Klarna's Quiet Q2, Casca's Loud Raise: Why Credit Is Reinventing Itself Without Waiting for Washington

Klarna's Q2 numbers landed on Monday, an AI-native loan origination platform banked $29 million on Tuesday, and Britain's proptech league table refuses to sit still. As of this week, the story of lending is being written outside the legacy stack.

The Klarna number that mattered was not the profit

On Monday 18 August, Klarna Group plc published its second quarter 2026 results, and headline watchers immediately fixated on the wrong line. Yes, gross merchandise volume hit $36.6 billion (up 18% year over year) and revenue reached $1.042 billion (up 27%), according to Klarna's own press release. Yes, adjusted operating income of $91 million was up a jaw-dropping 214% year over year. But the number that quietly rewrote the BNPL narrative was 0.52%.

That figure is Klarna's provision for credit losses as a share of GMV, down from 0.56% in the second quarter of 2025. Provisions in absolute terms rose only 11% while volume grew 18%. In other words, Klarna is lending more and losing less, at least on a rate basis. US Fair Financing delinquencies at 30-plus days fell 20 basis points quarter over quarter, per the company's supplementary disclosures.

For a sector that has spent five years being pitched as an unexploded ordnance of consumer debt, the operational reality looks steadily more mundane: an underwriting business that works, at scale. Investing.com and InsiderFinance both flagged that the market's initial disappointment came from a softer full-year outlook, not from any wobble in credit quality.

The Affirm comparison: same asset class, different weather

Contrast Klarna with Affirm, which reported first-ever GAAP net income of $103 million for fiscal Q3 2026 on GMV of $11.6 billion (up 35% year over year), as reported in company filings. Affirm's 30-plus day delinquencies, though, ran at 2.8% in March 2026, up 29 basis points year over year. Two BNPL flagships, two different credit trajectories, both profitable.

The takeaway (or, at least, the one worth screenshotting) is that BNPL has stopped being a single asset class in investor minds. It is now a set of underwriting philosophies, wearing similar branding. The Substack analysis by srimurthy this month made the point cleanly: BNPL sits at roughly 2% default rates versus 10% for revolving cards. The regulation Klarna and Affirm feared is not the regulation that shows up. Which brings us neatly to Washington.

The rule that will not be written

On 12 May 2025, the Consumer Financial Protection Bureau formally withdrew its 2024 BNPL interpretive rule. In June 2025, the Bureau confirmed it will not issue a revised version, calling the original 'procedurally defective' for applying open-end credit regulations to what are, in practice, closed-end instalment loans. As of this month, per Consumer Finance Monitor and American Banker's PaymentsSource desk, that is the settled position.

Regulatory absence is not the same as regulatory blessing, but for BNPL operators it has bought oxygen. The industry can now spend the second half of 2026 building product rather than briefing outside counsel. Whether Congress fills the gap is a different question. The Congressional Research Service flagged the 'myriad of BNPL issues' facing policymakers in March, and letters from Senator Reed's office are already stacking up.

The real action: AI-native loan origination raises real money

While BNPL held Washington's attention, the more consequential lending story was written in Series A term sheets.

This month, Casca, an AI-native loan origination platform, closed a $29 million Series A round, bringing total funding to $33 million, according to a press release distributed via PR Newswire. In parallel, Lama AI announced a $20 million Series A led by EJF Ventures, positioning itself to serve community and regional banks specifically, per Finextra's press coverage.

Both raises share a pitch: the legacy origination stack (think green-screen underwriting waterfalls layered on top of core systems that predate the smartphone) is being ripped out.

The replacement is cloud-native, model-first, and increasingly agentic. TIMVERO's 2026 sector review calls this shift the movement of AI in lending from 'pilot program' to 'operational baseline'. Pennant Technology's mid-year outlook says something similar in duller language.

Why community banks are the interesting battleground

Lama AI's decision to target community and regional banks matters because that is exactly the customer segment large lenders have been slowly conceding to fintechs. In February 2026, Mastercard integrated Small Business Credit Analytics into its Open Finance platform, allowing lenders to combine real-time sales data with analytics for SME credit decisions. That was a wholesale infrastructure play. The Lama AI raise is a distribution play against the same opportunity.

The Digilytics blog on AI in SME origination put the operational case bluntly: manual underwriting cannot scale to serve the long tail of SME borrowers profitably. What Casca and Lama sell, effectively, is the ability to say yes (or no) faster and with an audit trail regulators can actually inspect. That last part now matters more than it did last year, because the EU AI Act's high-risk system obligations went fully enforceable on 2 August 2026, forcing lenders operating in the bloc to formalise explainability, bias auditing, and human oversight. The regulatory sword and the commercial opportunity are the same sword.

Canada, quietly

Tracxn's July 2026 landscape report clocked 30 embedded lending startups in Canada alone, naming Flexiti, Financeit, PayBright, Credit App, and NetNow among the notables, with 13 having received funding and seven at Series A or beyond. Canada is not the market anyone leads with when pitching a global embedded lending narrative, but the density here is worth attention. Smaller domestic markets often mature underwriting models faster because the borrower universe is more concentrated and behavioural data richer per capita.

PropTech: the money is still in London, the drama is still in Dubai

Rotating to PropTech. The UK, per Beauhurst's 2026 top-100 review, is home to more than 845 active proptech companies collectively holding around £3.05 billion in equity funding, with £230.4 million raised in 2025. London took $340 million and Berlin $180 million of European PropTech capital over the trailing year, according to Qubit Capital's investment landscape review.

Habito, one of the UK's better-known digital mortgage brokers, still anchors the mortgage-tech segment. But the momentum, per Qubit's data, has shifted decisively toward AI-native platforms: proptech firms with an AI thesis grew 42% in 2025, versus 24% for non-AI peers.

Tokenised property: from hype to Dubai's Phase II

Meanwhile, tokenised real estate keeps promising to be next year's story. Dubai's Land Department moved its real estate tokenisation project into Phase II on 20 February 2026, enabling resale activity in a controlled secondary-market pilot. The rest of the world is watching. Forecasts from research houses tracked by SCN Soft and 4IRE Labs suggest global tokenised real estate could reach $3 trillion by 2030, representing 15% of real estate assets under management.

The Newmarket Pitch dataset covering August 2025 to July 2026 recorded $604.8 million in disclosed PropTech capital across 28 deals, and warned that 'smart-building and tokenised-asset pitches without revenue rarely work' in the current environment. Translation: investors will fund the vision, but only if the P&L shows up first.

What ties Thursday's news together

Three threads run through this week's lending, credit and PropTech news. First, BNPL is normalising as an underwriting business, not a regulatory drama. Second, the interesting money in credit is chasing AI-native origination for banks that cannot build it themselves. Third, PropTech has quietly stopped rewarding storytelling and started rewarding revenue. If you were waiting for a fintech reset, this is probably it: not a crash, just a maturity gate.