🧠 Klarna pulls IPO, Plaid Raises $6.1bn, Circle files S1 but pulls $4bn IPO.

Hey Fintech Nerds 👋

Wow. Klarna is NOT going to IPO. We don’t know what they’ll do instead. Plaid is NOT going to IPO. Instead, they raised $575m at $6.1bn Tariffs are crushing the macro environment. More in Things to know 👀

Circle paused and I’d written a full S1 breakdown this week 😭. A company that failed to exit via SPAC in 2022 re-invented itself not one, not two, but three times with a founder who marks its third big exit. It's an incredible story and might have more twists left. Things to Know 👀 section.

Visa is alleged to have offered Apple a $100m incentive to switch card networks, according to the WSJ. This isn’t surprising. What is surprising is how long getting away from Goldman is taking. More in Things to Know 👀.

📣 Rant this week was one I’ve wanted to do for a while. You’ve probably seen a lot of hype around Model Context Protocol (MCP). I wanted to ground it in Fintech use cases and reality. Nice short, snappy read for you this week :).

As many of you know, I’ve been passionate about actually stopping scams for a long time after my family was impacted; it's why I work at Sardine*. We’re doing an event on the 21st of August in SF: a workshop format on best practices in UX and risk to prevent and manage scam volumes. You can sign up here.

Here's this week's Brainfood in summary

• 📣 Rant: Model Context Protocol - Flash for AI?

• 💸 4 Fintech Companies: Paid - Stripe Billing for AI Agents, Ubyx - Building Visa for Stablecoins, Enza - Modern digital payment acceptance for African Banks. Openstone - Private Market Investing for French HNW

• 👀 Things to Know: Plaid won’t IPO, raises $6.1bn instead. Circle filled its S1 - They won’t IPO! Plus, the Visa $100m for Apple is business as usual, the state of Apple is not.

• 📚 Good Read: CFPB offers small-dollar lenders reprieve in a blog post

If your email client clips some of this newsletter, click below to see the rest

Weekly Rant 📣

Model Context Protocol - Macromedia Flash for AI?

Remember when the Internet was just static pages and broken image links?

Before Flash arrived, creating dynamic web experiences required serious coding skills. Flash democratized creativity online - suddenly, anyone could make animations, games, and interactive content without being a developer. We're at that same inflection point with AI.

AI still feels like a magic trick for most people. Anime selfies are fun. And in my case a major upgrade. (Although this does kind of ruin the Money2020 drawn selfie schtick.)

The vast majority of companies, consumers, and the economy haven't integrated it into their daily lives. That's about to change with the help of the Model Context Protocol (MCP). 

Although it may not be the perfect standard, it will unlock innovation in AI, just as Macromedia Flash did on the Internet.

Today’s mainstream consumer AI is Janky.

Apple Intelligence might be the most disappointing product Apple has ever shipped (or not yet shipped). AI in products is still, for most people, mostly an annoying Microsoft Clippy.

Google Gemini inside of Gsuite is awful and usually refuses to perform a task you know its capable of.

Tom Blomfield’s tweet here nails it

• Apple rolled back Apple Intelligence because some news headlines were incorrect.

• Anyone used Meta AI today? No me neither.

• Microsoft is the least-worst here with GitHub Copilot and Office Copilot at least showing up.

The big tech incumbents, who have all of the distribution, just can’t stick the landing on AI. Yet, AI is capable of such magic for those who do put in the time.

The mainstream experience of AI today resembles the mainstream experience of the internet in the mid to late 90s.

The Internet in the late 90s was similar

The early Internet was all static images and basic blue links. Search engines were poor, and we often relied on “portals” rammed with hyperlinks to things like sports scores or stock prices. In many ways, the early Internet was worse than TV or radio. Most people could make a basic HTML page, but doing anything advanced required incredible skill and patience. 

Nothing felt like magic, nothing felt like a product—until the early 2000s, when Macromedia Flash was a big breakthrough for the mainstream multimedia users of the Internet.

Before we had YouTube, Netflix or Vimeo, Flash was the primary protocol for online multimedia. If you watched a video online, or played a game through the browser in the late 90s or early 2000s, chances are you used Flash. Those of you old enough to remember spank the monkey or whack your boss can attest, the internet got a lot less boring and the browser got a lot more useful.

The breakthrough with Flash was the ability for almost anyone to host video or create online games. It wasn’t quite vibe coding, but it was much more accessible than putting together the raw elements of HTML and early video codecs of the time.

Ultimately, Flash fell by the wayside as it hit security problems, Apple started to block it. Then HTML5 and mobile apps ultimately displaced Flash with richer, more consistent experiences.

Yet by creating the tool for creatives, Macromedia Flash laid the groundwork for the future of the internet.

Vibe coding tools and Model Context Protocol are unleashing a wave of creativity.

What is Model Context Protocol?

Here’s how Anthropic describes it

You’re thinking, great, why do I need that? 

Today AI has tool use. As a developer you can tell ChatGPT, Claude or Cursor “this is a way to access a weather API, when someone wants the weather, here’s the bit of my API you need”

The problem is it doesn’t scale. 

The developer has to 

• Build custom code for every API 

• Define and drive every action a user can make 

• Manage that workflow across APIs. 

Imagine a scenario where you want to read emails from Gmail, update a slack group and update a table in Asana. If an API changes, or if you want to do something else, or add a new API, that’s custom code, every time, every workflow. 

Enter Model Context Protocol (MCP). 

It can take what might have been a 1 day of integration work down to seconds. This means if Plaid updates their API, your AI can still use it without requiring you to rewrite code the MCP server helps your AI code buddy manage that change.

As with all buzzwords read them backwards to understand them.

• Protocol: It is a standard, open source way for any data or API provider to make their service available to AI tools.

• Context: It gives your AI the ability to understand everything the API or data source can do. E.g. Gmail can read emails, send them, archive and much more.

• Model: AI services like Claude, Windsurf, or Cursor where developers (or anyone) can connect various APIs, and applications together into little workflows.

What experiences can MCP enable?

AI is a superpower if you're willing to tinker and have the time/patience. But when your AI understands other APIs you can get a lot further, a lot faster. Here’s some examples:

When you add MCP servers, and the ability to “vibe code” - to chat with the AI model instead of writing code. People can produce incredible things. I’ve picked a few more visual examples here.

Financial Services Examples:

So you can very quickly allow your AI to do, anything Stripe can do.

At Sardine* we launched our MCP server a few weeks ago to allow card issuers to set rules and manage fraud directly from their AI model or IDE (like Cursor).

I saw Shopify just launched theirs

Some of this is feeding the engagement engine and playing to the audience sure.

But most of it is about learning.

Non Finance Examples:

Have your AI be a Figma designer by adding the Figma MCP Server, or have it talk to the 3D animation suite Blender:

Most engineers live in Cursor, which unlocks incredible productivity for them. Perhaps it is not surprising that Cursor is one of the fastest-growing companies in history, reported to have passed $200m revenue and raised a $9.6bn valuation. 

You could buy 2x Circle’s for that.

Cursor (and its competitors) plus MCP is another productivity step change.

In financial services, MCP could transform how we build products. Imagine:

• Fraud analysts using natural language to pull transaction data, device signals, and merchant details into a unified investigation workspace (Sardine* does this today)

• Compliance officers having their AI assistant simultaneously reference regulatory requirements and customer data to automate risk assessments

• Personal financial management tools that can securely access accounts across institutions to provide holistic guidance

Model Context Protocol is a watershed moment for the industry. It has one major achilles’ heel. It is middleware. And middleware has problems.

Criticism of MCP is valid.

Ben Evans points out a couple of noteworthy flaws: 

• Brands will fight this: Consumer brands like your bank, Uber or Instacart become dumb pipes when all they do is provide an API to AI Agents. 

• Middleware is always limited: We’ve seen middleware like J2EE ultimately become a bottleneck for innovation and hold back progress rather than unlock it.

I also have some other concerns.

• What about security? Often the most useful stuff will still require an AI Agent to pretend to be you if an MCP server isn’t supported. What happens then?

• Setup is still hard for a novice: With some patience you can setup an MCP server, but you’re still copying and pasting command line inputs to a config file on your computer. 

MCP isn’t quite magic in the way YouTube or Netflix is.

MCP is not, however, a replacement for security testing, nor is it the answer to all of our hopes for Agentic AI. 

But, if you’re a builder, MCP is an incredibly helpful way to get from 0 to 1 on product. If you’re an API provider, its a way to help your customers build with your API much faster, and potentially, acquire more customers.

You should be playing with MCP

If you’re the kind of product person or builder who’s tinkered with Zapier, or tests APIs with Postman, then checking out Windsurf and some MCP servers shouldn’t be a hobby. It is now your job.

If you’re stuck in corporate laptop hell, you have some much bigger problems to solve first. Primary among those is ensuring the safety of your data as AI Agents, 3rd parties and APIs increasingly demand access to it.

The hope? 

One day pulling together different APIs and data sources is so easy that even a novice could do it. That everyone (or their AI Agent) could pull together their own personal financial dashboards, by finding and using the right MCP server.

Here’s my prediction.

The tension in financial services is as always, dealing with fraud, risk and compliance. Things can and will go wrong. So our priority is figuring out how we make all of this secure enough for commerce and banking. 

Those who can manage that tension get to make finance (or any experience) feel like magic.

In financial services, the trend is towards more APIs, more open data and building faster. Therefore MCP (or whatever one day replaces it), will collide with your day job. The question is when.

Flash transformed static websites into interactive experiences. 

MCP has the potential to transform finance from rigid, siloed experienced into fluid, interconnected experiences that work the way people naturally think.

Time to tinker.

ST.

PS. Unsure where to begin? Here’s a list of 300 MCP servers you could try.

4 Fintech Companies 💸

1. Paid - Stripe Billing for AI Agents

Paid helps creators of AI Agents capture revenue without relying on subscription or metered models. Paid helps Agents get paid for the value they create. It works by showing users the outcomes the Agent created. It also helps creators stay on top of their margins as AI Agent API fees can begin to get expensive.

🧠 Slick URL and homepage and a solid idea. While everyone is using Stripe Billing, there's a gap for dealing with all of the gnarly edge cases agents create. I have no doubt Stripe will ship many of these features, but there's a simplicity and purpose to Paid that does one job well. That's often how disruptive innovation starts.

2. Ubyx - Building Visa for Stablecoins

Ubyx creates a network of merchants, issuers, and acquirers of stablecoins. It creates an incentive structure for each participant to add value to the network and a legal framework for stablecoin cash equivalence. The breakthrough is a clearing system (think like Visanet for cards) to redeem stablecoins at face value.

🧠 The breakthrough is in the accounting and clearing network. The whitepaper describes how, by ensuring all stablecoins can be redeemed for cash equivalents, stablecoins themselves become IAS 7: Cash Equivalent. What happens next looks a lot like how participants work with Visa. A pre-funded settlement account, standardized redemption processes, and an established redemption pattern. Bravo. Devastated I missed out on the round (not investment advice, but I love this idea). That said, I think it has no real moat other than being a great idea. The issue will be execution.

3. Enza - Modern digital payment acceptance for African Banks.

Enza enables merchants to accept payments in person, via ATMs, or online and orchestrate the payment flows according to their needs. It supports all global card schemes and manages reconciliation and reporting.

🧠 African banks have lost out to companies like Flutterwave and and Paymob. I like that this is flipping the distribution model. Support the incumbents. Banks still ultimately connect to the global wires and have merchant relationships.

4. Openstone - Private Market Investing for French HNW

Open Stone helps buyers access private equity, credit and real estate investments through funds like Openstone Prime for Real Estate. This fund typically requires 10m to invest, but Openstone brings that closer to 100k.

🧠 This "democratization" moves fund investing from UHNW to HNW. Access to these sorts of funds still requires someone to be a multi-millionaire. Mid to high seven figures is becoming more common in the tech circles so its not a tiny TAM and its a high value audience.

Things to know 👀

1. Plaid Blinked - No IPO, Raises $575m at $6.1bn instead.

Plaid is NOT going public in 2025, raises $575m and $6.1bn instead. Plaid says they have ~25% YoY revenue growth, and will use the capital to cover RSU expiry and some secondary liquidity. They now count Citi, Robinhood, H&R Block, Zillow and Rocket as key customers.

🧠 Great timing with Tariffs landing. If you’re going to announce you’re not doing a rumored IPO, doing it the same day as Macro gets crushed by tariffs is very smart.

🧠 As Plaid goes, Fintech goes. I’ve said for weeks there’s a few waiting to IPO looking nervously at markets. How does Circle, Klarna and Chime feel about today’s tariff news?.

🧠 Plaid's story has more twists and turns than most novels. A $5.3bn Visa acquisition blocked by the DoJ, a $13.4bn round in Fintech’s bubble, and now a $6.1bn landing

🧠 This removes a distraction. Open finance is all for grabs. It’s unclear if 1033 and regulation will drive the next 4 years here. Perhaps product innovation will. Plaid’s strength has always been UX and mindshare. This funding gives them a clean run at that. In the age of AI Agents, nothing matters more than secure data access.

🧠 M&A on the cards? If Plaid were going to acquire something or someone, who would it be and why? A genuinely hard question to answer.

2. Circle filled its S1 - Incredible Data but no IPO

This would have been the biggest crypto listing since Coinbase. The data is a phenomenal insight into stablecoins broadly. Circle reported $1.7bn revenue, a 39.3% gross margin, and an operating profit of $167m.

🧠 How does Circle grow if Treasury rates come down?

• Market share isn't growing: Hovering between 20 and 30% (albeit the market itself is growing).

• A recession would mean rate cuts which mean falling yields

• In the S-1, Circle estimated that just a 1% decrease in interest rates could result in a $441 million decrease in its stablecoin reserve income.

🧠 Coinbase wins big in its partnership with Cirlcle.

• Circle earned ~$1.7b revenue in 2024 and paid out $900M+ to Coinbase as a distribution partner that likely comes with very little Opex

• It has replicated this "pay for distribution" partnership with Binance, likely in an attempt to grab market share as a moat from competitors as the market grows.

🧠 Circle says its growing through partnerships and product expansion.

• They added Nubank, Mercardo Libre and Grab this year.

• If stablecoins are the US dollar for everywhere else, Circle aims to win through distribution partnerships

🧠 Regulatory clarity could help Circle or invite competitors.

• Competitors from TradFi: Fidelity just launched a stablecoin, and multiple banks are planning their market entry

• Competitors from DeFi: USDe is the fastest-growing stablecoin in history

🧠 Circle has the most capable lobbying and policy arm in finance.

• They've successfully influenced everything from Europe's MiCA regulation to both proposed US laws (GENIUS and STABLE).

• They've arguably "won" the narrative on "stablecoins as a way to extend dollar dominance in the 21st century."

🧠 Circle's story is fascinating: It's a three-time phoenix:

• 2013: Launched as Bitcoin wallet ➡️ Failed to beat Coinbase

• 2017: Pivoted to OTC trading desk ➡️ Ultimately not defining for them

• 2018: Final pivot to USDC stablecoin ➡️ Now backing $60B in digital dollars and the second largest stablecoin. Also, the largest "onshore" stablecoin (which could be very important).

🧠 It's also survived countless industry crashes and failed exit attempts.

• Walked away from a doomed SPAC in 2022 while still unprofitable

• Survived the SVB crisis and collapse of Silvergate/Signature that nearly broke USDC

• Then built a stablecoin empire challenging Tether's dominance.

• Now they’ve backed away from IPO there could be more twists to come

🧠 Circle's founder's story is also fascinating.

• Jeremy Allaire was the CTO at Macromedia and the originator of the "Flash Player" which was how we watched videos on the internet before Youtube and Vimeo

• It's not an overstatement to say Jeremy saw Youtube and Netflix content like services before anyone else

• He then founded Brightcove, a video platform which he successfully led to IPO in 2021

• Circle marks the third time Jeremy has made it to IPO.

• I can't help but wonder if Jeremy is early to the market and early to exit again?

• One thing is for sure. Never bet against Circle, and never bet against Jeremy.

You'll find more like this in the Tokenized newsletter

3. Visa bids $100m to replace Mastercard as Apple’s Credit Card partner

The WSJ reported Visa’s bid is an attempt to woo Apple away from Mastercard. The bid comes amid the messy divorce between Apple and Goldman Sachs over the Apple Card BIN sponsorship and program management. The report also suggests Apple wants to pick a network before it finds its new bank partner.

🧠 It’s not surprising that card networks offer incentives to switch. This is the whole game. A lot of that $100m, will likely take the form of credits against other products. It’s a juicy headline, but a daily practice in the card network wars.

🧠 With 12m users, Apple is a meaningfully sized card program to win, but an even bigger logo. If you’re the sales team at the card networks, this is your entire year. Huge signalling value for the market too. That said…

🧠 Apple Card, Apple Wallet, and Apple are the disappointment of the decade so far. Apple is really phoning it in lately, and Google, somehow, manages to be one step behind them. And what even is Samsung Pay?

Good Reads 📚

1. CFPB offers small-dollar lenders reprieve in a blog post

The rule proposed in 2016 proposed for small-dollar lenders to check customers' affordability and make no more than two attempts to take payments to avoid "insufficient funds fees." The rules were due to come into force on March 30th but now the CFPB says it will "not prioritize enforcement or supervision actions."

🧠 This, too, will likely end up in the courts. The initial rule was spent much of the late 2010s in the courts before heading to the Supreme Court (as Jason notes). This debate will rumble on.

🧠 While the cat is away... The original rule as described sounds eminently sensible (try not to get consumers into NSF hell). But if nobody is going to enforce that rule, there's a trap consumers can easily fall into.

🧠 Incentives matter. If we can't have competent regulation without everything ending up in the courts and taking a decade, can we do something else? Maybe one for a future Rant.

🎧 Good listen: Ezra Klein and Derek Thompson on the Lex Fridman podcast are excellent because they present the need to shift from a focus on process to outcomes. Regardless of where you are on the political spectrum, this is, to me, the fundamental issue we preventing progress.

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