Things to know 👀
Nubank went live in the US today through partner Lead Bank while it waits on its full OCC charter (conditional approval came in January). The Nu Account pays 3.5% APY, rising to 4.5% on balances up to $10k if you hold the credit card and use it three times a month. The credit card has no annual fee and 1.5% unlimited cashback, with a route to 2%. Transfers to Brazil, Mexico and Colombia are free from day one.
Nubank also launched a stablecoin-linked card. Nu Global is one multi-currency account across 35+ countries: every deposit converts to USDC or EURC, dollars earn 3.5%, euros 2.2%, a virtual Mastercard spends anywhere with no FX markup, transfers are free, and you can hold Bitcoin and Ethereum next to your balance. Europe and Latin America go first. Brazil, Mexico, Colombia and the US get wired in "over the months ahead." Nubank says 300m people live across borders, moved $800bn in 2025, and lose about 6% of it to fees and FX spreads.
🧠 The US product order took Nu from zero to 15m accounts in Mexico. High-yield account first, card second, remittances third in sequence.
🧠 The corridor is the customer. Mexicans in the US send around $63bn home a year. Nu already banks the receiving end. The US bank puts them on the sending end.
🧠 The cost base is the moat. Vélez told AFP Nu's cost to serve is 4% to 5% of a traditional bank's. That is how a 4.5% rate and a no-fee card fit inside a budget US capped at roughly $200m to $250m a year. (They’ve budgeted one point of group efficiency ratio, which is a super interesting model).
🧠 The US market entry graveyard for fintech is real. BBVA sold its US bank to PNC. N26 and Monzo left. Chime spent $635m on marketing last year to add 1.5m active users. Nu's bet is that word of mouth crosses a border: your cousin in Guadalajara already has the purple card.
🧠 They’re on a mission against complexity. US CEO Cris, herself having immigrated to the USA talked about the shock of having to re-KYC when applying for a mortgage with her existing bank. Nubank’s mission is to tackle complexity that shouldn’t exist, and customers loved that in LATAM. Why not the US too?
🧠 Nu Global feels like a direct shot at Revolut and Wise. It’s starting in Europe, multi-currency, and aimed at consumers.
🧠 The biggest Neobank in the world went stablecoin-first. A stablecoin account needs no banking licence in each country. A deposit account does. Nu gets to 35 countries on launch day with USDC, EURC and a card network
🧠 What we don't know. Which regulated entity holds your USDC, and where the 3.5% comes from (wheredoesyieldcomefrom.jpg). Neither is in their banks territory. Both of these questions are on my list for my Vélez interview.
🧠 Nu Global is also a probe. Nu gets to watch where traction lands, then double down and plant a purple flag.
🧠 My guess is Europe, the UK, is next. Nubank is arriving on Revolut's home turf. This is about to get spicy 🌶️
Muse is a personal AI agent like Instinct. You chat to it through a messaging interface, and it reads your email, books travel, negotiates for you, and pays through Stripe Link once you approve. If it buys the wrong thing, a refund guarantee underwritten by XCover (Cover Genius) covers it. Zuckerberg told Alex Heath the model is 100M free tokens a week, $20 or $100 a month for power users, and eventually "a very small cut of whatever the transaction is."
🧠The AI Agent for 2 billion users. This could be a REALLY big deal. Meta has all of the distribution to bake a personal assistant into WhatsApp and Instagram. It could become the default AI that manages cards, passwords, and user context for those people. Instinct showed what was possible. Meta is scaling it.
🧠 Payments and subscriptions are the business model. Free to start, a subscription if you use it a lot, and a cut of payments.
🧠Will merchants want Muse agents on their website, or will they start blocking it if the take rate is high or there are a lot of chargebacks? The first wave of agentic commerce had worse checkout conversion than existing websites. Most big merchants like Amazon blocked them. This is where the Stripe Link partnership can make a huge difference; they know a lot about conversion and what merchants need.
🧠 Single-use virtual cards ftw. Link mints a single-use virtual card scoped to the approved purchase, so Muse never sees your real card. Eligible purchases get price protection and damage or loss coverage up to $500 per claim, plus no-fee returns up to $250
🧠 Only 9% of consumers say they'd let an agent complete a purchase with payment (Accenture, June 2026). The other 91% need the refund guarantee more than they need a better model.
🧠 Insurance is the killer app for agentic commerce. Agents will buy the wrong flight. The question holding agentic commerce back has been who eats that loss, and now an insurer does.
🧠 Card networks solved this in the 90s with zero liability and chargebacks, and that's how card-not-present commerce took off. The difference then was that the user was involved. Visa and Mastercard will want to own agent purchase protection too. An insurtech got there first. But we’re still early.
🧠 Privacy is the pitch, and Meta hired BIG for it. Zuckerberg and Nat Friedman recruited Moxie Marlinspike, Signal's founder, to build a Confidential VM where "even Meta cannot see the content." It ships later this year. So your data lives in a cloud Meta can’t see. That’s COOL.
🧠 Instinct showed what happens without that. It stored passwords and card credentials in its own vault; testers revolted, and in the last two weeks it added Stripe Link and 1Password. Muse launched with both. Credit the partnerships folks at Stripe and 1Password for locking in the personal agents this early.
🧠 Where’s Google or Apple here? Google has the email, the calendar, the cards and the search history. Apple has the user trust. Neither has shipped a personal agent. Meta shipped first.
🧠 Most Neobanks, Apps and Merchants are not ready for this. Consumer agents are coming. They’re still early, but you need to get ready. When Muse AI or something like it becomes the consumer control center and your interface. The everything app. Is your service ready for Muse as the customer?
🧠 Is Meta a fintech now? Is all AI fintech? Those lines are blurring.
The company is paying $590M in cash for Stride Bank, which has held Chime deposits for seven years. Stride has $4.9bn of assets and a 26% ROE; so the price is roughly 1.5x tangible book, and the deal closes in H1 2027 pending OCC and Fed approval. Chime raised FY26 guidance the same day: revenue of $2.76-2.77bn, up 26-27%.
🧠 Uno reverse. The fintech built on top of the bank now owns the bank.
🧠 There's your ROI. Chime expects $100M+ of net "synergies" (God, I hate that word) from sponsor fees, cheaper funding, and more lending. That is a 20% uplift on its ~$485M 2026 adj. EBITDA guide.
🧠 Chime becomes a bank holding company in return for that 20%. The Fed will supervise Chime Financial, Inc., the whole company, alongside the OCC supervising the bank.
🧠 My first question was, how do they stay Durbin exempt? Banks under $10bn in assets get higher interchange revenue per swipe, and the $10bn exemption counts the issuer plus all of its affiliates, consolidated.
🧠 It turns out they're under the line. Chime, the holdco, has $2bn of assets. Stride has $4.9bn. Call it $6.5bn combined once the cash leaves. But Chime now owns the line, and it wrote this into the release: it will "keep its assets below $10 billion for the foreseeable future."
🧠 So The Bancorp stays as a partner too. Bancorp is already at $9.2bn, and Chime will keep working across both. Expect deposit sweeps, loan sales and a very disciplined CFO.
🧠 Chime owns one sponsor bank and still rents the other. The $100M comes from removing sponsor fees and cutting the cost of funds on lending products like MyPay and Instant Loans.
🧠 Chime is a payments company that now owns a bank it needs to keep small. Whether the market will ever fully understand them is another question I, and I'm sure they, continue to wrestle with!
Bloomberg reports Ramp is in early talks with investors to raise roughly $1bn of primary capital at a valuation of around $60bn. Ramp declined to comment and the terms could change. The company says it passed $1bn in annualized revenue in June, is free cash flow positive, has 70,000+ customers, and runs more than $200bn in annualized purchase volume. Meanwhile, they launched a Broadway musical starring Billy Porter. They are the marketing benchmark.
🧠 They almost certainly don't need the money. My guess is that investors keep pre-empting and Ramp is taking it while times are good.
🧠 Ramp was marked at $13bn in a March 2025 tender, then $16bn, $22.5bn and $32bn across 2025 before the $44bn round. Now $60? Woof.
🧠 Ramp has an AI story. They want to be the way to manage people, vendors and now, AI tokens. It already issues cards to AI agents with Visa and tracks token spend across providers, and no ERP has a field for that yet.
🧠 On the multiple: $60bn is around 40x annualised revenue if you use Sacra's $1.5bn estimate, nearer 60x on the $1bn+ Ramp has confirmed. Brex went for roughly 7.5x. Ramp deserves a higher number than Brex based on its growth. They’re growing 170% a year and throwing off cash. But 40x to 60x assumes it keeps growing like this for years.
🧠 The show is about Bill’s, starring guys named Bill. Because Bill Pay is too damn hard. It was common in the 50s and 60s for corporations to make musicals. Somewhere we went towards “new,” and we lost IRL. Real-life marketing is the new hotness.
An attacker sent lawful government information-demand emails using a genuine government domain that passed domain authentication. Affected customers were notified on Friday. What may have been disclosed ranges from name, date of birth, and home address to account statements, withdrawal records, and complete Bitcoin transaction history.
🧠 This is a highly sophisticated, believable attack. Successfully compromising a real government domain is a massive threat. These attackers likely knew the domain would pass verification and the internal process that would follow.
🧠 Revolut says it has alerted the government agency to the domain. You have to hope that government agency, whoever they are, is taking steps to prevent this from happening again, or the attack could be replicated.
🧠 I also imagine other institutions will now whitelist only specific government-actor email addresses, not full domains. We’re heading into a world where these attackers are well funded and have AI tools that can compromise more or less anything.
ChatGPT for Financial Services combines the GPT -6 Astra model with built-in financial data from the likes of Daloopa, PitchBook, LSEG News, and Crunchbase. It can pull external and internal data, create models and pitch decks in minutes
🧠 Most enterprises have a neutered version of AI. Co-pilot and ChatGPT Enterprise are the “better Google” experience. Some information, Q&A, summaries. But rarely doing work.
🧠 The packaging is the product. OpenAI is offering “zero data retention,” role-based access, connectors, and approved tools from Wall St-grade data partners. OpenAI can tweak the underlying model, not just the harness around it, and says its error rates are much lower than competitors’.
🧠 Rogo and Hebbia offer this today, at scale. Both companies reported over $50m ARR this year, and count clients like JP Morgan, Citi, Rothschilds, BlackRock, KKR, and Morgan Stanley.
🧠 Every spreadsheet is an opportunity to sell AI tokens. AI labs win when people burn more tokens. Doing knowledge work burns 10x or 100x more than basic queries. Recent OpenAI models have gotten really good at it too. Give it transcripts, voice notes, other presentations, data, and what comes out after 30 minutes is pretty good. I’m giving a keynote to a bank’s top 250 next week that would have taken me 3 to 4 days. I spent about 4 hours on it (and they’re very pleased with it).
🧠 I think AI for finance will become the meta for the entire AI sector next year. The second-highest AI spender outside of tech is financial services. It is the world's largest profit pool, and second-largest industry by market cap. The opportunity is too big to ignore.
Split Pay pays your whole rent on the due date and floats up to 50% of it for up to 30 days. It charges 2% of the bill plus $10 a month, with no interest and no late fees. Miss a repayment and you're out. Around 1m members, a quarter of them monthly users.
🧠 This is a really nice income smoothing product, and I hope it gets used that way rather than as stealth BNPL for rent.
🧠 2% feels punchy. Do the APR maths and it looks ugly, but compare it with a late rent fee, an overdraft or a payday loan and it holds up. (I have beef with APR as a measure, because the time dimension really maters).
🧠 Income smoothing is needed! CEO Andrew Borovsky (ex Cash App) found their average customer earned $90k, generated $2,100 a month in cashflow, and still lived paycheck to paycheck. The cause was the timing of bills.
🧠 Rent is due on the 1st and paychecks land twice a month. That twice-a-month thing is a super odd quirk of the US system and, like most American quirks, seems impossible to undo.
🧠 Businesses solved this ages ago with net-30 terms. Split Pay is net-30 for renters.
🧠 Split Pay spent its first two years building an underwriting model for people under 40. The company claims 50,000 variables, no FICO, and a 97.5% repayment rate on $350m of originations (self-reported). Levchin (an investor and Affirm CEO), built Affirm on the same bet, that AI underwriting grows the lendable segment without growing the losses.
🧠 Rent is about as good as cashflow data gets. Biggest bill most renters pay, lands monthly, can't be skipped. Underwrite that and you know more about someone's cashflow than FICO does.
🧠 The power of a great wedge product cannot be overstated for consumer finance apps. Get paid early built Chime. Commission free trading built Robinhood. Split Pay's own history is the wedge story. It started as a rent payment app (RentApp, then Visible). Splitting rent was a paid feature. Now it's the whole company.
🧠 Most consumer finance apps are lending businesses waiting for enough data to underwrite. Split Pay found a way to get paid while it collects the data.
4 Companies 💸
1. Locus - Billing for AI agent platforms
Locus lets companies building AI agents give each customer a pot of credits their agent can spend across tools like OpenAI, Apollo and Tavily. Instead of opening dozens of provider accounts, the company pays one usage bill and can add its own markup.
🧠 Another wallet for agents to pay for APIs isn't enough; we covered AIsa three weeks ago. Locus's smarter wedge is letting the company selling the agent make money on every call, like an app store margin for AI work. Its first obvious user is Locus itself: Locus Founder uses the wallet, API catalog and checkout to run businesses for users, which is useful dogfooding. No named external customer means the market still needs proving.
2. Woven Intelligence - Compliance software for the consultants behind RIAs
Woven helps compliance consultants manage the rules and paperwork for dozens or hundreds of wealth managers and broker-dealers. They build one compliance program, tweak what is different for each client, and use it to track deadlines, evidence, and annual reviews.
🧠 The customer is the consultant managing compliance for dozens of RIAs. Win one consultant and Woven can reach dozens of regulated firms, while helping that consultant add clients without hiring a person each time. Will consultants standardize the process they currently sell as expertise? In reality, the processes are always similar; the work is in the nuances of risk appetite, which will vary by customer profile.
3. Healia - Cheaper healthcare for dual-income families
Healia helps a family where both adults work compare the two health plans they can access. If the spouse's plan is cheaper, one employer can pay that plan's premiums and medical bills through Healia, while spending less than it would to cover the spouse itself.
🧠 I love this because each employer buys health coverage on its own, while dual-income families make the decision as one household. Healia lets an employer share the saving with the family, so both can pay less. This is exactly the sort of financial engineering healthcare needs.
4. Alton - The grown-up Corgi
Alton helps AI, crypto, fintech, and other hard-to-insure startups buy business insurance. A company fills in one application, then a broker finds carriers, checks the policy for exclusions and stays involved through renewals and claims.
🧠 Alton feels like the grown-up Corgi. Corgi’s constant fundraising is part of the product: more capital lets it sell more policies before it knows what the old ones really cost, which is a lovely flywheel until claims disappoint and funding gets hit at the same time. Alton leaves that risk with the carrier and collects recurring commission. Less upside, much less fragile.
Good Reads 📚
Nvidia is the central bank because controls the issuance of the scarce base asset: GPUs. Data-center financing creates money. Loans fund new GPU capacity, expanding the AI economy’s “productive base.” Neoclouds are commercial banks. CoreWeave, Nebius and others borrow capital, acquire Nvidia GPUs and extend compute capacity to AI labs. AI labs are factories. They turn compute into tokens and inference, the equivalent of economic output. Token consumption is GDP. More inference spend means more activity inside this parallel economy.
🧠 This is a clever analogy but a step too far, even for me. Commercial banks create deposits. Neoclouds do not create compute; they simply supply it.
🧠 My view was that it's more like a private Fannie Mae. A backstop lender for a specific asset class.
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.
(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
