Things to know 👀
Per the WSJ, Stripe is in talks to acquire OpenRouter, the AI model marketplace, at a valuation of roughly $10bn, and a deal could be announced soon (though talks could still collapse, and other bidders are circling per WSJ). For context: OpenRouter raised at a $1.3bn valuation in May.
OpenRouter is a single API in front of hundreds of AI models. Instead of integrating OpenAI, Anthropic, Google, and every open-weight model one by one, 5m+ developers plug into OpenRouter once, then compare prices, switch providers, and route each request to the cheapest model that can do the job (if Opus is too expensive for the task, send it to Gemini 3.5 Flash or DeepSeek). OpenRouter takes a cut of the token flow. For an AI startup, tokens are usually the biggest input cost, prices move constantly, and new models ship weekly. OpenRouter is where that spend gets managed.
🧠 Stripe’s growth is predicated on being the default financial stack for AI companies. Stripe grew TPV 34% last year to $1.9tn and added more volume in dollars than the year before. OpenAI and Anthropic run on Stripe; start an AI company today and you default to Stripe. Stripe always does this; spot the next category, buy its infrastructure early (Bridge for stablecoins at $1.1bn, Privy for wallets, and now, potentially, OpenRouter for AI).
🧠 Stripe wants to grow the GDP of the internet and their take rate on it too. Stripe’s take rate on payments for AI companies has helped them grow. But a take rate on model routing isn’t entirely dissimilar.
🧠 A near-8x markup in two months reads like classic bubble behavior. But consider the visibility Stripe has. It already collects OpenRouter's payments. It is bidding with the target's revenue trend in plain view.
🧠 The Metronome acquisition was a step towards doing more in AI. In December, Stripe bought Metronome (reported ~$1bn), the usage-based billing engine whose customers included OpenAI. AI is requiring bespoke and complex financial products, and Stripe wants to be best in class at all of those.
🧠 OpenRouter is the other half of that thesis. Metronome measures what an AI company sells. OpenRouter manages what an AI company buys. Routing is the control point of AI spend: it decides which model gets the token, at what price, from which provider. Stripe built a $159bn company taking basis points on payment flows. OpenRouter takes a cut of token flows. Same business model, new commodity.
🧠 Money in, money out. Stripe Payments collects an AI company's revenue (money in). Metronome meters the usage that drives it (money in). OpenRouter routes the tokens all of it runs on (money out). That's both sides of the P&L. Stripe would see what AI companies pay for intelligence and what they charge for it, in real time, across the sector.
🧠 CFO’s are in revolt over token costs. We’ve seen headlines from various companies like Uber burning their entire token budgets for a year in 4 months.
🧠 They’re not the only company betting on the “financing of AI jobs to be done.” Ramp is working toward the same destination from the spend side having launched a token spend dashboard for CFOs, and its own model router. Most PSPs are competing for the AI company's checkout. Stripe is competing for the new sources of commerce.
🧠 Imagine baking Stripe link into OpenRouter. Stripe has a wallet with 300m users. You’ll usually see a checkout ask you for an SMS code when you go to sign up for a new AI product. Those stored cards could be used to buy multiple AI models with ultra low friction. Although the best place for a wallet is still the coding harness. Cursor, Claude Code, Codex.
Revolut confirmed a $115bn valuation in a secondary share sale, per the WSJ, with employees cashing out at $2,017 a share. Coatue, Fidelity, T. Rowe Price and a16z are paying, and Nvidia’s venture arm joined the cap table in November.
The company earned it with a full UK banking license in March (five years after first applying), a US national bank charter application filed, 65m+ customers across 45+ countries with India, Mexico and Colombia in motion, $6bn of revenue and $2.3bn of pre-tax profit in 2025, and new licenses in the UAE, Australia, Mexico and Brazil.
🧠 A company that launched as a travel money card in 2015 is now worth more than Barclays or Deutsche Bank. I remember executives from both banks sneering at Revolut shortly after it launched: “Anyone can launch a card and lose money on FX.” How wrong they were. Revolut is the first private company Europe has ever produced worth over $100bn, and among private fintechs anywhere on earth, only Stripe ($159bn) sits above it.
🧠 Now do the maths on the price. $115bn on $2.3bn of pre-tax profit is roughly 50x. That is a software multiple attached to a licensed, deposit-taking bank. Investors are pricing Revolut as a tech company that happens to hold your money. That said, at current growth rates they’re trending to $9bn by the end of 2026. Then it's closer to 13x.
🧠 Even Nvidia wanted a piece of a bank. In 2026 the marginal growth dollar has one obvious home: AI, and this one went to a bank (albeit one that built a custom foundation model running on Nvidia hardware).
🧠 There is about to be one HECK of a Revolut Mafia. Employees cashing out at $2,017 a share means a wave of new founders with money and operating scars.
🧠 The US charter is the swing variable. If it is granted, Revolut gets a shot at the largest banking market on earth, and $115bn will look like the early price. Never bet against Revolut.
Marqeta is bringing stablecoins into everyday card spending, with Zero Hash providing the stablecoin infrastructure. Stablecoin-linked cards will be usable at tens of millions of merchants worldwide, and any Marqeta customer can embed stablecoin payments into new and existing card programs without rebuilding core systems or taking on new regulatory burden.
The context: Visa ran 130+ stablecoin-linked card programs across 50+ countries last year and expects that number to roughly double in 2026. Volume grew 319% to ~$5.2bn. Tiny next to Visa’s $14 trillion, but growing faster than anything else on the network. Consistently.
🧠 Marqeta has a lot of existing scaled customers who might want this feature. Think about Square, Western Union, DoorDash, Uber, Affirm, Klarna and Ramp. Some of them already have stablecoin products.
🧠 The demand is genuinely there. Companies like Kast let people in Argentina, Nigeria or the Philippines receive dollars and spend them directly. Netflix, Claude, flights. No currency switch, no local bank account required.
🧠 What I wonder about is whether Marqeta can compete on the economics. The breakout player in this category is Rain. $1.95bn valuation, reportedly ~38x volume growth in 2025, and a structural edge: as a Visa principal member operating from Puerto Rico, it issues directly and keeps interchange the traditional stack splits three ways (sponsor bank, program manager, processor). Marqeta in the US requires a bank underneath, and that bank takes a cut.
🧠 Classic innovators dilemma. Marqeta has distribution, thousands of live card programs that can switch this on tomorrow. Rain has economics, more interchange per swipe to fund rewards and win issuers.
Augustus raised a $180m Series B at a $1bn valuation, led by Tiger Global with Hummingbird, QED and the founders of Nubank, Ramp, Circle and Deel joining. The company is building a clearing bank that gives fintechs and banks outside the US direct access to dollar accounts, with settlement that runs 24/7 across traditional payment systems and blockchains.
Augustus received conditional approval for a US national bank charter from the OCC in May, only the eighth granted since 2010. It already runs regulated euro clearing through a Finnish entity, processing billions of euros a year for customers including Kraken, and the new money goes toward serving fintechs and banks in Latin America, Southeast Asia, the Middle East and Africa.
🧠 The target is correspondent banking. A fintech in (or serving) Manila or Lagos that wants dollar accounts today rents them through a chain of correspondent banks, where every intermediary adds cost, delay and a compliance team that can cut them off. Augustus wants to be the one US-chartered bank those customers connect to directly.
🧠 This is the core Citi and JP Morgan Franchise. Citi Services made $21bn last year at adjusted returns near 30%. J.P. Morgan Payments made a record $19.4bn. A lot of that comes from serving the largest Fintech companies and other banks.
🧠 We haven’t seen “Neobanks” go directly after this space before. Now we have two. Erebor and Augustus both are aiming at companies. With Erebor more for the companies who want good tech and not to be “de-risked,” Augustus seemingly more focussed on FI’s and Fintech.
🧠 This team has some experience. While the CEO is 25, execs include the former CEO of Greendot (for the bank), the Chief Compliance officer comes from Column Bank (having worked at JPM and HSBC before. So its an interesting mix of entrepreneurs driving the growth, and experience driving the regulated bits.
🧠 Conditional OCC approval and a new core banking system. Augustus has conditional OCC approval for a national bank charter, plus Marble, its own core banking system built for 24/7 settlement across Swift, ACH, SEPA and stablecoins.
A New York judge refused to throw out the state's fraud lawsuit against Zelle's parent company Early Warning Services. Attorney General Letitia James alleges fraudsters stole more than $1bn from consumers because “Zelle launched without basic protections,” and Justice Phaedra Perry-Bond found she had sufficiently alleged that Early Warning “prioritized accessibility, convenience, consumer adoption, and market dominance at the expense of consumer safety.”
Zelle is the largest P2P money transfer service in the US, bigger than Venmo or Cash App, and Early Warning is owned by seven of the largest banks in the country, BofA, Capital One, JPMorgan Chase, PNC, Truist, US Bank and Wells Fargo. The suit says safeguards were proposed internally four years before they shipped in 2023, after the CFPB and Congress started asking questions. A motion to dismiss only tests whether the claims are plausible, so this is permission for the case to proceed rather than a finding of liability. Zelle says it will appeal.
🧠 State AGs have replaced the CFPB. The CFPB filed this same case in December 2024 and dropped it three months later as the new administration wound down enforcement. James refiled under state law, and other state AGs have the same consumer protection statutes available.
🧠 Discovery comes next. Both sides can now demand internal documents and testimony about Zelle's fraud controls, including why safeguards proposed internally took four years to ship. Whatever surfaces will attach to the seven banks that own Early Warning as much as to Zelle itself.
🧠 While this is working through the courts consumers are losing. We’ve seen countless headlines of pensioners losing their entire life savings from scams. Cambodia and Myanmar are flooded with criminal enterprises who force 1000s of people to work under slavery conditions to scam vulnerable people in the west. This is a scar on our industry.
🧠 The banks have a shred of truth to their concern though. Much of the scam starts on social media or messaging. By the time the bank tries to detect or prevent it, often the user is coached to push through any warnings by the bank. In that case, what can the bank reasonably do? Where does personal responsibility stop?
Corgi, the YC-backed (S24) AI insurance startup, has reportedly closed a second Series B extension at a $4B valuation, per Forbes sources (the company declined to comment). The company completed a $108M Series A in January (~$630M post, per PitchBook), $160M Series B at $1.3B in early May, a $106M B1 at $2.6B three weeks later, and now a B2 at a reported $4B. The founders claimed $40M in run rate at the Series A; sources say it's tracking $450M by year-end.
🧠 The go-to-market here is spectacular. And I’m surprised nobody did it before. They have dedicated offerings for pre-seed/seed, Series A, and growth-stage companies. We had this in spend management; why not insurance?
🧠 This is venture pricing at AI-lab speed, applied to an insurer. If the $450M lands, $4B is under 9x forward run rate. Cheap for an AI company, rich for an insurer. The multiple assumes insurance premium behaves like ARR. That’s a big assumption.
🧠 It won't. Premium arrives with claims attached, and on liability lines the loss ratio reports on a multi-year lag. So this is either peak AI hype's next cautionary tale, or Corgi has genuinely cracked something. Possibly both at once. They’ve built an entire front-to-back modern platform. Their cost to serve is likely orders of magnitude lower.
🧠 The regulatory posture is more early Uber than insurance. Corgi writes much of its cover through a Risk Retention Group: members pool capital, claims get paid from the pool, and RRGs sit outside state guaranty funds, so if the pool runs dry, members eat the loss. TechCrunch's driest line was a doozy: "Perhaps it's not surprising, then, that Corgi wants to grow its coffers."
🧠 Which leaves the capital-structure question. Insurers usually feed growth with reinsurance and debt because both are cheaper than equity. Either a valuation doubling every month makes equity the cheapest capital going, or the balance sheet needs building faster than cheaper capital can move. Lets see how this does if some claims start going under.
PS. Can we take a minute for the name and the logo? If that isn't all-time great branding, I don't know what is.
If you enjoy this kind of content, I can guarantee you’ll love being in a room of 1,500 other folks who love to go deeper into where finance meets AI. That’s a huge theme for us at this year’s Nerdcon in San Diego on the 19th and 20th November. I’m bringing my audience, the operators, the people who read this newsletter. And it’s the perfect place to find your next hire, client, or just get inspired. Let’s make events awesome again.
4 Companies 💸
1. Nebex - Trade finance for the space economy
Nebex connects sovereign buyers like governments, suppliers like launch startups/contractors, and institutional capital. It says it finances transactions at the speed of the deal rather than the speed of government procurement, so suppliers get paid sooner while buyers run all of their government procurement DD. Payments and settlement run through the platform's own rails.
🧠 Three-sided marketplaces are hard. This one adds sovereign governments, the slowest and most relationship-driven buyers on Earth, to the mix. "A de-risked asset class" is a bold way to describe demand that's lumpy, political and wrapped in export controls like ITAR. But nations increasingly want sovereign space capability and suppliers are starved for cash between milestones, so a rail that finances the gap is genuinely needed. But can you get government procurement to move fast enough?
2. KredosAi - AI collections for lenders and telcos
Kredos messages overdue customers across SMS, RCS and other channels, using reinforcement learning from human feedback to work out which message, timing and channel actually gets someone to pay. It targets enterprises with high delinquency (telecom, auto lenders, financial services) and optimizes on payment outcomes. The platform claims deployment in weeks and payments pulled forward by up to five days.
🧠 20x ROI is the kind of number that makes me want to see the baseline. Optimizing on whether someone actually paid, rather than whether they clicked, is logic. But Symend and TrueAccord have run behavioral collections for years, and "reinforcement learning" might just be the current label for the same A/B machine.
3. Nudge - SEO for the ChatGPT shopping era
Nudge tracks where AI engines like ChatGPT, Perplexity and Gemini mention a brand's products when a shopper asks "what should I buy", flags the citation gaps, then generates content and product-page variations those engines are more likely to cite. It's aimed at e-commerce brands trying to show up inside the AI conversation instead of the Google results page.
🧠 There are a lot of people selling AEO/GEO and not a lot of evidence it's effective. We know clicks are coming from the LLMs, but the agencies and tools selling this optimization are quite light on evidence of uplift beyond traditional SEO.
4. RILLA Shield - Invoice-fraud protection for Aussie tradies
Forward any invoice, email, SMS or call to RILLA and it returns a Safe / Verify / High Risk verdict in about ten seconds, flagging payment-redirection scams, spoofed sender domains and altered bank details before money moves. It's built for Australian trades, real estate and logistics SMEs, the businesses that can't absorb a single fraudulent payment.
🧠 This is the most obviously Claude-generated website I’ve seen this week. Not to throw shade, but also, maybe try hide that a little? I’m wondering about the user flow here. Often, the scams that land are the ones you don't second-guess. The clever wedge is the signed evidence pack built for insurers: catching fraud is crowded, getting a claim paid is not.
Good Reads 📚
Shannon Kelly covers the 2026 policy flurry in AI: the White House National Policy Framework in March, June's Executive Order on voluntary reviews of covered frontier models, the bipartisan Great American AI Act draft (which would create a Center for AI Standards and Innovation under Commerce), the FSB's consultation on responsible AI adoption, and a patchwork of state laws.
The big change is SR26-2, the model risk management guidance issued April 17th that replaced SR11-7 after 15 years. It narrows the definition of a "model" to complex quantitative methods posing material financial risk, effectively ends check-box validation, and allows conditional approval of a replacement model before full validation completes.
Generative and agentic AI sit outside its scope, so Kelly sketches what their governance should borrow from MRM: continuous monitoring, red-team challenger agents with circuit-breaker power over a primary agent, and named humans accountable for the outcome. Her boldest idea is a self-regulatory organization that certifies vendor AI models under NDA, because no individual bank can validate an algorithm the vendor won't let them see.
🧠 Banks have been paralyzed by SR11-7 for a decade. Finally some reprieve and clarity on where Generative AI sits vs “quant methods” involving risk.
🧠 So your AI customer support chatbot is different. It still has some regulatory considerations, and how you get comfort on that is TBD. Could a model validation SRO fly? They’re very hard to build in practice. Or do we need a shared responsibility framework instead (or as well as)
Tweets of the week 🕊
That's all, folks. 👋
Remember, if you're enjoying this content, please do tell all your fintech friends to check it out and hit the subscribe button :)
Want more? I also run the Tokenized podcast and newsletter.
(1) All content and views expressed here are the authors' personal opinions and do not reflect the views of any of their employers or employees.
(2) All companies or assets mentioned by the author in which the author has a personal and/or financial interest are denoted with a *. None of the above constitutes investment advice, and you should seek independent advice before making any investment decisions.
(3) Any companies mentioned are top of mind and used for illustrative purposes only.
(4) A team of researchers has not rigorously fact-checked this. Please don't take it as gospel—strong opinions weakly held
(5) Citations may be missing, and I’ve done my best to cite, but I will always aim to update and correct the live version where possible. If I cited you and got the referencing wrong, please reach out
