It usually starts with a customer request. Can you support RTP? FedNow? Push-to-card? Stablecoins? Then comes the next one. Because new ways to move money don’t replace what came before. ACH, wires, and even checks are still part of the mix.

Without the right foundation, every new rail means another vendor, integration, and more time before you can ship.

Modern Treasury changes that. With one unified API, you can build across today’s payment methods and be ready for whatever comes next. So as the way your customers move money evolves, your product can evolve with them.

Trusted by Procore, Navan, Morse, and companies of all sizes. Powering $600B+ in payments.

Start building today: moderntreasury.com

1. The Ninth Circuit ruled against Kalshi. Sports contracts are bets.

The Ninth Circuit ruled against Kalshi on Friday, in its ongoing battle vs the states on "predictions" vs gambling. The panel voted 3-0 that Kalshi's sports contracts are sports bets. Bets sit outside the CFTC's exclusive jurisdiction, so Nevada can enforce its gaming laws. The ruling binds nine western states, including California, the biggest state with no legal sportsbook.

🧠 The court opinion's first sentence was a killer. From Kalshi's ad copy: "the first app for legal sports betting in all 50 states." Rough way to open a loss.

🧠 We probably shouldn't be surprised it was Nevada. The state derives a lot of revenue from gaming. Roughly a third of Nevada's general fund comes from the resort industry (per the Nevada Resort Association).

🧠 It splits with the Third Circuit, which sided with Kalshi against New Jersey in April. The courts are not unanimous. That likely means this has to go to the Supreme Court. (In fact, Polymarket odds jumped from 30% to 64% that this happens this year)

🧠 The problem is sports. Sports was over 90% of Kalshi's trades and 95% of its revenue in 2025 (the court's numbers). Meanwhile, election contracts got sent back down, undecided. So the product Kalshi is famous for lives on. The product that makes Kalshi money lost.

🧠 Kalshi self-certified its sports contracts; the CFTC never approved them. CFTC rule 40.11 still bans gaming contracts on an exchange, and the court said the CFTC's pending rewrite of that rule changes nothing today.

🧠 CNN estimates 18-to 21-year-olds traded $5.4bn on Kalshi this year, $3.9bn of it on sports and parlays. Kalshi says that cohort is 3.14% of volume. Both are true when total volume is $173bn.

🧠 Katie Haun (Haun Ventures) made a really great point on X. Congress changes, administrations change, but court decisions stick. Wanting one federal regulator sounds great until that regulator turns hostile.

🧠 Right now the CFTC is on Kalshi's side. All three judges on this panel were Trump appointees and ruled against Kalshi anyway. So the Supreme Court will likely decide whether the US has one sports betting regulator or fifty. And that regulator might not always be kind.

2. Stripe and Advent walk away from PayPal, killing the $53bn take-private

Bloomberg reports the Advent and Stripe consortium has abandoned its pursuit of PayPal, ending what would have ranked among the biggest leveraged buyouts in history. The offer, made in July, was $60.50 a share, a 28% premium valuing PayPal above $53bn, backed by roughly $50bn in committed bank financing. The board saw it as inadequate and flagged regulatory and financing risk.

🧠 PayPal never formally said no. The offer sat unanswered from July until the buyers stopped waiting. Brave.

🧠 The market was priced for a bump. PYPL closed above the $60.50 offer on Wednesday, then dropped 12% overnight when the bid vanished. Clearly the market was more excited by a buyout than an existing management turnaround.

🧠 Every reported buyer has now left. Block exited the consortium in the spring, before the offer even went in. Stripe and Advent followed this week.

🧠 Stripe and Advent walking tells you there’s an upper bound on the PayPal price. The competitor with the most reason to stretch stopped at $60.50 for 439 million consumer accounts.

🧠 What’s PayPal’s realistic comeback story? Stripe and Adyen already dominate enterprise processing in PayPal's home market, and Checkout.com is heading for the US next. Each has a single unified platform, stronger unit economics, and a better growth story.

🧠 The board just bet the company on itself. The CEO now expected to build more than $60.50 a share of value chaired this very board until February. Every earnings call from here gets graded against that number. And if it fails, why would another PE try now? The board looks unwilling to do business.

🧠 Considering this is the same board that's been in place for a while, I’m not sure where we’re supposed to get confidence in a turnaround from? PayPal has a lot of legacy systems, and historic M&A that never got integrated. Former CEO Alex Chriss started this work, and then was quickly booted by the board, which didn’t move for this Stripe offer. They’re either determined to fail, or see something I can’t.

🧠 PayPal still has amazing assets and talent. Every interaction with operators at PayPal is a joy. Good people know what to do, but turnarounds take years of pain. I’m not sure this board has the stomach for that.

3. Stripe added Link to Instinct, the text-message AI agent that hit a $2.5B valuation this week.

Instinct is invite-only. You text it on iMessage or WhatsApp, and it books flights, buys tickets, and cancels subscriptions for you. Users who buy through it spend over $1,300 a month on average. It closed a $250M Series B led by Index and Benchmark, up from a $100M valuation a few weeks ago, and the app is still free. With Link, the agent never sees your real card. Link mints a one-time virtual card after you approve each purchase.

🧠 It works like magic until it doesn't. I've used it to book restaurants and haircuts for the kids. It found a free slot in my calendar, booked it and dropped in the invite. I've also watched it get blocked by merchants and fail to authenticate.

🧠 Stripe just removed the scariest credential from Instinct's security model. Until now you stored your card with Instinct, and it would do 2FA for you by reading your email, which testers found could be phished. The agent still holds your inbox. It no longer holds your card.

🧠 Link is becoming the wallet agents default to. It works with OpenClaw and Claude, and now Instinct. Stripe also handles checkout inside ChatGPT, Gemini and Copilot. Stripe now sits on both ends of an agent transaction.

🧠 Every consumer gets an agent, and the agent is the new channel. Web to mobile still had a human reading your page. Instinct and Grok Bot are both early signs of what will be normal in a few years time.

🧠 Most companies are NOT ready. Your e-commerce funnel isn’t ready. Your customer service isn’t ready. Your strategy certainly isn’t. When an agent reads your marketing, fills your forms, calls your support line, and trips your bot detection. Your CAPTCHA is about to start blocking your best customers.

2. Socure raises at $5.2bn, its first new valuation since 2021, and buys agentic fraud startup Fravity

The round is $156m led by Summit Partners. Socure last raised at $4.5bn in November 2021 (Accel, T. Rowe, Tiger Global), the peak of ZIRP, on revenue in the tens of millions. Today it reports $364m of total ARR, growing 63% YoY, 133% net dollar retention, and it's profitable. Fravity builds AI agents that do the investigation work in fraud, AML and KYB cases (the paperwork side of compliance). Socure says it cuts cost per case by 80% and false positives by up to 70%.

🧠 They didn't need the raise. The valuation moved 15% while revenue moved at least 3x.

🧠 That's what growing into your valuation looks like. Most of the 2021 class never did; Socure did it on revenue that's still accelerating, which is the path to IPO.

🧠 Fravity is a side dish (albeit a fascinating one). It's small and early, but with Socure's distribution it could go a long way.

🧠 Fraud, AML, identity and KYB used to be four vendor categories. They're collapsing into one, and the FRAML market will follow a power law.

🧠 A handful of platforms win, and every one of them ships agents. Socure is one; Sardine is another (disclosure: I'm an advisor). But there won’t be many in this category.

🧠 Buying Fravity is Socure paying for the agent layer rather than building it. It's also the second company Socure has bought from the Simility alumni network, after Effectiv for $136m in 2024.

🧠 What I wonder about is pricing for agents. Socure gets paid per API call today; agents that close cases get paid against analyst headcount, and Johnny Ayers put that at $100bn a year of compliance labor. Nobody really knows how to price AI yet.

🧠 This whole sector is about to get 1000x more interesting when AI is the customer too.

3. Revolut passes 80m customers and builds a research lab around its own foundation model

Revolut now has 80m+ customers in 40 countries, adding 5m in the last three months. This morning it launched Revolut Research, a division working with NVIDIA on PRAGMA, the foundation model it trained on its own customers' transaction histories. On historical data it cites 2.3x higher accuracy at spotting credit default risk, 65% more fraud caught with 17% more precise alerts, and 41% more relevant product recommendations.

🧠 Revolut is adding a million customers every 18 days. Nubank does it in 23, and Nu is in 3 countries to Revolut's 40.

🧠 This has compounded for years. Revolut took four years to go from 1.5m to 15m and 21 months to go from 50m to 80m; Nu has added roughly 20m a year since its 2021 IPO.

🧠 If both hold that pace, Nu lands near 220m customers by 2030 and Revolut near 170m. That would make them the two largest banks outside China and India by customer count. Chase serves 87m consumers today. If you’re ignoring these because they’re not big in your country, now’s the time to pay attention.

🧠 The biggest difference between Revolut and Nu is the loan book. Revolut has 58% of Nu's customers and 7% of its credit portfolio ($2.9bn against $39.4bn), plus a fresh UK banking licence to close it with.

🧠 Nubank was the first to build an at scale foundation model for underwriting. It bought Hyperplane in 2024, published nuFormer last year and now runs underwriting, service and growth decisions through it across 138m customers. Two banks with 220m customers between them landed on the same answer: train your own foundation model on your own data.

🧠 Revolut says it will open-source the technical frameworks, which tells you where it thinks the moat is. Head of AI Pavel Nesterov: "you cannot rely on third-party blueprints." And I’ve heard this from other researchers. The off the shelf blueprints aren’t tuned to finance. By open sourcing these, Revolut is doing a bit of employer branding to attract talent.

🧠 More customers, more data, better model. Every 18 days another million customer histories go into the training set, and the model gets better at the one thing the balance sheet needs.

🧠 I expect foundation models for credit to be the dividing line for banks over the next five years. The ones training on their own data will price risk better than the ones renting a model and bolting it on.

🧠 Then there's the stablecoin. Revolut launched EURR the next morning, a euro token issued by Stripe's Bridge, into an app holding $67.5bn of customer balances.

🧠 Every euro stablecoin combined is worth about $810m. If 1% of Revolut's balances moved into EURR, it would be the biggest euro stablecoin on day one.

🧠 Revolut built the AI model and rented the coin, which is consistent: the 80m customers and their data are the moat, and issuance is something you can buy. Customers, intelligence, settlement, all announced in one morning. That's what a global bank looks like when you build one from scratch in 2026.

🧠 Why launch their own and not adopt an existing one? The Euro stablecoin is likely a good operating currency for expats far away or non Euro markets. Under MiCA it can’t pay yield, but by issuing it Revolut gets to keep that yield.

🧠 Revolut will likely support and adopt other stables too. The app already has USDC, they had to switch off USDT (again because of MiCA). I’d wager they’ll support OpenUSD, and even Qivalis if it got scale.

4. Vanguard buys Altruist for $4.6B because America is running out of advisors

Vanguard is acquiring Altruist, an RIA custodian that handles account opening, trading, billing, and reporting for 6,000+ independent advisors through its own self-clearing broker. Axios puts the price at $4.6B in cash (Vanguard says undisclosed), which is 2.4x the $1.9B Altruist raised at in April 2025, and the biggest acquisition in Vanguard's history. Vanguard seeded Altruist in 2020 and made the first call this year.

🧠 The reason is buried in the press release. Salim Ramji: "The need is broad, but the capacity to provide high-quality advice is limited." Cerulli says 110,000 advisors plan to retire in the next decade, 38% of headcount and 42% of client assets.

🧠 So the constraint on advice is people, and Vanguard is betting software fixes it. One advisor serving more clients, on rails Vanguard owns, while $124 trillion gets handed down a generation through 2048.

🧠 Custody is the shelf. It holds the cash sweep, the model marketplace, and the account-opening flow. Schwab earned $3.2B of net interest revenue last quarter on $454B of client sweep cash, about half its revenue. That's what pays for "free" custody.

🧠 Vanguard is fifth on Altruist's model marketplace today. Owning the custodian could make it the default fund provider. It could also kill the value prop, because a lot of RIAs picked Altruist for being the small, neutral one.

🧠 This is the company famous for low fees walking into Schwab and Fidelity's cash cow. If Vanguard prices the cash sweep the way it priced index funds, custody stops being a margin business.

🧠 Robinhood ran the same play from the other end. $300M for TradePMR (350 firms, $43B AUC), a customer base that's 75% Millennial and Gen Z, and an advisor referral network that went live in June.

🧠 Robinhood bought the heirs' end. Vanguard just bought the parents' end. 50 million Vanguard investor-owners, a lot of them the generation doing the handing down. The advisor sits between the two.

🧠 Zoom out and every big player is reaching for a piece of the wealth transfer. The fight is over who stands next to the advisor when the money moves.

5. Salesforce and Anthropic launch Claudeforce

Salesforce and Anthropic announced Claudeforce on Wednesday. Claude becomes the default model in Agentforce Coworker, Agentforce Vibes, and Slack, and Salesforce ships inside Claude as a plugin with 37 prebuilt sales skills (open beta in September). Salesforce stock jumped more than 20% the next day, with Benioff declaring, "this nonsense of the SaaSpocalypse, I think it's time for it to stop."

🧠 The SaaSpocalypse thesis said Anthropic would eat companies exactly like Salesforce. That has not happened. Anthropic uses Salesforce as its own CRM.

🧠 Turns out not everyone vibe-coded their own CRM. Anyone could vibe-code a CRM with Claude Code now, and plenty of small companies have. At enterprise scale, you need the approvals, audit trails, and territory rules Salesforce spent 27 years building.

🧠 Anthropic gets distribution into legacy enterprise, where its biggest customers already live. Salesforce gets an AI story. In 2026, you're either an AI victim or an AI winner.

🧠 Salesforce has never put its "force" suffix on another company's product before. The name alone tells you where Benioff ranks Anthropic.

🧠 Benioff is the GOAT of enterprise software. He's survived every platform shift since 1999 and could probably teach Dario a few things about the game of big-company CEO.

🧠 Anthropic's July revenue run rate was $65bn. Salesforce guided to $46bn for its entire fiscal year.

🧠 Wild thought: Is there a world where Anthropic buys Salesforce? Anthropic's last round valued it at $965bn. Salesforce, even after this week's pop, trades near $200bn. Or is Claude worth more staying neutral?

🧠 What's this got to do with finance? Ask yourself: does your company run on Salesforce or Slack? Next question.

1. Sapiom - Stripe Atlas + Mercury + Okta for AI Agents

On the surface, this is a model router and agent sandbox. Sapiom helps companies auto-route requests for AI to the most cost-effective LLM, build agents and then continuously monitor them. It also manages spend and can attribute AI usage to work produced. Perhaps most usefully, this is all via a single integration; no additional contracts needed. They run models on their own GPUs and inference stack to keep costs low, and everything happens on a single ledger that holds value and an audit trail.

🧠 You can see how this becomes more in time. The companies that built this in-house are shipping 3x more features with half the team size. Where most enterprises are stuck in co-pilot hell, they don’t want to go all in with one (very expensive) AI lab, and they need cost tracking. This is that, so is primitive and Chamath’s 8090. These “control planes” are in landgrab mode. The question now is how big they get before Salesforce, AWS, Google, and Microsoft buy them out or build this. The founder previously was director of eng for Shop Pay. He has a long-term vision and a fascinating heritage for this problem space.

2. Maximum - “AI Native” core banking

Maximum is aiming to take on Fiserv, FIS, and Oracle with an AI-first banking system of record. The goal is actually to build a full stack of compliance, risk, payments, and operations into a single unified operating system. Agents can then run entire workflows making the bank much more efficient.

🧠 Core banking is a tough nut to crack. Often banks feel like prisoners to their provider, and that’s because the core is the heart of the bank. It’s the source of truth that is reported to the regulator, and experience shows known and a bit slow is much better than new and unknown. If banks make errors switching their core, it’s like messing up a triple heart bypass surgery. The consequences can be enormous. Companies like Thought Machine have won large, multi year contracts after grinding this out for a decade, and they’re only just coming through that. The best path here for banks is probably to stand up new products on this entirely new front to back infrastructure, that’s AI first, and then gradually migrate once proven. I like the audacity.

3. Patify - A spending policy platform for AI

Patify helps companies that use agents to buy marketing, for procurement or in trading to audit and manage what those agents are allowed to do. Companies define a set of policies that agents have to call every time they want to spend; Patify approves or denies and then provides an audit trail.

🧠 Is this a feature or a product? I feel like every spend management platform either has, or will soon have, this capability. But if you’re building a vertical AI company for marketing, then perhaps this is a simple drop-in embedded spend management for agents play.

4. Frame - The compliance and control layer for fiat, stablecoins & tokenized deposits

Frame helps financial institutions make compliance part of settling any tokenized transaction. It’s technology ensures all payment data is fully encrypted (frame can never see it), and then writes it to an immutable settlement record, with a cryptographic proof posted onchain. It builds a core payments engine for rails like FedWire, Target 2, stablecoins, or tokenized deposit networks as a single golden source.

🧠 The hardest part of tokenizing money is the internal transformation. Frame’s pitch is compelling. We’ll upgrade your compliance for all rails, including new ones. Then Frame becomes an internal source of truth, not a settlement network. It connects the networks old and new a financial institution already has and then makes settlement and compliance programmable and realtime. This is compelling because today, sanctions, peps, adverse media, or simple account limits could happen in batch. That’s a real bottleneck, especially where clients are looking for more 24/7 money movement.

1. Financing the AI Boom 3

Marc describes the new $500bn deal between Nvidia and the world's largest asset managers as “Nvidia the guarantor of last resort.” Which is the perfect frame to explain the deal. He covers how A100s a chip released in 2020, now has contracted demand through 2029, which meaningfully brings into question the thesis that the 6-year amortization period companies were using on chips which was a bear case for compute. And there isn’t enough Capex to fund the build out even in the hyperscalers

And Nvidia plays a really key role in this market.

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 18th to 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.

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