused on an asset-light and a referral model... reducing our reliance on capital and the credit cycle.
The warrant: from 'AI strategy' to an actual AI acquisition
Here is where the story turns company-unique. Yiren disclosed that between May and July it signed warrant agreements with the founders of four "AI-native" companies spanning education and entertainment, giving it the option to buy controlling stakes at pre-agreed prices. Then, days before the call, it served notice to exercise one of them: an interactive AI entertainment platform — user-generated role-playing and digital characters — with roughly 3.8 million users, 120,000 paying users, a presence across Southeast Asia, and, management stressed, already operating profitably. The framing is a deliberate evolution of a pitch that has been running for a while. On the March call Ning described the end-state as “an AI agent, AI native company, not only for credit and insurance subsectors, but also for more financial services subsectors and a few select industries.” — Ning Tang, Founder, Chairman and CEO · 2026-03-19 What is new is that the abstract AI ecosystem now has a name, a user count, and a warrant exercised. That is a genuine step-change from talk to transaction. It is also where skepticism is warranted. This is a $166 million market-cap company with RMB 1.7 billion of cash, down from RMB 2.45 billion just one quarter earlier, using options rather than cash to build an AI portfolio in verticals — interactive entertainment, education — with no adjacency to consumer lending beyond a shared user-acquisition engine. Management is candid that this is capital-efficiency masquerading as strategy: "a disciplined and capital-efficient way to build our AI application portfolio." Whether entertainment content is a moat or a distraction for a company that still has to fund customer obligations and credit volatility is an open question.What management quietly stopped emphasizing
The subtle tell is in what faded. For a year, internet insurance brokerage was the growth engine management led with — clients up 281% year-over-year to roughly 453,000, policies up 177%, gross written premiums RMB 838.9 million. This quarter those operating metrics were still trotted out, but the keywords tell a different story: Insurance Brokerage business and Internet insurance business were among the company's sharpest keyword decliners, while the narrower line-item insurance brokerage revenue rose even as it fell 23% sequentially on lower renewal rates. The star asset of 2025 has been demoted from headline to footnote — replaced, in the narrative and in keyword momentum, by AI warrants. Then there is the Q&A. Prior calls under the provided history routinely engaged analysts on credit risk, crypto, and the guarantee book. This one didn't:For a company mid-pivot, with a nine-figure related-party allowance, an Ethereum crypto position that produced a RMB 97.8 million fair-value loss (management expects a substantial Q3 reversal), and a freshly triggered acquisition, closing the floor to questions is the single most evasive thing on the tape.There will be no Q&A session on today's call.