Limitless Podcast - The $7 Billion AI Middle Man: Stripe Buys OpenRouter

Episode Date: August 19, 2026

We discuss two major AI infrastructure moves: Stripe’s reported multibillion-dollar acquisition of OpenRouter, and Cursor’s launch of Origin as a GitHub alternative. We cover OpenRouter���s model aggregation and routing business, Stripe’s interest in AI payments and agentic microtransactions, and how the deal fits into the growing importance of the middle layer in AI. We also look at Origin’s focus on code hosting for AI agents and the broader shift toward infrastructure built for high-throughput automation.------🔒 Check Out Our Sponsor: LEDGER AGENT STACK 🔒https://developers.ledger.com/docs/ai-tools/overview/?utm_source=Audio&utm_medium=Podcasts&utm_campaign=Limitless------🌌 LIMITLESS HQ ⬇️NEWSLETTER:    https://limitlessft.substack.com/FOLLOW ON X:   https://x.com/LimitlessFTSPOTIFY:             https://open.spotify.com/show/5oV29YUL8AzzwXkxEXlRMQAPPLE:                 https://podcasts.apple.com/us/podcast/limitless-podcast/id1813210890RSS FEED:           https://limitlessft.substack.com/------TIMESTAMPS0:00 Stripe Buys OpenRouter6:22 Agentic Payments7:27 Inference Becomes the Moat9:05 Alex Atallah’s Exit11:27 The Routing Wars14:22 Cursor Challenges GitHub20:29 Microsoft’s AI Problem22:19 The New AI Stack------RESOURCESJosh: https://x.com/JoshKaleEjaaz: https://x.com/cryptopunk7213------Not financial or tax advice. See our investment disclosures here:https://www.bankless.com/disclosures⁠Josh works with Anthropic as a contractor. All views expressed are his own and do not represent Anthropic, its leadership, or its affiliates. Nothing in this episode is investment advice.

Just three months ago, investors valued OpenRouter at $1.3 billion. And this weekend, Stripe agreed to buy it for more than $7 billion. That's a five-time markup in just 90 days for a company doing roughly $140 million in revenue. That's 50 times their revenue, which is outrageous when you look at what other companies in this industry are trading at. It seems like a top signal until you realize that Stripe processed $1.9 trillion of volume last year and only kept 0.36% of it. Open router charges 5.5% on everything that flows through it. So Stripe just bought this take rate of 15 times higher than what it's used to in the fastest growing industry on Earth. And this would seem like a one-off pattern if we didn't also get information from SpaceX, also known as

cursor, who just launched a GitHub competitor yesterday while the GitHub platform was down. In AI, the attention is kind of moving from the model layer to the middleman, and we have some proof that we're going to walk through today going through this open router acquisition and the SpaceX alternative to GitHub. These are pretty big deal. So the number one question, I think, if you're listening to this episode, is why on earth should I care and why do I care about some Tradfai, Striped finance company buying this other random company that I don't really care about, right? I only want to hear about Anthropic. I only want to hear about Open AI. Well, firstly, what is OpenRourter? Open Routter is basically a platform that you can go to and you pay one subscription and you pay once and you get access to 400 plus AI models. This could be open models. This could be closed models like from Anthropic, from Open AI, but they could also be the Chinese LLM open source models. And the idea behind open router is as follows. If you are someone that has a task or a project or a business that you want to build, or largely anything, you want to figure out the cheapest way to do it, the most cost-efficient way to do it, but also using the

smartest models when and where it makes sense. But that's sort of confusing. You would need to get manually many different subscriptions. You would need to kind of know which model to use at which different time. And then the models get upgraded and then you don't know which one to choose. like, are you using GROC? Am I using Cursorses model? Am I using ChatGBT, BT? I don't know. OpenRuctor basically solves all of that problem for you. So you type in a prompt into a simple chat box in the same way that you would in ChatGPT, and it has a smart route on the back end that determines which model to use at which specific time. And it saved people so much money, and it's given much better answers on average. Now, you mentioned, they're using or they're

processing around 30 trillion tokens a week at this point, 100 trillion tokens per month, This is up, I think, 15x from last year, and it's up 5x or 6x in the last six months alone. These guys are like absolutely churning out money, $50 million, AAR, every single year. And so the idea here is, why would Stripe buy a company like this? Well, the answer is pretty simple. When you look at what Stripe does, they basically looked at the financial system. They looked at backs. They looked at how people transact money.

And they thought, hmm, this is really clunky. This is really siloed. We are going to build on top of this and make it. really smooth process to create a transaction, to pay for something, to sell your goods or service. They then looked at AI and they thought, huh, AI is going to eat our entire business up. And these tokens, these AI tokens that people are like prompting and using look very similar to money. And I see that OpenRouter has pretty much the same business model that we have. How do we get access to this AI world? We should just buy OpenRata. And that's pretty much what

they did. Even in their own words, their stated thesis is to be the economic infrastructure. infrastructure for AI. Traditionally, when you are an AI agent going to make payments, you are subject to the traditional rails of a human being, which is 30 cents and 2.9% fees. So anytime you want to pay online with a credit card, that's what you're paying when you pay through Stripe. It's a serious amount of money when it's done at the scale that they're dealing with, which is $3 trillion of money per year. So how do you get around that? You have to build this infrastructure layer. And when Stripe's building the infrastructure layer, you could view it as small different parts of it. So they have bridge. It's stablecoin infrastructure. It's crypto. There's much lower fees. Then in June of 2020, they acquire Privy, which is a embedded wallet specific for agents. Then in September, they acquire agentic commerce, and they start to build this agentic layer on top of it. Then they build their own blockchain in March 26, and now they are making a bid for OpenRouter. And the number is not final. We believe it's $7 billion. No one has confirmed. Stripe hasn't wanted to come out and say the steep amount.

like a few weeks ago. It was rumored to be 10. Now the current consensus is 7. There's a chance it may have been 8, but regardless it is a huge amount of money that they're paying for this. And it seems to be like the next layer, the next step in building this agentic infrastructure situation that they have.

Because when you believe in the thesis that agents are going to get more capable, that implies that the spending that they're going to be doing is much higher. And not only is much higher, but it could be much smaller. Perhaps you're trying to pay for subagents to use, to do this really complex project and you just need to pay a couple cents for server hours. It doesn't make sense to do that with a credit card. There's no infrastructure built

in place yet to have these micro-transactions that agents may want to use. And also, it's kind of clunky infrastructure for agents to send money in large quantities. Like when you think about sending a wire transfer, an ACH transfer, sometimes it takes days, sometimes you're limited in the amount you could send. That doesn't really work for agents, and you kind of want something separate for these

AI entities. So that feels like this is kind of what that is building towards. But the difference is that open router is a different piece of the stack. Open router serves kind of like the meter. It's the point where the intelligence gets consumed and gets priced. So they've bought this middle layer that on the surface appears pretty lame. It's like, oh, they just paid $7 billion for a wrapper. But in reality, they just got the middleware that tends to soon, I'm sure, hundreds of trillions of tokens are going to get routed through. And the amount of data that comes through that on top of everything is huge. So this feels like they, the reason they spent 50 times earnings. They've paid such a huge multiple

on this company is for that reason alone. It's for having the right to be the middle layer of this new infrastructure tidal wave that's happening as it relates to AI agents and tokens. I've been trying to wrap my head around the $7 billion price tag because it just seems like they're overpaying massively for what is essentially a company that makes $50 million a year. You don't typically pay like 120x earnings for something like that. But I, As I start to unpack it in my head, I think there's a few things going on here. Now, one of them you mentioned is agentic payments. I think Stripe is basically making a big bet that the future of payments isn't going to be human at all.

In fact, it'll probably be 10 to 1 human agents to humans. So what a world looks like in that sense is you're never selling to a human in the future. You're selling mostly to AI agents. AI agents aren't going to have big wallets or big budgets. They're going to be making microtransactions. Now, right now, the financial system. sucks for microtransactions. If you go onto like Bloomberg, for example, right? And let's say I want to read an article. I don't have a Bloomberg subscription right now, but I would love to read one specific article. I would love to just pay five bucks to

read that specific article. I can't do that right now. And Bloomberg probably just wants to charge massive subscription fees annual, right? But they're not able to do that. With this new purchase, Stripe is one step closer to be able to enable this. You mentioned they have their own blockchain. This is kind of like a stable coin thing. This is like digital money. If you combine both of these things together, you suddenly unlock agentic payments. And that is basically Stripe's main vision. The second thing, and I think the real reason why they're paying $7 million for OpenRouter, is inference specifically. Now, inference is basically the cost or the ability to serve customers' requests or prompts for their AI model. So you route it to an AI model, you get the answer back, and you serve it to them.

Now, inference over the last, I think, year and a half has basically 10 to 20x. It's become a more valuable moat than training your own AI model. And this has been the case for like all Frontier Labs as well. They realize that post-training is way more valuable. So in a world where anyone and everyone is using AI for pretty much anything, any conceivable kind of like task, you can expect to be able to have high volume transactions. Like money basically just flows through this entire thing. Everyone's doing every single bit of work through inference. And Stripe wants to be the rails that owns all of that.

I was reading this really interesting case study. Josh, do you know where Visa came from or like how Visa came to be? Not a clue. It was in 1958. they came out of Bank of America. There were a pet project inside Bank of America. They were incubated inside another bank. And they built basically this network

to make Bank of America more compatible with other different banks, except no other bank wanted to use this thing because it came from a competitor. So what did they do? They spun out this thing called the Visa Network, and it became super popular because it was like a neutral platform.

And then everyone started using it and Visa itself became a bank. Now, we see the opposite thing happening here where Stripe is basically acquired open rights. And so the question is, allegedly, if it's true, will they keep it separate to their current business or will they kind of keep it as like a neutral good? Yeah, I don't know. Well, you could tell Bill Ackman about that visa story. I'm sure he'd be stoked. That's a nod to yesterday's episode. If you haven't watched, go check it out.

I do have to ask you a question, which is, have you ever heard of OpenC? Like, remember OpenC? Oh, yeah. I quoted an NFT or two. For those that aren't familiar, OpenCe, and we'll get to the reason why I'm mentioning this, but OpenC was the largest NFT marketplace in the world. And for those who were participating in the crypto space. If you ever bought an NFT, chances are it went through OpenC. Now, it had a meteoric rise and then unfortunate fall because, you know, the attention just kind of faded away from the space. But the reason why I mentioned that is because the founder of OpenCe is named Alex Atala, and he just so happens to also be the founder of OpenRouter, the company that just got acquired by Stripe. And we have his LinkedIn here open on screen and like, wow,

what a run. Dude, he was a forward-deployed engineer before it was cool back in 2012. Between 2012 to 2014, which means he probably got equity, and that equity is probably up a tremendous amount. Then he was lead front-end engineer, chief technology officer, and then co-founder of OpenC for four years and seven months, and now co-founder and CEO of OpenRouter. So two multi-billion dollar businesses, and it's about time he got his exit because, man, what a generational run he's been on.

And if you are not familiar with Alex Atala, we actually had him on. this show. We interviewed him on Limitless a while ago. This was back from, when did we do this, August 4th, 2025. So just over a year ago, we had Alex on. At that time, what was Elgin Router valued at? It must have been like, what, one-tenth or less of what it's valued at today. I think it was a, I think it was max like a hundred mill. I think he had just raised, which is why we got him on. Oh, my God. And even just in the last three months, why did we raise with him? We should have invested. We should have invested. We need the limitless fund. Someone give us a limitless fund. That would have been our biggest exit to date.

I know. With this guy. But that's the story about Alex's house. So clearly this guy is a pretty focus founder. He's pretty cracked. This dude understands where the puck is going. I'm really curious to see what he builds next. But I found that to be a really interesting story.

And not only is he following trends, but he has like some pretty interesting thesis that are behind these companies. The one with Open Round in particular is it's kind of there's another plane analogy. I've been liking these plane analogies where nobody really holds loyalty accounts with 10 different airlines. You normally either have one or you. you search by the cheapest flight. So for me, I like flying Delta a lot. I have a Delta membership if I'm not flying with Delta to increase my membership status or to really use their product suite, I will go to Google flights and I will buy the cheapest possible flight. And that's all I want.

I just want to get from point A to point B. OpenRouter serves as that layer. Open Router is the company that will sift through 400 different AI models and ingest your query and tell you exactly where you can go to get the best outcome for the cheapest price. And that was the thesis that people really resonated and that's what worked. So, I mean, congratulations to Alex, this is a pretty amazing exit by any means. And to exit to Stripe, again, who still is not public for some reason, Stripe seems to be like the last large company that needs to file their S1. Hopefully we'll get that soon, but pretty amazing run by OpenRouter and the company. And I'm like really stoked to see where Stripe takes us next. I think it's also worth pointing out that OpenRouter isn't operating in like

a silo. There are many other companies that are rumored to be building their own routing platform. look no further than OpenAI. So Open AI has a bunch of different models. Their most recent model release has about like three different versions. They have a routing feature that enables your prompt to go to the right model at the right time for the cheapest possible way. And then if you look at Meta, Zuck saw the rumors about Open Routary potentially being acquired and launched a project code named, I believe it's Switchboard, which aims to do the exact same things with Meta's own models as well as a bunch of new open-hosted models. And then if you look at the financial system itself, Ramp, which basically does something similar to Stribe

processes a lot of transaction volume. They created an internal routing platform that saved them. I think it was 40% of their annual spend, not AI spend, but annual spend in general. And they're now creating that or externalizing that as a product for anyone to get access to. So we're seeing this growing trend of people realizing that there's not just going to be one or two models that you're going to be using every single day. you're going to be using multiple and you don't need to pick and choose which ones. They're going to be hundreds. So just get a routing or aggregated platform to do it.

And again, Alex Atala has a history in building aggregated platforms. OpenCU is exactly that for NFTs. You have to imagine that companies like Anthropic and OpenA are seeing this and they're thinking to themselves, well, hey, wait, we just have like a suite of models from high to low. Very cheap ones, very high-end ones. Why can't we create a router ourselves? So I imagine, like you just mentioned with Zuck, we're going to see a lot of that, mostly just due to the nature of agents and how these agents are working.

Not all of them need Frontier tokens. But for the ones that do, we have a company for you. And it is called Ledger, the sponsor of this episode. Thank you, Ledger for sponsoring this episode. If you are building with AI agents, chances are you worried about security. There are break-ins all the time and increasingly happening. Ledger has a three-step process where the agent proposes the humans approve and then the Ledger signer enforces.

There comes equipped with the Ledger Stack, which is an open-source framework filled with tools that you could use to help you navigate the way that you use your agents. It works with CloudCodex and Curcer. It is open source and available today, linked down in the description. Thank you to Ledger for sponsoring this part of the episode. Now, we did mention in the intro. I uttered the name of this company named SpaceX or Cursor. SpaceX just bought Cursor for $60 billion.

And Cursor has been on this generational run so far, where they released GrockBot, which was a cursor product now under the SpaceX category. And just yesterday something funny happened because GitHub was down. And for those of you who don't know, GitHub is basically where the world's code base is hosted. And there were a few hours during the day, which GitHub was having some outages. And during that time, Cursor, that small little company that's SpaceX acquired for $60 billion, the fastest growing company on the world, they released their own

alternative called Origin. And Origin is basically the GitHub alternative. Now, why are we talking about this now in this episode? It's because the thesis is very similar to what we just talked about with the Stripe acquisition. A lot of the value is moving to being the middle layer, the middle man. You want to be the infrastructure provider for where all the tokens get generated. So these companies aren't necessarily racing to generate the tokens, although in the case of SpaceX, they are. But instead, they're working towards this middle layer. And this cursor origin deployment is another example of how people just really want to be the toll booths. They're building the place in which you have to pass in order to build with these AI-agentic systems. And the best part about this

is you can get public exposure to this thing by just buying one single stock, which happens to also own a rocket company as well. SpaceX. Isn't that crazy? Yeah, unlike Stripe, unlike Oberado, who are still private companies, SpaceX, I still don't think a lot of people realize that they are very much so a frontier AI lab, and the acquisition of cursor will go down as one of the best acquisitions ever, in my opinion. But yeah, like, focusing on this particular launch

and why it's actually useful, code hosting platforms are used by pretty much any and every software engineer in the world. So basically it allows you to upload your code repository and collaborate across various different engineers to kind of update the code base, submit kind of changes that you want to make, and you can do so collaboratively, online securely, privately if you want, or publicly, if needed.

And so GitHub had the moat on this entire thing, and Microsoft did the smart thing years ago, and they acquired GitHub, except they didn't leverage that at all for any of their AI products. So now what cursor slash SpaceX have is a fully integrated software stack, where they're building the models, they now have the routing platform, which is basically cursor for coding specifically, and now they have the code hosting platform. So what does this mean? Any future coding model that Elon Musk creates,

so Grok Code, Grok Build, will essentially have a really tightly integrated product loop and they'll be able to see what developers and engineers are building, how they're behaving, and be able to iterate and build much better frontier coding models, which makes me basically, I don't want to turn this into a SpaceX Bull episode, but it makes me incredibly bullish about the entire thing.

But the second really cool thing about this, which separates it from GitHub, in my opinion, They've specifically engineered Origin, this new code hosting platform, to be extremely high throughput. And that's specifically not for humans. It's for agents. So what I really like about this demo, I don't, I can't remember if it's this demo or another demo I watched.

They had about a couple, I think it was like 50 GitHub commits per second. So they ended up having multiple agents. I think they had like 100 agents at one point pushing commits to this new cursor origin platform, which basically, I think probably like 10x's the speed or minimum 10x is the speed at which humans or the best engineers are able to push code. So the point is they've engineered an entire platform around AI agents doing the bulk of coding in the future.

So that means testing the code, reviewing the code, submitting the code, making sure other agents can help review the set code. And so humans at the end of the day will just kind of sit back and manage a fleet of these agents as they code. GitHub currently isn't oriented to do that. but this platform is, and I think that gives them a major advantage going forward. Yeah, some of the launch claims, 296,000 clones per hour, 22 commits per second per repo. For those who don't know, that's just a tremendous amount of code being pushed.

This is superhuman. This is not something that GitHub ever built itself for because it was never expected that humans would ever do something at a rate this fast. So this is basically, if you were to redesign GitHub from first principles in the year 2026, this is what you get. It's just a tremendous amount of throughput. and this is coming at a time and with GitHub is really stressed. We showed a chart on the show a few weeks ago that shows the amount of commits to GitHub

going basically in a vertical line. It's a hockey stick. They've never dealt with this much volume before. And as a result, they're starting to deal with some issues, one of which was the outage that happened yesterday, was a global outage for six hours and 42 minutes. That is a tremendous amount of time for a company that hosts the Globes Code, which meant that a lot of people couldn't push or deploy. During that time, ironically is when this launch video came out.

And it's funny, you know the timing was a little bit rushed because there's not even audio in the launch video. I saw this great post online that said like, if your video, if your launch video has audio behind it, you've shipped you late. They didn't even get to processing audio. They didn't even put a song behind it. They just shipped it. They said, here, we're live. Go try it out. I have to.

I can't help but think about Microsoft in this situation. Because Microsoft's, we know they've been struggling. They've been having a tough time. They haven't been able to produce really anything of value in the world of AI. They have all this open AI IP. They haven't really done much with it. Their flagship product, I guess, is still co-pilot, which no one uses unless you are forced to through your employer.

Now, their crown jewel GitHub has a pretty serious contender in Cursor. Or I should say, SpaceX AI, the soon to be the largest AI company or one of the largest AI companies in the world. So when I think about Microsoft, I'm like, man, what do they have going for them? Dude, don't worry. They've only been flat yet to date. Yeah. Wow.

I mean, let's see it like, let's see on the day. on the day we're looking at, yeah, okay, seems to be unbothered. I mean, they did have a breakthrough earnings, quarterly earnings recently, which their cloud revenue basically went vertical. But again, if you just end up becoming a cloud provider in the world where you could have owned the AI or frontier AI model lab, you've fumbled a bag drastically. What does the five year look like on that?

Oh, don't even. How have they been doing for a whole 65%? Not great for five years being one of the leading AI companies. Like, that's tough. And like you mentioned, it's like, okay, The company will continue to do great, right? They have the infrastructure. They have Microsoft Azure.

They're a huge cloud service provider. It has incredible margins. It's growing very fast. Everyone needs cloud. Everyone needs these access to compute. But other than that, like what do they have? Are they going to become a cloud infrastructure provider? Is that the end state of Microsoft after all these years?

Well, there is a counterpoint to this, which is pretty much any company that isn't a tech company. In literally any industry uses Microsoft. Microsoft's suite of services. I cannot tell you the number of business meetings I've been on over the last six months that have been on Microsoft Teams, and it makes me cringe. But, listen, it works. And that just shows the deep embedding that their software infrastructure has for the majority of companies in the world. Now, that is a very sticky moat. Now, Anthropic, OpenAI, are trying to chip away at their mode, right?

Saying, hey, use our models, we'll deploy, we'll give you forward deployed engineers and try and figure out how best to integrate AI because they, they can't really get into the client's computer systems other than that way. Microsoft knows that they have this mode. Satya Nadella wrote this really long essay basically saying, hey guys, I know that you're trying to get into our data mode and we're going to do everything that we can to protect it. But I don't see how they can. They're going to have to eventually hand over the reins if they can't come up with the competitive AI offering. Code Pilot right now is that premium offering.

And I don't like to call it a premium offering because it's not that at all. It can't even do the basics of like a GLM 5.2 Chinese open source model. So again, it's hard for me to try and figure out how Microsoft improves from here, but they do have the time. They do have the data mode. And as long as they protect that, they have a shot. They have the cash balance. So we'll see. We'll see what happens with all of these.

As the thesis play out, I mean, we have now, it's clear like that there is a shift upon us where the last two years we're very much focused on like who was able to build the smartest model. All of the values occurring to that. Now people are looking further out that risk curve. Like where is next? We know that infrastructure is one of them building data centers. but we also know that the middle men, that middle infra layer is like a pretty good place to be. Stripe purchased the toll booth. We have cursor. I guess you could imagine as if they're like paving these new highways. And the labs are the ones who are actually building intelligence. So it's becoming more clear.

Everyone's starting to learn their role in this stack and they're accelerating because not one company is able to do it all just yet. So that's where we stand on Stripe and on cursor. And I think these are two really cool, really impressive exits on the stripe front with OpenRouter. like congratulations, what an exit. We're going to see how that plays off. Clearly, they're very bullish if they had $7 billion to spend. And like, man, I hope one day Stripe decides to go public. They feel like the final one to publish an S-1. So hopefully we'll get that soon. I would love to own part of the number one AI finance lab. And Stripe has all the potential to do that. Maybe some of these funds do. Maybe Leopold launches someone in the future. God knows. Yeah, I would love to get access to different sectors that aren't just generalized LLMs. And I think finance is a big one. And

Stripe's acquisition is huge. And Stripe is just like this unbelievable team of builders who are just so talented and so good at what they do. And if I were to place a bet on anyone actually winning this next frontier of finance and what that looks like, I would certainly pick Stripe. So very bullish on the future of Stripe as it relates to the future of payments. And I think they're a company to look out for OpenRourters, just fuel on the fire. And that is the update. So thank you all so much for watching. If you made it to the end, you are a real one.

We appreciate you sticking along for the journey. If you enjoy this episode, don't forget to share it with a friend. That's how we grow, which we've been doing so fast, and the comments have been so sweet. So thank you to everyone who has left. A nice comment. We read pretty much all of them. I wake up in the morning and like the kind of scroll first thing and just see all of the nice and sometimes not so nice things. But I appreciate the feedback both ways.

It's important to keep us humble. Eda, any final party thoughts before we head out of here? No, I think that's pretty much it. I'm curious what people's thoughts are on an episode like this. Is this something that's interesting? Is this something that's just boring for you? Do you want to just hear about muddle of it? updates in general, or is talking about different sectors and applications of AI in different sectors

useful to you? Let us know in the comments, DM us. We're available on X. Share this with a friend. And yeah, wherever you're listening to us, if you are subscribed, if you haven't given us a rating, please do so. It helps us out massively. And we will see you on the next one.