Coinbase Cuts AI Spend by 50% | Kalshi's $40B Valuation & Impending IPO | The Year for SaaS Roll-Ups

I have a lot of thoughts, but this got so much traction, right? And I think it's an important topic, but I'm just getting burned out on performative social media from struggling CEOs that aren't AI companies trying to act as if they're at the bleeding edge. I really don't care what a lot of these CEOs think about the performance of the latest LLM on their boring old AI or crypto company, but man, I've had enough of the crap from Brian Chesky and the rest. I mean, these are generational founders, to use Harry's words, but it's just performative. Like, put the numbers up, boys. Sorry, we're in a crypto winter. Sorry, Airbnb is still below its IPO price. Go do something about it. Go do something about it. I've had enough. I'm in a totally different place.

Right. So, because I actually thought it was a really great piece, and in one sense — hang on, let me finish. In one sense, let me agree in one sense. It's precisely because in the continuum from frontier amazing AI company to boring, stodgy corporate America, Coinbase is now plus or minus in the middle. This is your point. It's not a — you know, it's four or five years since it's been public. It's not a bright shining frontier AI company, which is what makes the piece so good to me. It was like just a common garden tech CEO saying, 'I was spending X a year ago. It exploded in the last five, six months primarily because of the ability to do code generation, and I got to grips with this thing, and I reduced my spend by 50% in the last two months, and here's how I did it.' Right? It's precisely because it wasn't some frontier leading company that it makes it more relevant, because I think every single company spending $50 million, $10 million on Claude is going to look at this and say, 'If we haven't done all these three things in the next month, someone's head's going to roll.' It's cost management 101. But what it said is, in the space of two months, by getting to grips with your spend, you can continue to innovate, continue to generate more tokens, but cut your spend by half. Right? So, I thought it was a really great piece.

It is. It's just Coinbase fell minus 30% was its last quarter.

Yeah, but it's those things.

No, I'm not disagreeing with you. I'm just — and I think the graph and the chart were great. I'm just burned out. I want a leader doing this. I want a leader doing this.

What about an even shittier company? And you know, we got lots of shitty picks. If one of the automotive manufacturing companies, which have fundamental business challenges, also did the same thing and said we were spending $100 million on AI, now we're spending 50, it would be just as relevant. I don't think — and in fact, I would argue it's precisely those companies that are under cash pressure and earnings pressure that are going to get their act together on not spending too much money on AI, which is why it's more relevant. And frankly, that's most companies. Very few companies have the luxury of a venture-backed AI-forward startup that can say, 'You just get this done no matter what.' Most companies have cost discipline.

Is optimizing your LLM spend really going to help if your revenue is shrinking minus 20 to 30%? I just don't see it's going to reignite growth in your crypto base.

Again, you're just being a grump, guy. He didn't say it's going to — I think it's a — Harry asked what we thought. I think it's a sign. I think the data is valuable. Like, let's move on, right? I think it's great, but I am getting burned out on struggling CEOs on Twitter sharing performative AI data when they're not AI companies. It's like, 'Show me the money. Show me the revenue growth.'

I'm going to argue again. I think it was a fact. I actually really liked it because it was a fact-based piece. And by the way, you know, it just showed here's our spend by month, here's our tokens generated by month, here's our peak, and here's how we were able to clip 50% off it. Right? Let me put it this way. Rather than arguing with you, Jason, which is usually not productive for either of us, I'm willing to bet that every single CFO in the Fortune 500 sent some version of that article to their CIO and said, 'Dude, look what this smart guy in the valley is doing. Figure your act out.' Right? So, I think actually, let's leave Coinbase out of it. Let's pretend we don't even know the name of the person who wrote that article, right? Let's ignore it. I think the really interesting question is what does this mean for revenue traction for the hot, sexy foundation models, right? Because I was looking at the numbers. In one sense, oh my god, you reduce your spend by 50%. That's terrifying if you're the company getting that revenue. On the other hand, the positive spin, it only went back to the spend they were doing in, I think, November. In other words, if you look at it, basically all that happened here was all these companies were spending and growing pretty aggressively in terms of their spend with probably Anthropic and OpenAI. And then in November, December, with coding, it exploded. It took about five months for everyone to get their act together and say, 'We can't be doing this. Let's cut the burn.' And then they figured out and kind of reduced it by 50% back to roughly the spend in November. Now the interesting question is, does that imply — and I'm not saying it does, to be clear — does that imply that, you know, Anthropic, which last year exploded from $1 billion run rate at the start to $9 billion at the end to $44 billion mid this year, you know, I'm not saying this, but you could say, 'Oh, you're saying that their revenue is going to go down by 50%, that they're going to be at a $22 billion run rate.' I'm not, but it's going to have some impact on the growth rate, and that's the —

I think so.

Yeah.

I think there's two issues here. One is it sharpens the question of does the rise of open source and others actually going to impact the growth of the frontier models more than we ever predicted? That's a big question, and I'm going to say I don't know on June 30th. It's easy to say that we could point to a lot of data. I think maybe that's the topic here. I think there's a second point in that where Coinbase was really helpful. It's that like the less dramatic version of what I was saying. I do think there's a second point that the post made that people maybe missed, which is that as we round into the second half of '26, folks are realizing they radically ramped up their AI spend on product. It seems to have worked subjectively, qualitatively, but the productivity isn't there to justify it. I think that's what Brian was really saying. He wasn't that the data didn't say it, but he's like, 'Listen, if we had shipped so many new products to Coinbase, if our product velocity had quintupled because our token spend quintupled, I'm all in.' Like, if that flipped him around from minus 30% growth to plus 30% growth, I don't think we'd be — he might have still done what he's doing, token routing a model, he wouldn't be making this point. I think if you look across even any of our portfolio companies that are doing well that are not purely reselling tokens, they're coming to the conclusion that I'm not quite sure what the hell — I know I want to do this. I can't put the genie back in the bottle. AI is great, but it's not lines of code. What the hell? And so many folks are not seeing the lift from net revenue, net productivity they thought from agentic coding. It's almost a conflict, but it's something that we're all going to have to deal with in the second half. And it's not the same as cutting costs. It's saying, 'Jesus, I spent an extra $10 million in the first half of the year, and we grew the same as we did the prior two quarters.' Like, where's the lift, boy? Show me the lift, right? And CFOs are struggling with that too. Even if the business — my point is even if the business is doing well, they're struggling with it now.

Yeah. Depending on the company, this AI spend on engineering — if you're a software company, it should credibly give you lift, revenue lift, because you're making more. The thing you make is software. You're making more software, you should give more revenue lift. Or if you're a digital goods company. So you're right, Jason, if you're Coinbase, you're like, 'I'd like to have seen revenue lift here,' right? I think it's even applicable for companies the further you are from a digital good. Yeah, silly example, if you have a tech team and you're a car manufacturer, going back to it again, you're not going to get a whole ton of lift from your extra software unless you're Tesla with FSD. But at a minimum, you should be seeing savings. You know, if you were spending $100 million on software and now you're spending $10 million on tokens, at a minimum, you should be seeing savings. And if you're not seeing either, you're going to be looking at this with a pretty jaundiced eye. Yeah. And I think that's what's happening. Yeah. I have a portfolio company that is every number — software company, every number is green, right? Way overloaded with investors, way overloaded with everything. Hit the first half of the plan. Everything's great, right? You would love all the numbers. But at the last board meeting, they came in and wanted to double their token spend, which was massive in the first half of the year. And it was enough to move the burn from no big deal to big deal. Even for a company in the top half percent, even folks were like, 'That's a lot of extra budget.' And for the first time, the board was like, 'Okay, but if you want our approval, tie it to ROI.' And this amazing team couldn't — they couldn't — the velocity is everyone wants to invest, but it didn't directly tilt the curve. So there is a point for even the highest flyers where you're going to say, 'Jesus, you got to — I got to see the ROI.' And I think that's the big — that's just time. You know, we went to token maxing where like everyone just try stuff that made sense, right? And that led to the early folks that got whiplash, like Cursor having to go open source really early. And that's an interesting niche issue on X, but the real issue is just we just can't show enough lift from this spend that it's going to stress even the best of us, not just Coinbase. It's going to stress everybody, and so be it. It's time for the next mature phase of token spending and software development. It's just time, boys, to grow up, right?

You at least have some concerns, which will segue to one is going to be in part our discussion next on distillation and Anthropic's perspective on these open source companies stealing their IP, as they would say. We will leave the irony for a later discussion. But yeah, I mean, it is plausible that you have a world where the frontier — remember, even if the bulk of the tokens are generated using open source models, it is plausible that the bulk of the revenue will still obviously come from state-of-the-art frontier models, right? So, but the question — and therefore there's clearly a very big business here, right? And that's all great. It's to your point, Harry, if you've constructed your world in such a way that only a trillion dollars is good enough and you end up with the consolation prize of half a trillion dollars, right? Which is still, you know, the largest company — I mean, I'm just trying to think here — either the second or third largest company by revenue on the planet, and it would be the largest digital company on the planet. If that's — if you can be the largest tech company on the planet and still not make money, you might have oversized your ambitions a little and it might pay to come back a bit. And that's exactly right, is that there's nothing in this Coinbase memo or a hundred Coinbase memos that implies anything like, 'Oh my god, these are not going to be amazing companies with great products that have differentiation.' It's just, as you say correctly, if you've built a cost structure and a capex spend that you need it all, then the last thing you need are cheap open source alternatives at one-fifth the price. Before we move on to Anthropic's perspective on distillation, Jason, I love you, my friend, but what do you want from these CEOs then? Like, candidly, he's being very factual and innovative in how he's presenting what the company is doing. Like, what do you want from him to just shut up and do the work?

I want to see how AI — Listen, I value the data, okay? I'm not being facetious. I do feel this way. I value the data. So, I appreciate that. But I want to see how AI — I want to see how AI, if at all, can give Coinbase a revenue lift. That's what I'd like to see. Even if it's just them — I mean, listen, Coinbase is subject to the whims of the crypto market, okay? And investors should understand that. And when crypto roars back, Coinbase has grown at rates that are almost Anthropic levels for brief periods of time, right? So it's part of being on a non-recurring revenue journey in a very volatile market. But I'd love to see how growth is 5% higher from AI — AI something in crypto. I'd love to see how it's driving up insurance premiums and insurance margins. I just want to see where this magical boost is from this utility. The LLMs are a utility, right? They're tokens. They're fungible utility. Like, we're kind of teasing at whether they're becoming fungible utilities, right? Is one token replaceable for another is the meta issue. I just want to see a boost. I'm tired of folks like Adobe saying we have $500 million of agentic revenue and missing the quarter. That's performative too. And listen, what would I be doing if I was the CEO of a company not accelerating the age of AI? Man, I might be doing the same, but I don't respect it. But I want to see the real boost. Everyone's faking, everyone's feeling like they have to be part of the AI age, but they're not delivering, Harry. They're not delivering.

I want to applaud Jason for his consistency and disagree slightly. I think you — first of all, I give you credit, you have been remarkably consistent on this, and I'm going to paraphrase what you're saying: software companies in the age of AI are either accelerating or irrelevant, and you're exactly right. I would say, in fairness, I'm going to — I don't think crypto gets an automatic lift from AI, so I think you're being a bit harsh on that company, but your Adobe example is exactly correct, and it'll come — actually, it'll segue to the Microsoft discussion later on. I do agree, and again, I give you credit for this strong principle strongly held. If you're not accelerating in the age of AI and you're a software company, you've got a problem. That's your point, and I think you're correct. So if that tweet — let's put it this way, if that tweet had come from the CEO of Adobe, you would be totally correct in saying, 'That's great, but dude, you need more.' Exactly. So I do agree with you. I would exempt Coinbase from that because I lump them more in the financial than the software space. But in the software space, you are correct. If you're not getting on board this train, you're getting left behind. Even sometimes — and I love them, we all three of us love them — even sometimes for a while, I would get tired of Aaron Levie's constant AI stuff. But to answer your question, it has led to a boost at Box. It hasn't turned Box into a 100% grower. It's gotten it back to double-digit growth. Some of Aaron's stuff is a little bit too — to me, and we all learn from it, right? Some of it's a little bit too AI reflective, but he ties it to his revenue and his plan. He's like, 'This is how it worked at Box. We're processing documents. We're processing content. Here's how it literally ties to our business model.' So it's okay if some of it is a little bit performative. I give Aaron a huge thumbs up on it, but the Brian one I just — anyhow, it's all good. It's great to have the data. I just have a little — I just become a skeptic when it's not tied to their business. That's all. I'm a skeptic of what the goal is here. That's all.

Sure. I mean, you have to power through the irony before you can have the discussion. And we all understand the irony, which is all the foundation models, including Anthropic, were trained on other people's IP to the point where Anthropic recently settled litigation with a whole bunch of copyright holders on books because they had, quote-unquote, unfairly — to use Dario's word applied to Chinese leverage their intellectual property. So I do admire the element of hypocrisy of being appalled when someone else does it to you after having done it to other people yourselves. Be that as it may, let's move on from that. Just wallow in it a second, but then move on. Is it — you know, what's happening again, stepping back for folks, is the allegation, which actually — Anthropic, let's not personalize it, Anthropic the company said actually in a letter to the Senate Banking Committee recently that basically the Chinese open source companies are bootstrapping their development of their state-of-the-art models by effectively breaching the terms of service of Anthropic and sending literally millions of prompts to Anthropic, recording the answers, and using that as training data to start training their models. It's basically taking — and to some extent they would say it's taking, yeah, Anthropic's IP and using it to build open source models which then compete against them. So that's the kind of comment here, right? And so, yeah, so other than — so what, and how that's funny, right? What happens after that? Well, the first thing is, the main thing is, is it illegal, right? And the interesting thing here is it's clearly in breach of Anthropic's terms of service, and that's very clearly expressed, right? But that's not a legal — that's a contractual problem between Anthropic and the Chinese model companies, and they're more than welcome to sue each other in — you know, you knock yourself out in Beijing. Good luck with the lawsuit, dude. Right? I think the interesting thing, and that's why the Senate Banking Committee thing is interesting, is you could imagine a world where, because as well as just being in breach of terms of service, it is also arguable that you have copyright issues and our trade secrets acts, and they go above the level of contractual and then start to get into actual legal issues that the government might take an interest in. Or maybe at its most extreme, I could see Anthropic saying to the government, 'Hey, these are strategic US assets. You know, we're getting regulated separately on how we produce our products. You've got to stick up for us and say you're not going to let this happen and put the full weight of the US government behind it.' And that takes what was a contractual dispute between two parties and makes it the US government putting the thumb on the scales. And clearly that's what they're angling for. Like going back to the Coinbase comment, what happens if in return for Anthropic kind of complying with US government restrictions on overseas access, the quid pro quo in the next piece of legislation is no Chinese model that has been proven in a US court of law to distill using US foundation model technology can be used by a US company? It's not crazy. This is the kind of — so you can see where they want to go. I mean, they're not just kind of crying because it's unfair, in my view. They're laying the pipe for a set of trades to push back on this open source stuff if they're in fact viewing distillation. So that's, I think, what's going on here. That's what I — I'm not claiming to be — that's what I — whether to Harry's point, whether he's no longer become a successful communicator right at this point, whether the spite startup vibe of 'we're the safe guys' has expired and led to conflict with Trump. Putting that aside, I think he wants the models banned for use by US companies. I think he wants Chinese models banned for use by US companies. And I think in an era where SBF may get pardoned, the founder of Binance already got pardoned, this is pretty imaginable. I think it might be perfectly logical for them to get around the table, especially when we're jostling on a geopolitical level and say, 'Listen, we're just going to ban it, or we're going to say any — or we're going to do some weird tariff on any US startup that uses a Chinese model. I have to pay 100% tariff tax.' That's beyond my skill set. But clearly they just want US companies — you can't stop China from doing what China's — I just got back. You ain't going to stop China from doing it for a million reasons. But just put an end to what Cursor and Harvey are doing. No more. You guys can't do it anymore. You guys have to, on grounds of national security. This is theft of our IP. This is theft of our data. We can't trust them. And hey, Cursor and Harvey, your models are just destroyed. Your business models. Thank God you sold for $60 billion because Chinese open source is banned. I don't think it's implausible. I — let's — is it on Khi? It seems to be more than a 10% chance it's going to happen.

Well, you should have, because you know this is — and Bill Gurley is great on this — this is regulatory capture in the extreme, right? The truth is there's two separate — because there's two separate issues that they're brilliantly conflating. One is should they be, quote-unquote, banned because they distilled our prompts and as such got a leg up on that. Naughty them if they did, just like naughty Anthropic and naughty OpenAI for stealing your 1.5 million books, right, to leverage your property three years ago. So pay the naughty fine and move on, right? No one's banning OpenAI and Anthropic because they were naughty. So logically the distillation thing shouldn't result in banning. Separate thing is — and conflating the two is — are you really saying, are you really trying to find a motivation from the government that says ban not because of the naughtiness but because Chinese open source running US AI sounds scary, right? And I can imagine that there are, you know, products that the interest of national security — like Huawei are banned in the US. So it's not crazy that if we believe frontier models are equivalent, that you could see that kind of thing. Now the argument that every tech CEO will and should be making is these are open source models that are sourced and running in the US on US inference. There's literally nothing going — yeah. The code is open to inspection. There's no back door here. There's zero risk involved. Right? But to make that argument implies that you have on the other side of the table a government entity willing to listen and do nuance, and that's hard. So I think that you could imagine, in the absence of a sensible regulatory function, you know, that you conflate the naughty tax for stealing the intellectual property, the security risk because it's China, and then the deep dark secret is the frontier models are actually just trying to defend their vast capex spend by eliminating a low-cost competitor. And it all comes together in a big kind of policy mismatch in return for some of these restrictions on, you know, security and our usage, right?

There's also a middle ground too, or a partial win for Anthropic and OpenAI, is just listen, we can't stop Cursor — which they could — we can't stop Cursor and Harvey, but you know what we can do? We can make every single Fortune 500 company uncomfortable banning open source, like they're not comfortable with from security. There's enough ambiguity out there that it's just not worth it. Right? Whether a startup can take — startups can cut corners that enterprises are not comfortable cutting, right? All you have to do is make it look dangerous to enterprises and they can just ban any open source use in their company. Right? That's the fallback position. That's a good win.

I'm going to distinguish something. They wouldn't say ban open source. They would say ban non-US, Chinese-based companies distilled on US models, because you know they're not a poolside — because look at some — one of the interesting things here is clearly, I mean, all these US companies don't want Chinese open source, they want open source, and at some point there's going to be an interesting opportunity for a US open source company like Reflection, like Poolside, to take some of this revenue. But you're right, Jason, you know, some version of banning these companies that have been found, quote-unquote, guilty of distillation that are based ex-US in China is plausible. I mean, stepping back, one thing I noticed here that Anthropic wrote to the Senate Banking Committee. Obviously, the minority head of the Senate Banking Committee is Liz Warren. You really have to want to get something done. Whether it's a tech company, you say, 'What I'm going to do is pull in Liz Warren on my side.' Very few tech companies say, 'That's what I really need here today, right?' And it just shows that you're trying to, you know, get something interesting done here.

No. Because I think you have to be very pragmatic in terms of what you — and again, there's two separate issues. If they've done the naughty, then they should pay the naughty tax. They should pay exactly as much as Anthropic had to pay to the book guys, and they should have paid to Anthropic. It's easily priced, and we could use the argument — I mean, and that would make things, quote-unquote, fair. So from a distillation perspective, no. Separate comment from a national security perspective. Again, as I understand it, if you are downloading the model, the code is open for inspection, the weights are available, and there's nothing running and no telemetry back to China, then I don't think there's any danger. So I think you probably shouldn't ban them, right? That would be my take. I think that AI is going to be like the oil situation in the Persian Gulf today. And what I mean is it's too — we are so, as an economy in the US, we are so addicted to this. Our 401ks, our stock market, everything we do, every bet, every re-election for every politician — we are so tied to the AI boom. I'm not going to — whether we can debate whether it's a bubble, right? I mean, at some level it has to be a bubble. We are all — we're all on this journey together. 40% of the S&P 500 is tied to this bubble. I think we have to protect it.

And you know, Sarah Frier or whatever a year ago saying that the US government should back all of our data centers. That was, you know, I think it was her that was flamed for saying that, right? She's probably — it was true. And she probably shouldn't have said it the way she said it. I think it was offhand in a comment. I mean, everything — I think what I mean is everything's going to be circular because we — it's like we can say we don't need that oil because we make plenty of it from fracking, but look at the impacts, right? Even if gas prices go up a dollar a gallon, people are freaking out. And if their 401ks go down 20%, if Nvidia stops selling chips, if all of this moves to — I don't know. I think we're going to end up doing everything we can as a society to prop this up. I think we're all going to be — and I think it may even be more important than the other issues. We just don't want unemployment. We don't want these things to happen. And we're all in on this whether we realize it or not. I think we're all tied to the economy. It's all about me. I care about unemployment. I care about — but my 401k, don't touch it, man. You think I'm kidding? I think we're all in, just like the price of gas. Like, yeah, you know, I feel a little bad about — but don't raise the price of my gas. Don't touch my 401k. These are — don't touch it. Don't touch those high-paying jobs.

What you're saying might be true, Jason. I'm not sure it is, but it might be true. But if it is true, what is implicitly — what's implicit in that statement is really very negative. In other words, because what you're saying is keep up the price of AI as an input to the rest of the economy by protecting these foundation models' pricing structure so that they can get a trillion dollars in revenue. And the loser, as always is the case with trade restrictions, is the rest of the economy who won't get cheap intelligence. They'll get dear intelligence. So what it means is the stocks in your 401k that are making AI will continue to make money, and the stocks in your 401k that should be getting the lift from AI are going to lose money. It would be like as if in 1981 or '82 the government's — I give it IBM — IBM licensed MS-DOS and they should have protected it, and instead everyone was able to clone it like Compaq. It would be like the US government coming in and says, 'No Compaq, no Dell. You can't make clones of the IBM. We've got to keep IBM stock price up, and therefore we're going to ban the clones and we're going to keep the PC industry a small, tiny, profitable industry for IBM.'

It could happen in this case because the dynamic of overseas national security makes it happen, but it would be, for the record, so freaking dumb. So I hear you.

Well, here's the counter-argument. I'm not saying you might be right. It might be dumb, right? I think if you step back for a minute — to be clear, dumb things happen all the time, especially when governments are involved. I have high confidence in dumb. See Harry for details. I'm in the UK.

No, no, no. Sorry. If you look at the history of LLMs on this show since it started, right, what has really changed — and Coinbase is just one example of this — we are probably leaving the oligopolistic age, right? And we had an oligopoly when we started the show. Actually, you could argue they were two monopolies. Okay, Anthropic owned coding. Elon Musk, when we started this, said those Anthropic guys have something special. I can't compete in coding. OpenAI owned consumer, right? They both had sort of weird pricing. They made up and were losing lots of money. But it was — and they had these sort of dominant — Now then, as we began this year, we entered an oligopolistic era where we had two leaders. Okay. And for a while, what happens in oligopolies traditionally is you compete aggressively on features and not on price. You essentially soft-collude on price, and the models were somewhat similarly priced. OpenAI would be like, 'Oh, we're half the price for Codex,' but not really, we're as competitive, right? And then typically — I mean, Harry's the master here — when you leave oligopolistic stages, there's massive price erosion because you're no longer competing on features, you're competing on price. And that's exactly what we're seeing. And governments do support oligopolies all the time when it's — and the number one example in the US is medical device companies and Medicare, right? You fix pricing so you can get innovation, and I just —

When you fix pricing because of regulatory capture — I agree with what you're saying is yes, oligopolies are great for the people involved because yeah, you can have a — sometimes for innovation — if you look at the study, when you have folks that are brutally like, 'We're still competing,' but if we agree that basically, listen, we're 200 bucks each for our max program and we're going to charge this much per token. How do you win? You win because Anthropic comes in and builds this disruptive coding model and takes all that revenue, right? That's how you win in an oligopolistic market when pricing doesn't matter and features do, and it actually can be great for innovation in the short term, but maybe not in the long term, right?

I mean, include me out on that. Competition works. I mean, again, going back, yeah, I think any number of examples say you just got to let competition rip. You got to let the open source providers rip. We're going to want — I mean, again, it's the Coinbase example. You know, the counterfactual is if those open source products didn't exist, the Brian Armstrong quote tweet would have been, 'We were paying $10 million six months ago for our AI intelligence. Now we're paying 60. What the freak do I do?' It wouldn't have been as good a tweet, right? We need competition to make this happen. So I hear you, Jason. It might, by the way, it might happen. You might see some pressure, but I think it would be a bummer. Move on.

Ironically, just when they had one of their big announcements and Satya made all this conversation about their direction on AI, maybe it suddenly opened people's eyes because we've been saying for the last year, look, they don't have the standalone model. Their AI story really is we spent a lot of money on capex, which we do to support OpenAI. We have a good investment in OpenAI, which is looking pretty good, though perhaps not as good as it was a year ago. But our core software business doesn't have a compelling AI product. And let's get real, Copilot and Claude Code are eating the two core parts of the Microsoft franchise back in the day, which was, you know, offers for the individual knowledge worker, which is what Copilot is becoming, and developers, developers, as Steve Ballmer used to say, which is what Claude Code is becoming. So when you zoom out a million miles, Microsoft, which is in the software business unlike Apple, and therefore can't afford to stay apart from the great AI wars other than its equity ownership in OpenAI, doesn't have anything compelling to bring to the table. And I think the market's looking at that and going, 'You're not going to zero, Harry. Don't panic. You've had your hit.' I mean, you don't get the lift that you get from the growth story. It's Jason's comment. Do you really have an — I mean, they'll say they have an AI growth story, but if you break it out in the numbers, which they don't do a lot of, that is just, hey, we're selling inference to OpenAI, so we have a growth story. But what they don't have is a compelling end-customer growth story here that they own themselves. So I think the market's finally caught up with the — Look, I don't know the answer either, right there, because it's a very complex business model, Microsoft, right? It just has — it's a conglomerate with — it's got Xbox, which I'm not a total expert on, and a whole bunch of pieces. I think for — so I don't have any opinion. I think stepping back for a minute, I think what is troubling is Microsoft guiding that Azure growth is decelerating. And I think anytime you see deceleration at all, right? And Rory can say, 'Jesus Christ, how can you keep growing 40% at this size, right? Law of gargantuan numbers, right?' Sort of massive. But I think just as a student, the only way you can — your stock price is going to trade up is if you beat, raise, and grow. You have to accelerate in this market. And you can be an oldie like Okta and Twilio and just your stock can blow up, or even Navan if you reaccelerate. But the market's expectations are so high that, you know, guiding to 37% from 40% growth is a fail. It's a fail for Azure, and it's just — this is why maybe that regulation is coming for those Chinese open source models because we just can't afford the deceleration. We just can't afford, you know — so, but it's not the only — Azure is not the only factor, right? But it's so important, and it's almost a canary in the coal mine that if there's any issue in Azure, we should at least reflect on why it is, right? If there's any slowdown in this era where we're claiming we're all going to be running 20 agents 24/7, Azure should be — shouldn't it be accelerating even at this scale?

Well, yes, because — I mean, again, because as I mean, I cited developers and end users and knowledge workers, but you're talking about the separate aspect of the business, which is the cloud business. And you're right, and yeah, the Azure cloud business deaccelerates, but I think more fundamentally, a huge slug of the Azure cloud business is simply just inference for, you know, other providers like OpenAI. So yes, I agree.

But it should accelerate, right? That should accelerate.

Yeah, if you recollect, and this is a harsh comment, three years ago Satya did the 'we're going to make Google dance' comment, and the truth is since then Google's massively outperformed. And for all their faults, and I still think they have significant issues and risks, they at least have their standalone model and a product to sell. The truth is today Microsoft has a 30% ownership interest in OpenAI and doesn't have a state-of-the-art frontier model themselves. That's a big difference.

People like to bet. And you know, the US had a prohibition, and we talked about this last week. They had a prohibition on gambling and sports for the longest time. Poor Pete Rose never got into the Hall of Fame because he bet on himself. And now we have, you know, a couple hundred billion dollar a year industry doing exactly the same. So they're just riding that trend. And they, you know, they got other products. It's not all sports betting, but it's about, I think, 70% plus sports betting. And, you know, it's killing it in revenue. So it totally makes sense. Will Kalshi be a hundred billion dollar company in 12 months' time? I doubt it. I mean, look, you get there one of two ways. Either sports betting continues to expand and they can —

take disproportionate market share. That's one option. Or the other at that is the whole crypto perpetuals business, which I just really started to learn about, which is effectively, you know, kind of futures on crypto, which is, you know, which is which is gambling to the end, right? Either that business turns out to be much bigger than we realize, which is not impossible, right? I mean, look, ICE, not the ICE, but ICE as in the inter the owners of NASDAQ, the I think Intercontinental Exchange has a big ownership stake in Polymarket. Why that's relevant is that's a company that runs ICE, that runs stock trading and, you know, real financial markets, taking a significant, I think around 20% ownership interest in the other big kind of online betting prediction market player, Polymarket, and they're clearly saying this is a huge thing that can have 50 or 100 billion dollar kind of valuation. So it's not crazy, but it happens one of two ways. Either A, sports betting gets ultra huge and they don't get tied up by regulation and the whole issues around that, or B, the non-sports betting side of prediction markets, in particular the financial products, becomes huge. I don't think predicting who's the next who's going to win the general election in the UK or who's going to be the next president in the US is a huge, enormous, ginormous business. I love it. I find it fun. It's actually the most interesting part of Polymarket and Kalshi, but that's not going to get you to 100 million bucks because the number of people who actually really want to bet on the next president is actually pretty low. It's either sports betting, which works because we all love sports. Good luck tomorrow, Harry. And then or it's financial betting, which we all love betting because we all love money. And I'm sure if you, you know, you've got sports, you've got money. If you could bet on sex, you'd have the trifecta. It's the human desires. It's the big to get to make a big picture. To get a big ass company, you need to cater big markets, and predictions on politicians is a small market. Betting is a big market on sports. People love sports. Betting on money is a big market. That's my point.

Dude, I totally agree. I did a deal into Fomo, which just got their new round done by Index and USV, and their new product is Perks, which obviously allows you to do much simpler, you know, up or down on stock prices for consumers.

No, I mean, you know, poor Warren Buffett is like, "It's time for me to die because you people have lost the plot." But if you want action, if you want action on the table, then Perks is your boy, you know, right? If you have a strong feeling on where the market's going the next hour, then go for it. Also, also like when you look at TAM expansion, Perks is a brilliant way to expand a TAM to a mega mega market. Like,

I mean, worth pointing out that the anti-AI IPO is about to happen today. Lily, we're recording this on Tuesday, I think June 30th. Bending Spoons is going to go public tomorrow, July 1st. And this will come out on July 2nd. And literally, Bending Spoons is the anti-AI IPO. It's an Italian company and they own AOL for God's sake, right? Which by the way, I remember has killed their AOL email product, but still is an ad network. They own Evernote. They own a bunch of stuff like that. So that's a company that literally is 20-year-old software companies going public at $20 billion. So it's not like the IPO market is shut, right? Which is what I thought you were saying. What you're basically saying is, is the hidden question here, Harry, has the volatility around SpaceX made OpenAI and Anthropic nervous about going out? Is that really your question?

Yeah, 1.5 in trailing, but 600 million in Q1. Yeah. So probably, yeah, 8, 9x forward revenue. Yeah. Healthy multiple.

It is funny because you're exactly right. It's like you have a whole bunch of single product crap, you know, B2B SaaS companies that have slowed out to 10% growth and are trading at three times revenues, and then you have this company which is an amalgamation of a whole bunch of tired consumer products that were growing at 10% until these guys took them over. And when you read the S-1, you realize they don't get organic growth, they don't get new users, they just raise prices, cut costs, and suddenly that's being valued at, you know, 8 and 10 times revenue. So it definitely feels kind of a little weird. It would be like as if all the SaaS, it's not quite the same. It would be like all the SaaS companies that were slow growth were trading at two times revenues and Constellation Software, which is a roll-up of SaaS companies, was trading at nine times. It definitely feels like, wow, that was a clever way to make money, right? And you do kind of, you know, it's a well-executed strategy maybe, and it's worked. Clearly there's clearly value being created. You do wonder is it a little frothy, but you know, good luck to them. I mean,

I think it's going to do well. I think over the medium term. Here's why. I mean, listen, God, don't take risk factors in a prospectus seriously, right? Or whatever prospectus said. But there's a grain of truth. They said there's at least a thousand businesses they've already identified that are material, can move the needle. I think if they're this good at buying, repackaging these companies, and there's a thousand targets, they can maintain outlier growth rates for longer than we would be in the stock, right? So, can they rate? I guess the meta question is going back to the Microsoft issue. I know Rory's going to think it's an odd tie, but can they maintain this outlier growth for five plus years, right, to justify some sort of premium? We can debate whether the premium's too high, but outlier growth gets a premium. I think for without question, if they can execute at the level they have, it's justified for 5 years because there's a thousand targets. And I think there's also going to be another Bending Spoons in B2B that's going to nail this. They're going to buy these horrific products like Marqeta and others, put a few smart people in it, and just boost the NRR. They're just going to boost the NRR and a few other things. And so, I think there's a chance for several Bending Spoons to take the struggling software companies and do a revenue arbitrage because they can package them together into something high growth. I think there's five good years here.

I think that's smart, Jason, because you're exactly right. Like, because if you're one of those thousand privately held companies, and let's assume they're all consuming, there's nowhere else to go. I mean, that's the big attraction. These guys are the only way out. So they can lock, at least until someone else comes along, they can lock in some kind of rev arbitrage.

I suspect, and I don't mean to be unfair,

I mean, Bending Spoons has one trick, which is it uses a lot of folks in Italy that are lower cost for engineering, right? But I assume the other trick it has is it's able to incent GMs to do a decent job on these. Like find kids, and then these kids may be 60 and not 16, well, 26, but find kids to run these Eventbrites and AOLs. And let's assume you have a steady enough flow of these kids they give a [__]. Okay? Then I would start to buy up almost any B2B company with nine figures in revenue with a sticky customer base because I mean literally our worst, I know I talk about our worst, and the worst product that we use today is our most expensive product. It's Marqeta. Rory will remember it. And they threaten us. They threaten us. The API doesn't work. It breaks every day. They just told us they're raising prices 20% next year like Bending Spoons but without any features or functionality. The site went down for a full day the other day. You put a kid in here that gives a crap, okay, and you say, "I just want you to take whatever Marqeta still has at Adobe, 300 million of revenue left, right? It's probably decaying. Take that 300 million and give a crap. Don't threaten your customers. Like actually launch some features. Like remove the rate limits on the API. Like do something so that to retain your base." It would work. Like even our worst software, we would stay. It's not that hard. So I would take all these ones that have a sticky base and buy. And listen, you then you take 10 folks at 200 million. You got a $2 billion business growing 30, 40% and stack it. You got something pretty nice.

But the problem is they're just putting mediocre people in charge of these, like the non-Bending Spoons are having PE firms put these recycled mediocre executives in charge of it that are going on learning tours for 90 days and coming up with no ideas. That's just not going to work.