👋 Welcome back to another edition of Fintech Radar!
If you’re new, here is a breakdown of what you can expect from each issue.
If you missed our recent editions, you can catch up here. Some previous issues you might want to check out if you’re new include “A Deep Dive Into The Cash App’s Growth Machine”, “The Future Of Payment Initiation”, and “Current: Doing It Differently”.
Fintech Radar is a must-read for fintech founders, operators, and investors. If that’s your target audience, placing an ad right HERE is a cost-effective way to reach them.
If this sounds like a good fit for your brand, head to our sponsorship page for more details and to secure your advertising slot. Prices start at $100 per issue!
If you have any questions, reply to this email and ask away!
🤖 Stripe buys OpenRouter, CNBC
🏃 The Rundown: Stripe confirmed last week that it is acquiring OpenRouter, the gateway that lets developers reach hundreds of AI models through a single API. No price was disclosed, including in the letter Stripe sent its own investors. In that letter, Patrick Collison, John Collison and Will Gaybrick called it Stripe’s largest-ever acquisition, following Bridge, Privy and Metronome, and said they expect it to close in the coming weeks. They described OpenRouter as the largest token routing engine in the market, with token consumption compounding at 9% a week year to date, and said Stripe’s H1 net revenue rose 41% year on year.
🥡 Takeaway: And there it is, Stripe is now in the token resale business. What was originally a rumoured deal valued at $10 billion ended up coming in, according to Bloomberg, “above” $7 billion (the New York Times had it at roughly $7.5 billion, and Axios puts it above $8 billion), and mostly in stock. So the deal sits somewhere near 5% of Stripe’s value. It’s sort of interesting that Stripe has handled OpenRouter’s payments basically since day one, and I can imagine that was a helpful datapoint for the Stripe BizDev team, if you know what I mean.
The letter itself is quite a document. It declares that the singularity began on 1 January and that Stripe has been operating on that basis ever since, which is not a sentence you often find in an investor update. They walk it back a paragraph later to mean a large inflection in long-run trends rather than anything millenarian, but they did choose to lead with it, which is a grand way to kick off a letter about why you bought a token reseller.
On the actual deal, and less on the AI hyperbole, the letter does make some interesting points about the logic behind the acquisition. Specifically, capital and intelligence are the two digital flows every business now runs on, so having built the thing developers use to manage a revenue pipeline, Stripe wants to own the thing they use to manage an intelligence pipeline. Token spend behaves like payment volume: high frequency, per-unit pricing, sensitive to routing, and awful to reconcile. That is the shape of problem Stripe has been building machinery for the entire time it has existed, and that machinery will apply just as well to the token economy.
So that’s all well and good, and for the most part it makes sense. What I hadn’t really thought about was how this stitches together with some of the other products they’ve been assembling in, loosely, this space. Take Metronome, which Stripe closed back in January for a reported $1 billion, and things sort of get interesting. Metronome does metered billing for the likes of Anthropic and Nvidia, and they note in the letter that metered billing in an AI context is inseparable from token serving and consumption itself. Radar is the other one worth throwing into the mix, built to catch card fraud and now apparently catching token fraud at some of the largest AI companies. Neither of those is the reason for the deal, but in combination they show why the opportunity might make more sense than it did at first blush.
On the price, they got in front of the obvious question, pointing out that the core payments engine is profitable enough to support acquisitions like this and that the share count is lower now than three years ago despite the M&A. That’s a real answer, though a roughly $8 billion mostly-stock deal is still issuance, and it nets out only because they’ve been buying back faster than they print. What they draw from it is the more interesting bit: that staying private is an advantage going into all this, which suggests the IPO filing docs stay on the shelf a while longer.
When this was still a rumour I said the thing to watch was whether the labs would keep sending the same share of their traffic through a router once a payments company owned it. I still think it’s an interesting question. Stripe sits between every major lab and a large slice of developer demand, deciding which model handles what, while also being a significant customer of those same labs. I could see them staying relaxed about that, or deciding fairly quickly that the routing decision is one they’d rather make themselves.
🛠️ Ramp built its own model router and is giving it away, TechCrunch
🏃 The Rundown: Ramp launched Router last week, a model routing service that lets companies reach OpenAI, Anthropic, DeepSeek, Moonshot, Minimax, Nvidia, xAI and Z.ai through a single API. Customers can route on provider flex tiers, on up to three benchmarks of their choosing, or send only the harder problems to the expensive models, with a dashboard covering token spend, cost, latency and fallback attempts. Ramp built it internally over three years for its own use before deciding to sell it. It’s free until the end of 2026, with a $26 launch credit and inference costs paid separately, and it’s US-only for now. Pricing for 2027 hasn’t been disclosed.
🥡 Takeaway: In the same week Stripe paid somewhere north of $8 billion for a model router, Ramp launched one it had built in-house over three years and is handing it out free until December. Amazing timing for a release, and it sets up an interesting natural experiment on the “build vs buy” front for model routers.
For Ramp, this is a classic scratching-their-own-itch story. They needed to route their own model calls, built the thing that does it, and have now decided the tooling is good enough to sell to everyone else. So a corporate card company produced a competent router as a by-product of running its own stack. Sounds familiar.
It’s obvious Stripe isn’t buying the tech here. They’re buying OpenRouter’s market position: the traffic, the developer defaults, and a view of who spends what across every major model. Fine. Except developer defaults in this market have proven fairly portable, the model mix changes month to month, and swapping a router is closer to a config change than a migration. The alternative hypothesis is that the lock-in is thinner than the price assumes, and a free alternative turning up inside a week doesn’t strengthen the case.
The reason Ramp bothered is the more interesting part, and a peek into what’s happening in the world of wallets to workflows. Ramp’s whole pitch is stopping companies spending money badly, and token spend has become one of the fastest-growing and least-governed lines in a lot of startup P&Ls. Ramp already sells AI token monitoring, and routing is the step where monitoring becomes control, because you can’t enforce a spending policy on something you don’t sit in front of. Looked at that way it’s a card programme for tokens, which is a very Ramp thing to build. Deeper and deeper into the workflows they go.
It’ll be interesting to see which way these two bets go. Stripe’s view is that the routing position is worth owning outright and gets more valuable as token volume climbs. Ramp’s is that routing is a feature you bolt onto something you already sell. The free version showing up days after the expensive one is not the strongest opening for the expensive one.
🤝 Visa and Mastercard join the Agentic Payments Alliance, Rain
🏃 The Rundown: Rain, the stablecoin payments infrastructure company, launched the Agentic Payments Alliance last week with more than 25 founding members. Visa, Mastercard, Fiserv, Circle, Solana, Remitly, Shift4, Evertec, Lithic, Sardine, Chainalysis, Fireblocks and Uniswap Labs are all on the list. Rain says the alliance will be run collectively by its founding members rather than owned by any one company, with early work covering shared research, testing standards for agent identity and authorisation, and advocacy on the regulatory questions.
🥡 Takeaway: Visa and Mastercard already have competing products here. Visa shipped Intelligent Commerce, Mastercard shipped Agent Pay and then Agent Pay for Machines, and both have spent the last eighteen months telling anyone who will listen that their version of agent credentials is the one to build against. Getting the two of them to sit in the same standards body this early is not nothing.
One major reason they’re all in the tent is likely to sort out the liability question, and it is conspicuous that the launch material doesn’t mention it. When an agent buys the wrong thing, or buys the right thing twice, or a merchant disputes a purchase no human ever clicked on, somebody has to eat it. Cards took decades to settle that question, and the chargeback framework that came out of it is arguably the most valuable thing the networks own. There is no equivalent for an agent acting inside a mandate, and nobody wants to be the one who writes the first rule and discovers they’ve written themselves the losing side of it.
This also sits neatly inside the pattern I’ve been banging on about all year, which is the schemes steadily moving up the stack. Core volume growth in mature markets is flat, so Visa and Mastercard keep buying and building into the layers above it: fraud, identity, value-added services, and now the authorisation layer for agents. Turning up early to a standards body is how you make sure the eventual standard looks a lot like the product you already shipped. On that note, I’d expect both of them to turn up on more or less every agentic standards body going while this shakes out. A seat costs next to nothing, and if you’re sitting at all of the tables you can’t really pick the wrong one.
I’d temper the excitement a little though. Standards coalitions are also where good ideas go to move very slowly, and this one has 25-odd members and a charter it hasn’t written yet. Let’s see whether the two networks stay this agreeable once the standard starts to have money attached to it.
📊 Experian puts your credit score inside ChatGPT, PYMNTS
🏃 The Rundown: Experian upgraded its ChatGPT app last week so that UK consumers can pull their Experian Credit Score and score history inside ChatGPT itself. Edu Castro, managing director of Experian consumer services for the UK and Ireland, said it was about meeting people wherever they choose to engage. Experian’s reasoning is that AI platforms otherwise answer credit questions from generic public information that may be wrong. It follows a feature launched in June that lets people explore personal loan offers in the same place. Experian says it powers 8 in 10 UK credit card applications.
🥡 Takeaway: The standalone PFM app has been quietly dissolving into whatever assistant someone already has open, which I’ve yapped on about a fair bit this year from the startup side. This is the same move made by a credit bureau, to itself, on purpose.
Look at what Experian’s consumer business actually is. The free score is the hook, and the money is in the marketplace behind it: card and loan offers matched to your file, with a commission when you take one. That funnel has always assumed the consumer opens Experian’s app to check their number and sees the offers on the way past. Putting the score in ChatGPT gives that up. They’ve already moved the loan offers across too, so they’re relocating the whole funnel rather than just the top of it, which suggests somebody there did the maths and decided the traffic was going anyway.
The leverage they have is that a credit score is one of the few numbers about you that a model genuinely cannot infer. It can talk about credit all day, but it can’t tell you yours. The bureau holds the file, and if it’s going to get quoted inside an assistant it would rather be quoted accurately and get paid for it than be approximated for free. Owning the authoritative version of a fact is a decent position when everything else is being generated.
Whether that survives contact is another thing. The lender relationships are where Experian’s power sits, and those are not obviously portable to a surface where the assistant decides what to show and in what order. I could see this ending up as the version of the model everyone copies, or as the moment a bureau taught the assistant to do the comparison shopping and quietly became a data feed. It’ll take a couple of years to know which.
The distribution question is shifting from whose app someone opens to whose data an assistant reaches for, and Experian has just shown what it costs to be in that second group.
✈️ Revolut is building its own airport lounges, Finextra
🏃 The Rundown: Revolut announced last week that it plans to open a network of branded airport lounges across major European destinations, and has signed an agreement with Copenhagen Airport to open the first of them next year as part of a push into the Nordics. Revolut’s Hadi Nasrallah pointed to the 75 million customers on the platform and said the company wants to set a new standard for airport hospitality in Europe. Premium and Metal customers can already get into more than 1,000 lounges through the app. It follows the opening of Revolut’s first physical store, in Barcelona, earlier this year.
🥡 Takeaway: It’s sort of funny that we’ve spent years lamenting the disappearance of the branch, but slowly what’s old is new again and some kind of physical presence does matter. Yes, this isn’t a branch, but it’s a touch point for Revolut customers that’s likely just as costly (if not more so!) and is designed as a way for account holders to interact with the brand.
Now to be fair, the economics are less strange than the optics. Premium and Metal customers already get lounge access, and Revolut currently pays somebody else per visit for it. Running its own rooms turns a variable cost that grows with every subscriber into a fixed one, and subscription tiers are where the money is for this generation of neobanks. It also converts a benefit nobody sees into a branded room in a departure hall that thousands of people walk past daily, which is advertising that happens to pay part of its own rent.
Brazil got there well ahead of Europe on this. Nomad opened a nearly 600 sqm lounge at São Paulo’s Guarulhos airport back in 2023, and Nubank followed last year with a 1,000 sqm one across three floors in the same terminal for its Ultravioleta tier. Those markets figured out sooner that a premium tier has to feel like something rather than just unlock a list of perks in an app. And I think that’s what’s driving it for Revolut. The problem they’re solving is how you make a premium tier actually feel premium in a market where discounts and rewards are a dime a dozen.
The obvious question is how they actually pull this off. Airport hospitality is a real business with staff rosters, food safety, lease negotiations and a queue at 6am, and none of it responds to shipping a fix on Tuesday. Revolut has a long record of announcing more things than it finishes, and this is a category where a half-finished version is a bad room that people remember. Copenhagen next year is the one to judge it on, and I’d give it until there are three or four open before calling it a network.
💰 Rezolv raises a $12.5M Series A, FinTech Global
🏃 The Rundown: Rezolv, a Mumbai-based AI-native lending technology platform, announced a $12.5 million Series A last week led by Norwest, with Vertex Ventures Southeast Asia and India coming in alongside existing investor 3one4 Capital. The company was founded in 2024 by Karan Mehta and Sonali Jindal, who previously founded the lender Kissht, and works with more than 22 banks and NBFCs.
🥡 Takeaway: Collections is the least glamorous corner of lending and probably the most obvious place for AI. Rezolv says it handles 6.5 million minutes of borrower conversations a month across more than 12 million loan accounts, which gives you a sense of how much of the job is just people making calls in a language and a register that has to be got right.
Indian lending runs on volume and thin spreads, so recovery rates move the P&L directly rather than showing up as some vague efficiency gain a year later. I assume that makes it an easier sale than most AI pitches into banks, and it helps that the founders ran a lender themselves and are now selling to the institutions they used to compete with. They know exactly which part of the month the collections floor falls over.
🎧 The SMB Context Margin Paradox, Fintech Takes (August 19)
Alex Johnson and David Snitkof of Ocrolus get into why small business credit is so expensive to underwrite that most lenders don’t bother, and whether AI genuinely fixes that cost or just relocates it.
🎧 Stripe’s AI Strategy: Build More, Not Less, The a16z Show (August 17)
Will Gaybrick sits down with David George to explain why Stripe treats AI productivity as a reason to build more products rather than run a smaller company, including the detail that internal coding agents produced around 30% of Stripe’s pull requests in a single week. Add this one to your playlist.
📧 Feel free to reach out if you want to connect. I’m @alantsen on Twitter, or you can D.M. me directly by clicking the button below ↴
Ps. If you like what I’m doing with FR, please share it on your social disinformation network of choice. I’d also appreciate it if you forwarded this newsletter to a friend who might enjoy it.
I Love It! ◌ I Like It ◌ Not Bad ◌ I Don’t Like It ◌ It’s Awful
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.