20VC：SpaceX以600亿美元全股票收购Cursor｜Stripe押注80亿美元OpenRouter｜Anthropic首次盈利与实现6000亿美元收入背后的数学？｜Lovable和Higgsfield完成超大额融资

Your gross margin problem is my revenue opportunity for my Colossus cluster. Pessimists sound smart, optimists die rich. Only a fool denies that Elon Musk is wildly effective.

I would much rather initially work for Elon than for Zuck. I think it'll be like the Scale acquisition. I don't even think this product will exist in 5 years.

You can't add expenses below the line fast enough to stop yourself making money. The only thing that matters will be the growth rate and the 2027 and 2028 projected revenue. Someone who was hired with a million-dollar package in 2023 ended up making $51 million 4 years later. There are a billion knowledge workers in the world. Absolute bollocks.

I think we'll give each of our best engineers $100,000 of tokens. I had 2 board meetings in the last week where they finished the roadmap for the year. They're into 2027. If you're not deep into your 2027 roadmap by August of 2026 in the agentic world, your team is not good enough to survive today.

Base44 is where that wall disappears. You describe it, yeah, Base44 builds it. Apps, websites, AI agents, real working products built in minutes using nothing but plain language, and it's all batteries included. The back end, the database, the authentication, the hosting. The heavy lifting is handled, so you just really stay in the flow. This doesn't just take the busywork off your plate, but it gives you an advantage and pushes you past what you thought you could build alone. So in this market, fast is the baseline. To win, you just have to be first. Base44 is that edge, the move that skips the troubleshooting and gets you straight to the breakthrough. Build your next thing at base44.com. That's base44.com. While Base44 turns ideas into apps, Plaud turns conversations into insights. Founders and operators spend way too much time every week jumping between meetings, investor calls, brainstorms, customer conversations, and then trying to piece everything back together afterwards, and that's why I've been using Plaud. Plaud instantly captures conversations, voice notes, meetings, random ideas with one press and then turns them into clean summaries, action items, mind maps, and searchable notes that you can actually come back to later. Honestly, it feels less like a recorder and more like an AI-powered, like brain memory system. And the crazy part is the hardware itself. The Plaud Note Pro is literally as small and thin as a credit card, so it's just always with you when something important comes up. There are already more than two million founders, operators, investors, consultants, and professionals using Plaud to stay organized. So think more clearly and stop losing great ideas and important details. Go to plaud.ai/20vc and use the code 20VC for 10% off. That's P-L-A-U-D.AI/20VC and use the code 20VC for 10% off. While Plaud captures the conversation, Fin helps continue it. As AI agents become more common in customer experience, teams often end up juggling multiple siloed tools for every job. Well, Fin was built to change that. It's a single unified agent that works across your entire customer experience, from service to sales to success and beyond. Fin is the agent making perfect customer experiences possible for thousands of customers. It's powered by custom models, trained on years of real customer interactions, so it understands the nuance and complexity of customer service better than any other agent. That means faster resolutions, more consistent support, and just better experiences for every customer. It's also designed to be fully self-manageable, so you can easily improve and adapt it as your business evolves. No third parties required. Leading companies like Gamma, Asana, DoorDash, and Crypto.com already use and love Fin to deliver better customer experiences. So for a limited time, you can get $500 a month in Fin credits for your first three months. Learn more at fin.ai/20vc.

You have now arrived at your destination.

Boys, we are back. We have some mega news this week. SpaceX closes the $60 billion all-stock takeover of Cursor, minting 1,000x returns for the likes of Annie Partovi and Neo. OpenAI's Startup Fund is a mega-winner, having invested $6–8 million very early. Rory, I thought of you when you said before that Elon Musk was giving Sam billions of dollars through gritted teeth. Many others—Thrive and Andreessen, most importantly—netted huge returns. What do we have to say on this one? It's the closing of an already-announced deal. Thoughts?

1. Cursor's Unlikely $60B Journey

I just had 3 things. One is that we've been doing this show for about 70-something weeks or so, and so much has changed. In the middle of it, or maybe a third of the way in, it almost seemed like Cursor was dead. None of our portfolio companies were using it. I don't even remember if Claude Code existed when we started this show. It might not have.

It sounds so crazy. I'm not sure it had launched. So Cursor was super easy. It rockets to half a billion in revenue. Claude Code comes out, and it seems like everybody's moved, like Cursor's dead. Cursor goes multi-model really early. It changes everything and rockets to a $60 billion outcome.

I mean, forget that it started as an email client. That fun little thing from Hacker News. I can't imagine what a rollercoaster it was on behind the scenes. This was not actually 100% linear progress to $60 billion. Pretty crazy, I think—the rate of change.

The one thing I kept thinking is how important it is to be beyond agile because I think so many teams would have given up on that journey. “Oh my God, fucking Claude Code came out? I have to build my own LLM.” It's so hard to keep up with the rate of change, and there's probably been 3 different Cursors since we've started, other than the email client. That was my main thought.

The second one is how it ended up not even being that expensive by the time the deal closed.

And you say that because it's going to be at $6 billion by the end of the year, and then you're paying 10x?

Yeah, you're paying 10x forward revenue. Earlier in the show, it looked like it would be gross-margin-negative. When we started the show, it was like, “Well, Cursor's a joke because they're selling a dollar's worth of tokens for 80 cents or 50 cents to the world.” Of course it works, right? This was the classic thing that VCs would mock when we started this show, and it was true.

Turns out, with open weights and everything, it's a pretty darn good business model, selling at 10x forward revenue. Elon was a shrewd buyer.

2. Why Elon Bought Cursor

I'm intrigued why Zuck didn't buy it. He's building the model capabilities with Alex and co. He's missing the enterprise capabilities. This would've solved that in a similar way to how it solved it for Elon.

Yeah, interesting tangent. If the logic is, “Hey, you've got a whole bunch of compute but not an obvious business on top,” the 2 people for whom that's true are Meta and SpaceX. Anyway, SpaceX did it, and Meta did not. At a high level, it's a fair point.

One of the things we pointed out in the agenda is that they didn't do one of these weird acqui-hire things. They just bought it like a regular old corporation. They didn't have any antitrust issues. They actually filed for antitrust, and they got quick clearance. They didn't have any compelling issues with that.

I don't know if the seller would've had the same confidence that Meta would've gotten through, just given their DOJ issues. They probably are going to have the DOJ crawling through more of what they do. So that would be one argument.

It may also be, to be very direct, that no one else had the stomach for the bet in the way Elon does. Meta is relevant here in terms of fitting the characteristics of lots of compute and no obvious business on top, but Elon has the biggest advantage, which is that his stock is trading high.

I mean, it's $8 billion in revenue last quarter. Call it $30 billion growing. Round up to $50 billion. What the hell? It's still 40 times revenue. Picking up a big asset, as Jason says, at 15 times current revenue—maybe less than 10 times year-end revenue—is dirt-cheap for him. It's net accretive on day 1 in a way that probably wouldn't be as true for Meta.

I haven't thought about it all that much because, frankly, until you mentioned it, I hadn't thought of that. But one of the things for me is, if you zoom out, and just reflecting back on Jason's comment on the gross-margin negativity, that's true. That story was true when it happened, and it's still true today. There are challenging margin issues.

When you reflect on the journey, the negatives that you can cite along the way tend to be true. They're correct. Those were real, and when you look at an investment, the positives in terms of market trajectory just outweigh the negatives.

I mean, it's easy to sound very financially smart and say, “Oh, in the end, everything has to generate free cash flow. This doesn't have gross margins, so, quote, ‘In the end, it's worthless.’” But it's wrong because along the way, when you have a market that's exploding like coding, that's a huge market.

Remember, this is the biggest market for AI. 70% to 80% of Anthropic's trillion-dollar market cap is predicated on this. If you have the number-two player in that space and you're growing hyper-fast, then even though you have gross-margin challenges, especially in an optimistic, forward-looking market, the buyer is going to look past that and say, “There are only 1 or 2 ways to play in this space at a meaningful level.”

It was just a perfect fit. Frankly, in very different capital markets, it could have been a very different story. There could have been no capital, concerns about gross margins and free cash flow, and maybe you would have had to slow down and follow a very different trajectory as Cursor. But in this market, they were able to go balls-out, for lack of a better word, have those tough gross-margin stories, and then find a buyer who was not only willing to look through them, but who actually had every ability to solve the problem.

He was basically saying, “Your gross-margin problem is my revenue opportunity for my Colossus cluster.” It just shows that the negative issues didn't go away. They just got swamped by the optimistic take, and that's why it's the old cliché we talk about: pessimists sound smart, optimists die rich. Jason said it. There were probably some very tough days, but they had the guts to keep moving forward, and because the market's huge and because, frankly, the environment is risk-on, they've had an amazing result. Good luck to them.

And on the Meta thing, it's just a detail. I guess it's a parallel-universe question, but Zuck would have had to pay $70 billion or $80 billion in a week. That would have to be very core because, don't forget, Cursor was about to close a round: $2 billion at a $50 billion valuation, from Andreessen and friends. Elon did what you have to do in that situation. What does it take? He bid $10 billion more. You've got a deal at $50 billion.

Zuck did Instagram and WhatsApp in an hour on the back of a napkin and paid high, right? But I think Elon did even better. What is it going to take? They were already working together with Cursor. They were already working together: “We're going to do the round at $50 billion.” “How about $51 billion?” “No.” “How about $60 billion?” “Okay.” “Well, the deal might not happen.” “What if we pay you $10 billion if it doesn't happen?” “Well, okay.” I've removed all the objections from the deal, and I'll let you run the company the way you want.

I think it was 3 points, and they shook hands and did the deal. Zuck can do the same thing. He's done it at least twice, but you have to want it badly to do it. At $80 billion, he would have had to move even faster. Rory is of course right: Elon had the ultimate stock and currency to do this deal, and the ultimate match.

But to do any of these deals, I think you have to be Elon or Zuck because you have to strike this deal in a week at $60 billion. Only a handful of people can do this. Only a handful.

Agreed. I was just reading Noah Smith, who's a Democratic, left-of-center but moderate centrist blogger. He wrote a great piece about a year ago that said, “Only a fool denies that Elon Musk is wildly effective.” Regardless of your opinion on the merits, he's possibly one of the most effective people on the planet at getting shit done when it comes to industrialization, physical AI, and AI.

From a standing start a year ago, he built the cluster, and then he bought the product to stay on top of it. He took SpaceX from, literally, a year ago being a really amazing rocket and satellite-to-cell connectivity story to being, as their S-1 says, at least in terms of the “future prospects,” 89% an AI story.

You saw his tweet. You mentioned it a year ago: “I've underestimated AI,” or some version of that. “I've underestimated AI. Time to go.” He went from a standing start to owning more compute than pretty much anyone else, and owning the most important product to fill that compute, in less than a year—just over a year. That's wildly effective management. It's world-class in getting shit done.

The other small factor, just thinking about it—we can move on—but imagine you are Michael at Cursor. Things are going pretty well. You've got a term sheet from your, what, $24 billion? You're a paper deca-billionaire, and then Nvidia and Thrive want to put in money at $50 billion.

You're not that cash-motivated. You could take out $1 billion, or $500 million. Things are going okay. These are very rare deals, but even though working with Elon in a year might turn out to be terrible, I would much rather initially work for Elon than for Zuck, personally. I would do it. I'd be like, “Elon is the guy I want. If I had to work for someone, it'd be Elon. He is better than me. This guy does fucking rockets and electric cars—he does everything—and Zuck's firing everybody and going crazy because he doesn't have an LLM.”

Not that he isn't one of the greatest entrepreneurs, but I wouldn't want to work for Zuck. I would want to work for Elon, and that actually matters. It may be a mistake in M&A, as a target and as a CEO, because we've all made mistakes here when we've been founders on the other side. I tell founders to ignore it. Ignore the brand. Ignore what you think the job is today, because you have no idea what the hell you're going to be doing in 24 months.

But it is incredibly emotionally important to founders to land in something they want to land in, and I would not want to land at Meta today. If I were Alex at Scale and I got $24 billion and had a tough business, maybe. But this one, man, I'll take Elon over that one.

Do you think Amazon or Meta go, “Eh, we'll take Cognition instead”? Is there a knock-on effect for the second player in the market, which I think arguably now would be Cognition?

I think it's interesting you cite those 2. There's actually a quality of absolute imperative to do something. It's about who has to. I think SpaceX had to, because they had all this compute and it looked like they had to fill it. Subsequently, they've also been able to rent that compute to Anthropic and Google. I don't think it's nearly as existential, for different reasons, to the 2 you named, and I'll name 1 for whom it is.

For Amazon, they're in the AWS business. They've got lots of contracts with Anthropic. They basically have the compute for Claude Code, so they're getting the inference-side revenue for that. They don't own the model, but it's important, not imperative. They've never done a $60 billion deal. I doubt they're starting now.

And Meta, to me, is again unlikely, and the reason they didn't do it earlier is that their core business is an ads business. It's freaking amazing. This is literally: I have a wonderful ads business that kicks off north of $100 billion. I've chosen to do this new AI thing. We can pretend it's strategic, but really, I'm just fucking really interested in it. I don't know if you have to do another $60 billion deal on top of that one, so it's not as imperative.

I think, just to put it out there, I'm going to name the one company for whom this market matters: Microsoft. Remember, Steve Ballmer has long since gone and owns a basketball team, but he would jump around the stage, sweating and screaming, “It's developers, developers, developers.” The fact that they've lost that connection with developers, that GitHub is now a trailing-edge product, is, to me, over the medium term, a pretty significant loss.

Operationally, the numbers are fine. It's a well-run company. But if you wanted to name the company that should want to own a leading, state-of-the-art coding product in this brave new world, clearly the number-one company is Microsoft. The antitrust issues would be a longer discussion, but I don't think owning the developer is existential for AWS. I definitely don't think it's existential for Meta. It's 100% existential over the medium term for Microsoft.

There was a fantastic tweet that said, “Satya should buy it and then make Scott Wood CEO of Microsoft.” I thought that would be a rather ridiculous thing—to replace himself—but I actually thought Scott would be a rather brilliant CEO of Microsoft.

My experience with big-company M&A is that the idea that—Harry, what did you say?—someone else feels like they have to jump in to respond to Cursor, that everyone else...

Yeah.

...my experience is that doesn't happen. It doesn't actually get everyone else to say, “Hey, I've gotta go buy Cognition.” My experience is that usually the other thing happens. I wanted to buy Cursor. I lost Cursor, or I didn't even know I lost Cursor 'cause Elon swooped in. Now it pushes it up my existing priority list. That's how number 2 and number 3 get bought, not because there's a panic for land grab, but because I didn't get what I wanted. That's when you gotta be really thoughtful as number 2 because, uh, that's when you get bought as number 2 and number 3. When just number 1 just gets taken off the table. It's not so much a land rush. A lot of times the acquirers are like, “I thought I had more time.” Maybe Satya's like, “I thought I had more time with Cursor. Andreessen was gonna do it at 50. I could do it at 100 billion next year. I wanted to wait and see.” And they thought they had more time, and they didn't, so then they go buy number 2. The couple times I've sat on the other side, I don't know that it creates such a strategic arms race that everyone just gets picked off instantly. That might be a VC partial myth.

3. Stripe Buys The AI Router

It’s not quite the $10 billion that was reported, but OpenRouter is a $7 billion acquisition by the Irish Collison brothers, Patrick and John. I mean, what an incredible journey. Alex Atallah, CEO, who I just had on the show—he founded OpenSea before. It’s obviously the leading LLM routing company.

It raised a Series B at a $1.3 billion valuation just 4 months ago, so it’s 5X that for CapitalG. It’s a 12X for Menlo and Andreessen. How do we think about this? It’s widely reported. Now it’s confirmed.

You can see intuitively how Stripe gets there from here, right? When you look at their existing business, they get paid a small percentage of the money flow to manage the complexities of collecting cash via cards and also via ACH now. Here, they’re going to get a small amount of the money flow to manage the complexities of picking models and running, as an enterprise, using a single API to run tens and maybe hundreds of different models.

At the conceptual level, I can see it totally makes sense. A lot of their lift recently, even on their payments business, has come from their customer base being so AI-forward that every time you spend money with OpenAI or Anthropic on a credit card, they get some of that money. So I can totally see how they get there, and again, it’s some version of the same thing as the Cursor comment.

You can do the old intellectual exercise: “What are the barriers to entry for this business? There will be lots of people over the medium term.” There are a ton of weeny router companies out there, and everyone’s building one. But it turns out that in an early land grab, when people are moving—the market here is growing 10X year on year if you take Anthropic’s growth rate as the big-picture comment here—if you move early and build a useful part of the infrastructure, you will probably find an acquisition at a price that doesn’t make any sense on a DCF-to-you basis but makes huge sense to the acquirer.

Just like Elon will turn Cursor into money and cash flow far quicker than Cursor could have turned Cursor into cash flow, I’m willing to bet Stripe will turn OpenRouter into money probably quicker than OpenRouter could do it on a standalone basis. This is what happens in a crazy market. If things slow down, a lot of these acquirers would run the buy-versus-build analysis and say, “There’s no hurry. We’ve got 5 years.”

When things are moving as fast as they are now, you’re going to see, in my view, until such time as you see the correction and the acquirer currency diminishes, a whole bunch of people like Anthropic say, “Screw it, I want to be in world models. I’m just going to buy the code. I don’t have time. Screw it, I want to be in AI. Screw it, I’m huge in payments. I want to be in the AI inference flow. The quickest thing I can do is spend $7 billion, some of it in stock, get these guys, and be rolling in a week.”

This is what you see. You saw it at the early internet stage. You saw it back in times even before that that would make you cry, Harry, if I even mention them. When things are moving really quickly in a build-out, you just see these amazing acquisitions where the value to the acquirer dictates a very healthy price. It’s one of the reasons why venture works. You go right out there on the risk continuum, but if you time it right, you can get these kinds of returns. Well done, a16z. Well done, Menlo. Well done, CapitalG.

It’s a reminder of how weird revenue is in M&A. If you’re bought by PE, the top and bottom line are incredibly important, down to the significant digit, down to cell G38. If Workday goes private, exactly what its DCF will look like in 2032 is so important.

The weirdest thing about M&A with big companies is that revenue is so important to argue over multiples and the price. It’s so important to price, but it’s irrelevant, because it’s all about what Stripe can make out of OpenRouter. It’s such a weird thing that your revenue going into big M&A actually doesn’t matter at all, even though it’s probably the biggest input to price.

Stripe literally—what’s OpenRouter doing, $70 million, $80 million today? Stripe does not care about that money. You often see acquirers abandon even the existing revenue, right, to do the revenue. It’s just such a weird paradigm.

Two things. Rory’s right. Stripe actually appears to be very good at acquisitions. It’s how it accelerated into crypto and otherwise. They’re good at it. The flip side, you could say, is maybe they should be better at building these themselves, right? That’s the grouchy version: Why didn’t you build it?

But if you’re good at M&A, and this is 5% of your market cap plus cash, and you want it tomorrow, it makes sense. You have to be good at M&A, though, and then you do it. The counterpoint is that I love OpenRouter. I talked about it on the show 6 months ago. I’m a customer. I’m a user. It was brilliant.

It was one of these pieces of software, like ElevenLabs, that was instantly easier to deploy. It was elegant. It was a beautiful piece of software. But it’s pretty niche.

In what sense, Jason? Just a genuine curiosity.

Okay. So OpenRouter, as I understand it—and I think this is right—is really strong in developer-type tools where you want a simple way to pick a model. You can pick any model. You don’t need to get on Fireworks. You don’t need to set up anything.

It’s really, really, really strong with chatbots, where they don’t have to be perfect. When you’re talking with my digital Harry or digital Rory, you don’t need perfect outputs, right? You can route between models based on availability and cost. Those are their 2 niches.

Now, let’s talk about workflows with a lot of reasoning. For B2B, when it has to be accurate, you’re going to down-spec to 1 or 2 models because you can’t have model drift. You can’t be routing from Kimi to Qwen to GPT-4.6 to Fable, and all of a sudden your B2B workflow that has to be perfect drifts from all of them. It drifts.

Even just going from 1 Opus model to another, you see drift. You have to QA it, requalify it, fix it, and test it. So for high-reasoning models, people do frontier-esque outputs, right? People don’t rotate through 11 models, and I don’t think OpenRouter is the right product for that, and that’s fine. They get that too.

But I think Stripe is saying, “Hey, listen, any transaction on planet Earth we can take 2-point-something percent of,” right? But that’s not going to be true for OpenRouter. It’s a niche, wonderful niche product, but in the world of routing—which everybody does, Databricks does, Replit does it, Lovable does it, Vercel does it—it’s a niche product with 2 really good niches.

This is the risk to Stripe: that they end up owning a niche, successful niche product, and that’s not their DNA. Their DNA is not niche. It’s just not their DNA.

And first of all, I do think that’s fair, because we internally agonize about this space, and that was exactly what we were angsty about. Your framing is exactly correct.

Let’s spell out the positive trend. As long as you have the frontier models trying to extract $100 billion in revenue from you this year, and you’re an enterprise, you’re going to want a Plan B, right? At least to keep the thing honest. So you are going to want some kind of routing.

But what resonated with me a little bit, Jason, is what you said: Remember when cloud was starting, people were like, “I want to be multi-cloud”? It’s really hard to be multi-cloud. Here, maybe I want to be multi-model, but maybe I only want 2 or 3 models, and therefore I don’t need this kind of routing functionality. That is the risk.

If your enterprise customer decides, “I need to flit between 3 models but not 10,” then you’re right, your value here goes down. I would imagine the positive spin is that your value goes up to the enterprise if you can build on top of just picking a whole bunch of models, normalizing all those options, and trying to commodify the models.

That’s the tension point. The more you can do that, and the more you can service the people who don’t care all that much, the better your business. But you’re right: If JPMorgan says, “I want more than just Anthropic, but I’m not going to qualify 10 models. I’m just going to work with Poolside as my Plan B and then offload the rest to something else,” then you’re right. You’ve niched down, and you don’t get that revenue.

For example, this week Rippling posted their view, as a B2B player, of what models they use, right? They had it all, and they said, “In the world across Rippling, we looked at 2 things that were best for us: Opus 4.8.” It’s an N-minus-one model, but it’s well-trained with their harness. Then there’s price, performance, and speed.

I think they picked whatever, GPT-5.5 Medium or something, and they said, “The rest isn’t worth it for Rippling today.” That could change in 60 or 90 days. So they down-specced to 2 at a time, and then they have to manage the outputs from these.

You may tune 1 set of workflows here, right, that are long-reasoning in another. Even at Rippling’s scale, managing 12 models is too much. If you’re a dev tool and let people pick, so be it, right? That’s great for OpenRouter.

Or if you want to build a fallback into your own product, OpenRouter is a 10 out of 10 for this. Let’s say something’s down, right? OpenRouter automatically falls back. But I think it’s a niche product. It could be a massive niche, though.

Yeah, I’m remembering the conversations now, because you are right.

The great thing about the core Stripe product is that all payments are equal, and Visa is the rails for everything. That might be the case here. I’m going to check out that post, because one of the big questions will be how much pricing pressure enterprises can put on the closed foundation model companies and how they put that pressure on, because I think it impacts a lot. Sorry, Harry, go on.

In 5 years’ time, will this be considered a successful acquisition or not? Bets on. Prediction.

I think it’ll be like the Scale acquisition. It will be the start of something that gets bigger. Whether this brand exists or whether even this product exists in 5 years, I don’t even think this product will exist in 5 years. But I think there’s a high chance, more than 51% chance, that it builds into a 20% or 30% revenue stream for Stripe, and that’s enough.

But does OpenRouter, as part of Stripe, exist in 5 years? I’ll bet you dollars to donuts that 5 years is so much time, and it’s such a niche product, that this product itself, if it does exist, will be deep in a dropdown menu on the top of Stripe, like 11 layers down, because it’ll be subsumed into the whole token management platform, right? Their TMP.

Yeah, I don’t know. I think Jason’s answer resonates with me. If it works, it’ll be seen as a TAM expansion play. What’s fun about Stripe right now is they’re doing that acquisition, which is very much, “Hey, we don’t play in this space. Let’s put a stake in new ground,” and at the same time they’re talking about a PayPal acquisition, which is very much, “We own this space already. Let’s buy these guys, fold them into what we already have, and just make a shit ton of money consolidating,” right?

Actually, I think that’s a clever strategy. I think they’re playing a very clever hand. There’s probably a 1-in-3 chance that they have a massive AI routing business in 5 years, but if they do, that’s a big second leg. At the same time, if they get the PayPal deal done, that’s the kind of deal where you have a high degree of visibility that you keep those revenues and remove the entire G&A, if you’re good at M&A and good at consolidation.

You get more of a two-sided network because you have consumer wallets, which Stripe doesn’t have, and you’ve done core consolidation acquisitions. Doing them both together, provided you can pull them off, is super interesting in terms of building enterprise value. And they’re doing it all privately, again back to the comment: doing what look like public-company-sized M&A and pulling it off while private.

I know they got the investors to take Stripe stock in the OpenRouter deal. I think some portion of it was stock. The PayPal deal is more complex and probably requires more thought. But again, being able to do a—what is it? I think a $40 billion or $50 billion deal and a $7 billion deal, issuing paper while private, is pretty impressive.

Stripe’s corp dev team need a bonus at Christmastime. They are busy this year.

They are busy this year, but isn’t everybody?

4. Anthropic Turns Profitable

We mentioned margin pressure on foundation models. Anthropic turns its first profit on $11.5 billion of Q2 revenue. The business is getting better for Dario. This was also in a week where Gavin Baker said about Dario that he believes they will be the final private company. Did you see this?

We did, and let’s separate the hyperbole and the future from the facts and the present. It’s not surprising they’re making money, right? If you just go back to last year, they did $4.5 billion last year, and I think their gross margins went from negative the year before to roughly 30%, right? On track, I think, for roughly 40% at the end of the year.

When you have decent operating margins, like 40%, and you go from $4.5 billion in a year to $10 billion in a quarter, that means you have $4 billion of gross margin, right? And that’s literally in 2 quarters. You can’t add expenses below the line fast enough to stop yourself making money. So it’s inevitable.

They 12x’ed growth, which means they probably 14x’ed gross margin if it continued to increase even slightly, and the trajectory has been increasing. You’re not going to 14x headcount or below-the-line training costs in 6 months. So yes, I’m totally not surprised they are operating in the black.

We had run numbers at the start of the year, and it kind of came to that conclusion. The interesting thing will be, as they continue to grow and buy that expensive compute from Elon—if you remember, that had a big price increase 2 months in—I doubt they will forecast for their IPO a base case of continuing profitability. I could be wrong, but this profit didn’t surprise me.

It’s amazing performance. It’s amazing revenue. Revenue with any kind of decent gross margin cures almost all else.

I think the other question is, as we gear up for an IPO, which could be very imminent, is what numbers does Anthropic get away with? You’ve got off-balance-sheet liabilities. You’ve got massive commitments. You’ve probably got stock-based compensation like we’ve never seen in the history of mankind.

So if you get asterisks and daggers on your numbers, they will be jaw-dropping. If they have to fully account for that, and some of that’s non-GAAP, these off-balance-sheet items—if they’re going to be hammered like a poor Wix or someone for SBC—everyone’s going to write up the horrific downside, right?

But I think everyone’s going to look through all the nerdy negative things you could see in the numbers. They’re just going to ignore it, right? I do think it’s important that it gets ignored. I think it’s important for Anthropic that it gets ignored.

And I think none of that shit will matter, to use a technical term. The only thing that matters will be the growth rate and the 2027 and 2028 projected revenue. Provided the revenue comes, everything else will be fine.

If the revenue comes, then you’ll need the off-balance-sheet stuff, and you’ll have the revenue to buy it. In other words, all this off-balance-sheet stuff is basically, “I promise to buy a whole load of compute from you in 2 years’ time, because if my revenue grows 10x for 2 more years, I’m going to need all that compute.”

Well, if the revenue grows, you need the compute. You’re happy to have it. In fact, you’re insisting you get it. If the revenue slows down, then you don’t need the compute. It all gets hard.

So almost everything is going to boil down to what number do you underwrite for the next 2 or 3 years. Then, as you said, the stock-based comp—no one’s going to care, because the reason you worry about stock-based comp is because, in a steady state, like Workday—we can cover that in a second—if you’re giving someone $500,000 every year to show up and be a middle manager, they’re probably mentally putting those RSUs into their comp, and they think to themselves, “I’m paid $400,000 in cash and $500,000 in RSUs.”

If you stop giving them the RSUs, they’re going to want cash. So it really is a cash number. In a mature business, it’s totally correct to worry about SBC. But the SBC numbers here are going to be huge because all these people got grants, and then it turned out to be worth way more than they ever thought.

The classic example is someone who was hired with a $1 million package in 2023 and ended up making $51 million 4 years later. That doesn’t mean you’d have to pay the next guy $51 million. If that person had gotten the $1 million they signed up for, that’s all the real economic stock-based comp it takes. The other $50 million is just dumb luck. You got lucky. It’s not a run rate.

I actually think it is okay in a hyper-growth company to look past a good slug of the SBC and normalize it out. Conversely, it’s not okay in a mature company. That SBC, stock-based comp, in Workday or Salesforce—that’s real money that people are spending.

It’s a little bit unfair because you’re giving the hyper-growth company a free pass, but they get a free pass. You get a free pass. It’s the same thing we said about Cursor: you get a free pass on margin, you get a free pass on off-balance-sheet liabilities, and you get a free pass on SBC, provided revenue goes up. Once revenue stops going up, all bets are off.

5. Anthropic's Massive TAM

Once revenue goes up, all bets are off. What would it take in usage for Anthropic to hit the $200 billion ARR plan for 2028, and then $600 billion the next year?

The simple version is: how many knowledge workers are in the world? How many folks can take a subscription? Being generous, is it 1 billion human beings?

If Anthropic has 100% market share at $200, that’s $200 billion. If Anthropic has 300% market share, that’s $600 billion. I don’t know. Rory’s thought more. The $600 billion seems complicated, but our demand for AI has only just begun.

You can see $200 billion, which is a number, right? Once you start getting to the $600 billion number, it gets really hard, because no one ever looks at the big number. I’ve just been doing some work on this. No one ever steps back and looks at the big numbers.

You said 1 billion knowledge workers in the world. Absolute bollocks. Hard-nosed comment here: the US is typically 50% of the world’s software budget because we’re 50% of the world’s high-end knowledge workers. We’re 25% of the world’s GDP. So, at a minimum, if spend tracks GDP, it’s only 4x the US.

But every software company is typically 2x the US. Why? Because the rest of the world can’t afford the same software we do, because they’re poorer and they have more people at lower wages and less software.

That’s why we have crappy internet when we go to Europe. So the truth is, the hard-nosed comment is this: you probably take the US knowledge-worker spend and double it. There are 83 million knowledge workers in the US, and then roughly 86 physical labor workers. So that’s what you start with, and you start cutting it down.

I literally was doing the math this weekend, thinking about it. You start cutting it down real quickly. The truth is, knowledge workers include everyone in healthcare. I don’t think we’re going to replace the nurses. It includes the teachers, right?

The sweet spot—the uber-sweet spot of the whole damn thing—is that there are about 1.8 million people doing coding in the US. Including QA and all the other bits, there are around 5 million people who do software-related shit—systems admin stuff, all the rest of that—and they get paid, in total, grossing up to about $600 billion a year. $200 billion means you’re replacing a third of them. That’s a lot.

And remember we said this before: the single most important ratio—and I asked you about what you thought it was, Jason—is, in a steady state, what’s the ratio of salary dollars to AI dollars? Because if it’s 50% of salary dollars, you can easily get to $200 billion. $600 billion is hard in coding. Well, you can’t get there. If it’s 10%, then it’s hard to get $200 billion across the whole thing.

So it really boils down to this in the steady state: if software is the tip of the spear in terms of maximum adoption, what do you think? For every $100,000 you spend on an engineer, or $200,000 you spend on an engineer, are you going to be spending $100,000 on AI, $50,000 on AI, or $200,000 on AI? That’s the number.

Yeah, we’re testing it. Over the last 60 days, every single scale-up is capping its AI budget for real. It’s not just the Ubers of the world. Everyone’s capping it because it’s grown truly exponentially. Everyone’s capping it at $6 million a year, $8 million a year.

I think it’ll land at $100,000 per engineer equivalent. I think we’ll give each of our best engineers $100,000 of tokens, and in return, we’ll cut the size of our dev teams by 30% to 40%, effectively. It won’t exactly work out that way, but close enough is how it’s going to work out. So there’s $100,000 here for running inference 24/7 with 10 agents in parallel.

For what it’s worth, I actually agree. That was about my mental model, too. Let’s assume it’s not just dev engineers. Give the sysadmins and the QA guys the same thing. You get $200,000 of wages, fully loaded, including all the benefits, and $100,000 worth of AI, but we cut 30% of you. That turns out to be terrifyingly about a $200 billion, plus or minus, market in the US.

Next year.

No, ever.

Well, yeah. That’s Anthropic’s estimate for next year, right?

I know, they’re too optimistic.

They’re getting 100%.

But my point is this: if you count all the heads and apply the Jason math, you get $200 billion in the US, which probably means you struggle to get $350 billion worldwide. That’s the TAM. And then you’ve got to go beyond software. There is obviously revenue beyond software, but it’s nowhere near as fertile, and the percentage isn’t going to be anywhere near as high.

Yeah.

It’s funny—that’s exactly the number I come out with, because you see the Ramp data that says the top 1% of their sample, which in turn is obviously a biased sample of tech-forward people, are spending $7,000, and then the median is spending about $100. It’s amazing, the dispersion. And $7,000 times 12 is only $84,000. That’s 1% for all employees. That’s the pointy edge of the most optimistic spend: 50 cents of a salary dollar.

I think we’re going to get to $100,000. In the investments I’ve made that are the best ones—the ones growing faster, especially the ones that are pre-2022 or 2023, so they have a frame of reference—they literally are shipping 2 to 3 times faster. Only recently. That was kind of bullshit last year, right? People would say that, but it was all performative, like token-maxing.

I had 2 board meetings in the last week where they finished the roadmap for the year. They’re into 2027. These are my 2 fastest-growing, but not brand-new, companies. They finished the roadmap. They’re well into the 2027 roadmap.

So you’re going to spend $100,000 on your team to do that, but it’s adding up to so many millions, it’s overwhelming. I do really think this $100,000 makes a lot of sense. You could justify more or less. People will ratchet it, but I think it’ll be the new normal, and you’ll cap your team. That’s just what the CFOs will do: $100,000 of inference and you get to hire this many engineers.

But the idea that they’re pulling their 2027 roadmaps in—it’s not just performative, it’s not just PR. You want to invest in that up until the maximum where it works, right? But the absolute numbers are just getting really big.

Yes.

I’ve got to say, man, if you’re not that way, you’re losing today. If you’re not deep into your 2027 roadmap by August 2026 in the agentic world, your team is not good enough to survive today. This is your last chance to make changes.

Listen, if you’re OpenRouter, you didn’t even have a 2027 roadmap. It didn’t even mean anything. You’re just remaking it day by day. But if you’re running the classic playbook of, “I can get this much done each quarter, this much done each month,” and you’re not into 2027, you’re going to lose to the competition. You’ve got to be honest: how deep into 2027 are you? Not deep enough.

Yeah, Jason always gives me these terrifying soundbites that I go back and think about. We did this survey. We tend to be fact-based people. We did this survey of all our companies, and we saw something similar to the Ramp dispersion: some companies are all in, some companies are adopting, but still dramatically less spend per head. I can’t remember the average, but it was dramatically less.

And what I didn’t do—and actually, now that I think about it, I should do and I will do—is go back and see if you can find a strong correlation, which you believe you should be able to, between output and spend. Can you justify the spend? Then you should be saying to the laggards, “You’re just going to fall behind.”

6. The IPO Timing Advantage

If it goes out at $2 trillion to $2.5 trillion, would you be a buyer?

First of all, I want to be clear: I don’t think the software market is definitionally the end of the TAM. I think the average knowledge worker won’t have 50% of their salary in AI, but they’ll have a meaningful percentage. So the TAM is significantly bigger than just developers, because you have lawyers, but I think lawyers won’t—

Look, the Kirkland & Ellis guy who’s pulling $2 million a year as a partner isn’t going to be doing $200,000 worth of tokens. He’s definitely not going to be doing $1 million worth of tokens. A lawyer would die before they gave $1 million of tokens instead of $1 million of take-home pay.

So the market is bigger than software, but there’s nowhere else that’s such a sweet spot as software. I don’t want to be limited to $200 billion. But I think the really challenging thing is, I definitely want to be first out rather than second out in terms of going public, especially if you have some kind of near-profitability story or bouncing-around-profitability story.

I think it’s a far more attractive strategic position to be going out as Anthropic in the fall with a, “We’ve been profitable. Okay, we’re unprofitable again, but we’re the winner in the enterprise,” than going out next year, where maybe the growth rates have started to slow, both for Anthropic and the public markets, and if you’re OpenAI trying to access the markets then. I definitely think they’re in a strategically more challenging situation.

I think they’ve just capitulated to it. Of course, you want to be first, to your point, right? I think OpenAI has had to get their house together—more executive turmoil. Apparently, a great last 30 days, right? But in the first half of the year, it was slower than its previously junior competitor, right?

And they’ve had to do so much to say, “Listen, we’re going to go public second, and then we’ll have a comp out there, and the comp is what it is, and we may not trade with the hype that SpaceX and Anthropic did, and the world will not end.” We will trade at a very precise number. We will know what we’re going to go out at. And the world will not end if we trade at $1.3 trillion.

I just think that they’ve given up on worrying about that, because ultimately Rory’s right: it’s much better to be first, but in the long run it doesn’t matter. If you don’t need the capital, it is what it is.

But Jason, I’m going to push a little. That is the sentence. There are no 2 companies on the planet that need more capital than these guys. In a world where you do need the capital, being second sucks.

I agree with you in general—you are correct. It doesn’t matter. 2 companies go public plus or minus a year, and a decade later, no one cares. We’ve definitely seen that over the years. The thing that’s challenging in this particular case is both companies still have enormous, many-hundred-million- to billion-dollar capital needs.

In that situation, I would much prefer to be first. I do. I'm putting an asterisk around “first” there.

I think you're right, but the thing is, let's say—and pick your number—let's say Anthropic is public at $2 trillion. It really doesn't matter, right? OpenAI is going to be able to sell stock at a discount to its implicit valuation before it goes public. There's still enough capital.

Let's say they're both implicitly worth $2 trillion. OpenAI's going to be able to sell stock next year at $1.8 trillion. People will do it, and especially if you're a CEO in your own company, it's okay to sell at a small discount.

Yes, agreed. And look, I'm not catastrophizing here, but I think the interesting thing is, if you're the smaller market-cap company and you have the bigger capital need—which right now OpenAI does, because they have a more ambitious capital-need target—would you prefer to be the guy trading at $1.5 trillion who only needs to raise $100 billion, or the guy trading at $1 trillion who needs to raise $300 billion?

At some point, these things become troubling. And yeah, price clears all markets. This is the best new technology market we've seen in—ever, perhaps. And if you are the founder in that market, and even now the number two, you're going to attract capital, but you just don't know the terms under which it happens.

And going back to my comment, I think you will regret not being able to access the capital markets this year.

No, no, of course I agree. My only point is the media and social media will make a big deal out of this, right? Who goes public first and who does better. I just think Sam and the OpenAI team have said, “This is our fate.”

They could go public tomorrow, right? There are enough people to buy these shares to go public. They've decided that while this isn't perfect, this is the best on the board, and we're going to live with the doubt. It's not the end of the world. You can't solve every problem tonight. They've got to solve bigger problems, right? The cards are the way they are, right?

7. Workday's LBO Math

Jason, you said about management-team churn, that the churn, for those that don't know, most recently was Denise Dresser, who was the CRO, who left, and Dali Rajic has replaced her. For those that don't know Dali, he's one of the most respected CROs. He was a fricking master at Wiz, and I think the best CRO or sales leader in the business. Chad Peets says he's the best of the best. So I'm feeling a little bit more confident for their Codex and enterprise division today.

Yeah, it's just a lot of change. I don't know anything inside. I just think Greg Brockman took over, right, and brought in the Wiz guy. He'd just had enough of the Salesforce crap, right or wrong.

Actually, if you look across all of AI, a ton of Salesforce executives have been recruited to come in and help. And you can make fun of it. I used to make fun of how, back in the day, Salesforce hired Oracle executives because they took shots at Oracle, but you need folks that know how to scale.

What is Salesforce at, $45 billion run rate, $50 billion run rate? I mean, Anthropic's past that now, right? OpenAI is past that. So you don't want to hire kids. You want to hire someone that has some idea how to play. So Salesforce is about it, right?

But if you step back from it, I'd rather have someone from Wiz who is close to technology and in a hypercompetitive space, rather than someone asking how many seats of Slack you want. It's just a very different go-to-market motion, right? It's very different.

Jason, you said if you have not already hit your end-of-year goal in terms of product and you're not well into 2027, you're behind. I'm making assumptions. I don't imagine Workday's quite at the cutting edge like 2 of your companies are, already hitting those goals in 2027.

And Silver Lake circles a $43 billion take-private bid for Workday, one of the biggest SaaS buyouts ever. We have 2 of the best SaaS minds in the business here. Guys, what should we take from this? SaaS isn't dead. One of the biggest firms, one of the biggest buyouts. The stock popped 18% afterward. Wow.

I think what you can take from this—the “SaaS isn't dead” thing is just too simplistic—is that a very financially oriented, wildly savvy buyer is willing to bet money that they can buy this at a constrained price, lever it, and generate a return because the revenues are sticky enough to allow them to pay down the debt over 5 years. And with reasonable multiple stability, sell it on and make a 20% IRR, plus or minus. I mean, I ran the numbers. That's the bet.

So it's not, quote, dead, but what it is not is wildly exciting. What it says is SaaS—this is the mature phase of an industry when it's not about wild growth, it's not even about untempered growth. It's literally about someone saying, “This thing is growing at 13% year on year. We can buy this thing for, what is it, roughly 5 times revenues, 16 times trailing EBITDA.”

We'll probably leverage it, you know, 2 or 3 times, 4 or 5 times EBITDA, but it's going to be a big equity check. Then you run the LBO model and you say, you keep it at 35% operating margins for 5 years. You use all that cash. It's roughly $10 billion a year in revenue, so it's like $3 billion a year of cash. You pay down the debt and the interest and, provided you buy right, you can make 20% and almost a 2x over 4 or 5 years.

I look at that deal and I go, I'm torn. First of all, I think Silver Lake are wildly smart. It's very interesting when you run the sensitivities. If you pay, like, 20% too much, it dips down into the mid-teens. It's almost the exact opposite of venture.

In venture deals, if you're in the right thing, it almost doesn't matter what you paid. You see Cursor for details, see OpenAI for details, right? This is the exact opposite. This is fine, precise financial engineering. If you're wrong by 20% or 30% on price, your IRR dips from 20%, which is totally acceptable at scale, to the low teens, in which case you wish you hadn't done the deal.

Can I ask a question? Precise financial engineering for a 4- to 6-year hold period. 6 years ago, ChatGPT didn't exist. Are you able to do fine, precise financial engineering in a world where we move so fast?

I think a system of record is a moat, but I don't think it's a ticket to growth. This is, I think, super important, and it's something that everyone on X gets wrong. It's great to have a system of record, which Workday has. Even with AI and LLMs' help, it's very hard to churn, or you just don't want to churn, but it sure as hell doesn't mean I want to spend more money with that vendor. That's their challenge.

But it sure as hell means the next 5 years are far more predictable than 95% of companies, including poor Monday.com, which we love, or others. We have no idea where Monday.com or even HubSpot will be in 5 years at the SMB level. We know pretty much where Workday is going to be in 10 years, right?

And so I think this growth versus retention is misunderstood. There is a little bit of upside in this deal, which may—I don't know if it's part of Silver Lake's calculation. The CEO came back. One of the co-founders came back. Aneel Bhusri came back.

He hired his successor when times were easy. Just before AI, he brought in a great knobs-and-dials co-CEO, like our friend Daniel Dines at UiPath and others, and realized, “If I ain't going to work today...” He came back.

So I don't think Silver Lake is planning on Aneel radically changing it, but I think if he does, there's real upside to the deal. Maybe instead of their 20% IRR, it could be a game changer if he creates the agentic version of Workday. They at least have the founder back in the saddle doing it, and that would make me feel a lot better if I were Silver Lake, that I have upside.

But it wouldn't be in the damn base case. Jason, you framed the base case exactly correctly. It's 5.3 times trailing revenue. In other words, what this says is financial minds will pay 5 times revenues for a system of record growing at 13%. Anything that's not a system of record, anything that's not growing as fast, price accordingly.

There's no way you'd apply the same kind of leverage to, for example, a to-do, task-management, project-management, or website-building software. What this gives you is a sense of what the baseline is for best-in-class LBO takeouts.

The Airtable–Bending Spoons deal gives you an idea of what it is: if you don't have that kind of system of record, you get 2.7. If you're vaguely profitable and in a space where, as Jason says, you can predict 5 years, you get 2.7. And what Workday says is, if you've got 30% operating margins, modest growth, but you're a system of record where you really can believe in the next 5 years, then if you're lucky, you get 5.3 times revenues.

That's the bid-ask spread right now. Contrast that with the game for OpenWater, where they're going to get, I think, a trailing-revenue multiple of plus or minus 100. You know, they're going to get 70 times trailing revenues. Which game would you prefer to play?

Workday has something that makes it a better deal for PE, I think, than anybody else on the target list, which is that it is a somewhat closed system of record. Now, Salesforce is out there working their freaking tails off because they are a much more open platform.

You can build your own agents on top of Salesforce tomorrow, and a lot of the hot GTM startups are built on top of Salesforce. They're not necessarily only on Salesforce, but it's open. Try building on Workday. It ain't so easy, right?

It is like LinkedIn, right? It is intentionally barely open. So there are negatives to that, right? But it also means you're going to capture more budget overall in your ecosystem than you would for others.

It has more of a buffer against agentic damage to your growth than an open ecosystem has. Open has negatives today, so I would want a system of record, churn-impossible, closed AF. I want the most closed system that can't churn, because the reason systems of record aren't that great is that you need your system of record, but if you're remotely open and you can produce a better agent yourself or through a third party, the value will extract to the agent even if the system of record is retained.

But Workday is so closed they've got a leg up, right?

Jason, how open is Salesforce?

They are a tollkeeper, like Shopify, but they're pretty open. Shopify and Salesforce are pretty open. The 3 of us can use OAuth to ship a Salesforce app tomorrow.

Just to prove that, Harry, really quickly, look, there's a bunch of companies, even in the pre-LLM world, like Gong, Outreach, and Salesloft, that are all effectively built on top of the Salesforce platform. You can't name the equivalent with any ease in Workday. There are a few, but it's much harder. There's some of the planning tools, but pretty much most of it gets sucked into the gravitational pull that is the GL and the accounting system.

So I agree. That's a good point, Jason. Whatever dollars are in that ecosystem, if they're careful and shrewd, Workday will get most of them. On the other hand, if they get too greedy and they don't invest enough, then the customers start thinking, “Oh my God, this is just not advancing. Over 5 years, maybe I do need more of this agentic workflow on top.”

Maybe the smaller customers start evaluating NetSuite, start evaluating the next generation. Even at the very small end, you've got the Willis, you've got the Campfires, you've got people like that. You can't be such a greedy bastard in your ecosystem that you incentivize people to start trying to move out, right?

But Silver Lake are smart, and Aneel's smart. You could have this be a profitable, self-contained universe. But remember, the most exciting version of that is you pay down all the debt in 5 years and you double your money. That's as good as it gets now. It's on a lot of money.

You're probably putting in plus or minus a $20 billion to $30 billion equity check, because you're not going to get infinite debt. Maybe $20 billion—you might get $18 billion, $15 billion of debt—which means you need a $25 billion to $30 billion equity check. So you're going to turn $30 billion into $60 billion, which on a multiple basis is not amazing, but it means you've generated $30 billion in gains and 20% of that in carry. So someone's about to make $6 billion if they can pay down this debt and just knuckle down for the next 6 years. Go team.

Yeah. And Aneel gets to rebuild his company outside of the public-company eye, which is slightly overrated because he has to hit the underlying numbers, but it's much better.

Instead of large numbers of stupid comments, he will get 1 very focused comment from 1 of the world's smartest investors. It's probably a trade-up.

Just 1 last thing on this versus Salesforce. It's just interesting. We run Salesforce entirely headless, so we have our own agent, 10K, our own AI VP of revenue. It runs Salesforce under the hood.

The pro is it makes Salesforce much more powerful than it ever was. I didn't log into Salesforce for 7 years. Now I log in every day because I have an agent. The con is it can connect anything—the agent. It literally can connect to any other agent, including competitors, including other data sources, data lakes, data everything. The agent doesn't care.

It's really a weird world as a system of record or a core system. Do you want to be extensible and open? Salesforce has said you can be headless. It creates risks and opportunities, because you make it much easier to abstract you away or to compete with you, even while you may retain a few seats, right?

Logo retention may be high, but it makes you have to run faster. Workday doesn't have to run that fast. Everyone can't run it headless and integrate any single thing, or pull out all your employee data and push it into my own ATS, my own system, or my own financials. It's a shrewd deal because it's the best moat out there with a system of record.

And I go back to my comment. If it is a shrewd deal, it also, by definition, means it's the high-water mark of what deals are going to look like. Plan accordingly, people. You get 2.7 from Bending Spoons and you get 5.7 from the Silver Lake guys, and you pays your money, you takes your choice.

Lamkin, you have a buyout firm. Which other asset would you buy next?

I'd want to know who gave Jason money for buyout. I would give Jason money for venture, but I don't see him as the spreadsheet guy.

I don't. Harry, I just think more and more about the fact that systems of record are going to retain their customers, but I think we just underestimate that that's not enough to grow. It's grow or die today, right? It's grow or die. This whole show, everything: grow or die.

Who cares about the stock-based comp or anything at Anthropic? My God, it's OpenRouter at 192× revenue. Just because your customers are prisoners does not mean, in today's world, they will spend 1 more dollar with you. In fact, the CIOs want to cut what they spend when they're hostage, right? They're like, “Okay, I want to spend 80% of last year. What can we cut from our bill from the vendors we're stuck with?”

I've got to think, but yeah, Rory's right. I'm not the best at the spreadsheet.

You know, I'm actually going to cancel my comment and disagree with myself. Actually, I think you'd be great, because I'll tell you what you would bring to the table that I think a lot of these PE buyers missed. It's this idea of mission clarity around growth.

If you don't have growth of some sort, you're in a desperate race against the debt, and the best you can get is a mid-teens IRR if you work there and you buy cheap. And remember, that's when you buy at 5.7× revenues. Some of these PE deals were done 4 or 5 years ago at 10× or 12× revenues for not as good a quality asset as Workday today.

The PE firm should hire you as their operating partner, where, for every new deal they do, you explain the facts of life. It's really clear here, people. The only thing that matters: you can't just stick it to your customers. If you don't give them value, you're going to get shafted in the end.

Rory, for me, the death spiral here is the exec guy who's got no idea about AI and has a load of logos and a load of middle management, and I think Jason would be the freaking best.

I agree. I change my mind, because you're done if you don't.

I don't want to spend money on the positive side. If you look at it like a more SMB version, I'm not saying how widespread it is, but if you look on social media, a lot of folks are like, “Okay, I'm lifting off Airtable now.” And they're like, “Bending Spoons is going to raise my prices 3×,” right? “Let me start doing it now.”

It's just an extreme version of what you have to be careful with everywhere, right? I mean, Bending Spoons may lose 20% of Airtable's customers who finally spend a week lifting off of Airtable, but when they triple prices, it's a good deal for Bending Spoons. But it's going to happen a lot faster than Workday.

8. AI Apps Become Platforms

Growth at all costs on the consumer application side. 2 big fundraises: Higgsfield, which raised at a $5.5 billion valuation from DST and hit $700 million in ARR; and then you have Lovable, which raised a new round from Menlo. They're around the $600 million to $700 million ARR range too, raising at a $13.3 billion valuation. Big price divergence for very similar revenue numbers, which I find interesting.

Guys, we've talked about these companies a lot. How do we think about them?

On the Lovable thing, the thing I was thinking is, it's so crazy considering when we started the show, right? Lovable and Replit were both raising at around $2 billion and were really terrible products when we started the show. Now they're great products. They're truly generationally great.

I do think today engineers and developers will mock me for saying this, but I do think that they deserve to be in a somewhat similar conversation to Cursor in terms of stickiness, strength, and capabilities. They were not when we started the show. Did Menlo pay up a little bit as an existing investor who was already in at $4 billion? Maybe. But is that multiple that far off the Cursor multiple that we just saw? It's not radically off, is it?

Yeah. A lot of it is for the end of the year. It's probably a little pricier, but whatever.

But it's not as out of whack as it might have seemed with Cursor as a comp. These platforms are becoming very rich. They're very good now, right? Cursor's very good. They can do so much more than they could 6 months ago. There's so much more complexity.

I mean, Cursor launched Origin, right, which would beat Graphite or whatever. It's going to become a GitHub entire-workflow replacement in a couple of months, right? I'm closer to Lovable, but they both just launched automatic deep penetration testing as part of their products, right? So you can go really deep on security.

These aren't just little hacks from a year ago, and it also makes it harder for startups to beat them out. As Cursor, Lovable, and Replit become true platforms, they're great software today. So when the Cursor deal was announced, our jaws dropped.

Now it's a comp. It's just a comp, and I don't think this is such a bad comp for Lovable to Cursor. Maybe that sounds wacky, but that was the one I thought. Higgsfield's cheap, although when the deal was done, it was at $500 million. So it's funny: in today's world, by the time the deal was announced, it's at $700 million. So it's still cheap, but that's what happens if you don't announce a deal the hour the term sheet is inked, right?

You guys know what both of them have done really well is parlay that kind of massive bottom-end demand for AI—in Lovable's case, for website building and coding; in Higgsfield's case, for video. Start with a PLG motion and then add mid-market and enterprise products on top. It's a well-trodden path. It was well-trodden in the SaaS days. We did a bunch of that. It all works. You build your top of funnel, and then over time, you just add the enterprise features.

But they've both done it really well, and you're right. Lovable has punched its way into being a big-picture coding alternative. There are different ways of going at it: you've got the Cognition style, you've got the Cursor style, and you've got the Lovable-Replit style. They're not direct comparables, but the big-picture point is that the thing AI does best is write code, and Lovable is a tool that uses that to write a lot of code, so it's got a lot of lift. I think they've both built good enterprise businesses on top of good consumer businesses.

It's not as—Higgsfield, and I know you guys are in them, so you know much better than me—I think there is clearly a market for enterprise video. It's a good market. It's perhaps not as deep as the coding market, but great to see them do it.

The one meta-learning for me is—I do think, and it took me a little while to see this—that these products today, not forever, maybe only for 6 months, or who knows, are defensible and have moats. For example, Higgsfield: I was one of the first 10 customers, I think. What could you do? Make a 4-second video using Kimi or Qwen. Who cared? It was a great way to do it because I didn't even know how to use a Chinese model, okay? But that wasn't particularly defensible. Now you can make a full-length motion picture.

And now with Lovable and Replit—even though I've been a user for a year, you could have made fun of these products when Harry invested, when we started the show—they really can almost build production-grade, highly secure apps with everything across the stack. I know we're building so quickly, and you better be into your 2028–2029 roadmap or you're failing, but they are starting to get these layers of moats.

The folks who work at these companies are so smart, right? Higgsfield has the smartest mathematicians in Kazakhstan. Lovable and Replit have become talent magnets. I know the team at Replit better. You walk into Replit—I mean, these are the smartest people that Amjad could recruit for years. These layers are not impenetrable, but they start to get thick and crusty, this crust around them, right?

Staying with that, because I think you're exactly right on that moat comment. There was a whole bunch of, "Oh, what's the moat?" I think the truth is, in any new software market, out of the gate, moats are light. But the companies that execute and get traction accrete moat over time.

Just to give 2 historical examples, the Netscape browser wasn't that hard early on, but as you parlayed that into other things, ultimately you only got acquired for $10 billion, which at the time felt like a failure, oddly enough. But the initial thing was relatively simple. It got complex. The classic example is Microsoft. The MS-DOS product was mind-blowingly simple, but over time you just accrete more and more value, and the same thing is going to happen here.

Will there be some guys who stumble along the way? Of course there will. But you're right, Rory. If 2 years ago it was probable that someone could have built a Lovable competitor with the features that it had relatively quickly, as they add more and more features, that just gets harder and harder.

Yeah, and I don't know that it was obvious 6 months ago that this would happen. I don't even know it was obvious to these companies we're talking about that it was obvious, right?

Yeah. I think, going back to Mike, it just shows: push forward, add more stuff for your customers, revenues grow, and good things happen. If you're faster than everybody else and better.

Yeah, absolutely.

You just have to be faster and better. That's all. Just faster and better, then it will accrete.

Faster and better is a more tangible thing than thinking some kind of... There are businesses that are much more moat-centric, right? You know, they have massively high IP. Some of the model companies, obviously, to some extent, and definitely things like the bioinformatics companies. But there are also businesses that will become wonderful businesses where the moat is, as you say, Harry, faster and better. And you just have to know which game you're playing.

Speaking of high-IP businesses, literally, like, 3 weeks ago, Etched raised at $17 billion. Today they've announced they've raised $700 million at $21 billion from Jane Street, Kleiner, Sequoia, and Andreessen. Four weeks after, double the price.

It was a good month, Harry. You only need 1 great month to raise today. I don't know whether you're pre-seed or north of 30. You used to need 3 to 4 good months to raise. Now you just raise on the 1.

Listen, I don't know the details of the deal. Jane Street wants to be a customer too, or something, right? It's not that they're suspect; it's just you never quite know how it's all tied together. So that was the only asterisk I had in the deal. But I don't know the details.

What story have I missed that we should discuss?

9. Andreessen Faces Antitrust Scrutiny

There's one that—I don't know if you missed it while you were vacationing there, Harry—but the Department of Justice is picking on poor Andreessen because of these overlapping boards.

I put it in. It's in my schedule. Thank you for mentioning the vacation.

But why is it? What's the story behind the story? There must be a reason, right?

I did the story because I often think one of the jobs we try to do here is let everyone who listens know what's interesting this week, right? And I will admit, this time yesterday, I knew exactly jack. I was like, "Huh? What gives?" I looked at this, and I'm like, "Why is..." Let me say something cynical and then retract it. Why is the Trump administration picking on Andreessen Horowitz? One would have thought that there is honor among thieves and gratitude. You know, the definition of an honest politician is: when he gets bought, he stays bought.

Yeah, Elon got his deal done in weeks, his $60 billion deal done.

So I did the research, and it turns out that this initiative, no surprise, was actually kicked off in the prior administration, which, frankly, was much more willing to, quote-unquote, get involved in business and kind of try and tell them what to do.

What's happening here is—zoom out—there's something called the Clayton Act, which I think was an antitrust act from the early 1900s. Section 8 basically says individuals can't sit on 2 boards of companies that are competing. There are all sorts of definitions of how you define competing. There's a de minimis threshold. That's on the statute books.

It turns out that under the Biden administration, the DOJ—I think it's the FTC within the DOJ, but don't quote me—had actually initiated some actions on that. There were a couple of general business folks who were on overlapping boards. Interestingly, Thoma Bravo had a couple of companies where, in one case, they had spun off a separate company from an existing company, so they had a lot of overlapping boards. The DOJ got on them, and eventually they said, "We'll just take the board members off." Right? And it was fine. So this is a thing, right? This is apparent.

It's a low-consequence thing, because what invariably happens is, if the department just comes around, you just pick the less interesting board and you come off it. What sounds like happened here is, even though you'd have thought the administration change would have killed this, apparently as part of it—remember when Fivetran and dbt were merging? The DOJ had to look at that because there were antitrust issues, and that got through and it was passed.

But as part of that, the light went on in someone's head at the Department of Justice: "Hmm, do we have a Section 8 Clayton Act violation here?" Because Andreessen's on the board of Databricks, and they're also on the board of, I think it's Fivetran, right? And now they're competitors. So now this has been percolating, and now they're investigating, right?

This is one of those things where I know why the law originally exists. It's all back to J.P. Morgan and overlapping boards and antitrust and whatever. You look at this and go, "Really? Is this the biggest fish you have to fry?" But my guess is it peters out into some version of the venture firms saying, "We'll take the board member off Fivetran," or whatever. Now, there are—it's interesting—there are ways you could contest it.

If you gave a shit and wanted to litigate, there are all sorts of things you could do. The legislation says individuals can't be on 2 boards, but it's not as clear on whether 2 separate individuals can be on 2 separate boards.

There are a whole bunch of reasons why you could decide, if you had the stomach for it, to litigate and see whether the Department of Justice would back off. But the truth is, no one's going to bother. I shouldn't say no one's going to bother.

It feels to me like, if this thing rumbles on and the Department of Justice doesn't back off, or they don't decide the competition issue is de minimis, at some point—if it got really serious, that's my point—no one's going to get into trouble for this. If it gets really serious, they'll go, "Okay, we'll take a board member off."

Yeah, it's probably a non-story in the end, thinking through it more, right?

Yeah. Pain in the ass, though.

There is a remedy here. You resign, right?

Yeah.

It's not damages.

If you're the compliance officer at Andreessen, you're wasting a lot of your time on this. But you're right, a non-story is it.

The only weird niche thing is that, classically, you might ask the founders if they're okay with it at each company. Maybe that's not even a permissible out under the Clayton Act. I don't know, right?

It's not, actually. The funny thing is, to your point, you're exactly right. This is a very interesting example, because it's an example of you and I both knowing that that's the acid test.

What we would be worried about is whether Founder A is pissed about Founder B. Are we sharing information? Are we damaging the other company with this information? But the classic antitrust thing is all about consumer damage.

What they're hypothesizing, absurdly, is that the Databricks guy and the Fivetran guy get together and say, "Why don't we raise the price of AI tools, and we'll stick it to all our consumers like J.P. Morgan and BofA and whatever," right? And that's so far from what's happening.

And this is the problem: You pass a law for one reason, to prevent U.S. Steel from raising prices in 1909, and here we are in 2025 or 2026. Do we really think that Databricks and Fivetran or dbt are colluding about the price of data tools? You're right. The logical test would be, "Founder A, are you cool with this for Founder B?" And if they're cool, we're cool. But it turns out that's not the way we write laws.

Yeah, I mean, if Martin Casado had to step off the board of Fivetran after exiting Cursor at $60 billion, it's probably okay, given their position in Databricks. It's all right. That guy just got us a $60 billion exit. We're sitting on $200 billion at Databricks. Fivetran, good luck. We'll switch to an observer seat.

Yeah. Anytime you want.

I'm going to move to observer status. Like, that doesn't work either.

Doesn't work either.

You got to retire.

They've actually thought of that. They're literally like, "Mr. J.P. Morgan can't bring his big banker nose in 1909 into any of the meetings." Yeah, no, that's what we're trying to solve.

So be it. Just CC me on the updates.

Genuine comment here. It's why—and this is a serious comment—it's why, when we talk about all these regulatory regimes for other things, you have to remind yourself these regimes go on forever, right? And if we pass some law about AI regulation now, you've got to be really careful about the unintended consequences months, years, and decades later. Once the regulatory law is passed, it doesn't leave.

You know what? One last thing, in all seriousness. At Andreessen's scale, everyone has to take the Series 62. There's probably 40 or 50 legal things going on in the background at any given time, right? It's probably not even 4. It's probably like 40.

They don't even talk about this one, right? It's mentioned: Let me know if I have to do anything. I've got to go to that pasta lunch with Michael for the closing lunch.

The $12. $12.

$12. Yeah. Let me know if there's an issue, because there's like 50 other lawsuits. Everyone's coming after Andreessen, right?

Yeah. It's the old "no conflict, no interest" comment. Exactly. They have lots of interests, so they have lots of conflicts. It'll be fine.

Boys, thank you so much. That was awesome. I loved that. But before we leave you today, you have the idea, but with most AI tools you hit a wall. The setup, the config, the gap between what you pictured and what you actually ship. Well, Base44 is where that wall disappears. You describe it, yeah, Base44 builds it. Apps, websites, AI agents, real working products built in minutes using nothing but plain language. And it's all batteries included. The backend, the database, the authentication, the hosting. The heavy lifting is handled so you just really stay in the flow. This doesn't just take the busy work off your plate, but it gives you an advantage and pushes you past what you thought you could build alone. So in this market, fast is the baseline. To win, you just have to be first. Base44 is that edge, the move that skips the troubleshooting and gets you straight to the breakthrough. Build your next thing at base44.com. That's base44.com. While Base44 turns ideas into apps, Plaud turns conversations into insights. Founders and operators spend way too much time every week jumping between meetings, investor calls, brainstorms, customer conversations, and then trying to piece everything back together afterwards. And that's why I've been using Plaud. Plaud instantly captures conversations, voice notes, meetings, random ideas with one press, and then turns them into clean summaries, action items, mind maps, and searchable notes that you can actually come back to later. Honestly, it feels less like a recorder and more like an AI-powered like brain memory system. And the crazy part is the hardware itself. The Plaud Note Pro is literally as small and thin as a credit card, so it's just always with you when something important comes up. There are already more than two million founders, operators, investors, consultants, and professionals using Plaud to stay organized. So think more clearly and stop losing great ideas and important details. Go to plaud.ai/20vc and use the code 20VC for 10% off. That's P-L-A-U-D.A-I/20VC and use the code 20VC for 10% off. While Plaud captures the conversation, Fin helps continue it. As AI agents become more common in customer experience, teams often end up juggling multiple siloed tools for every job. Well, Fin was built to change that. It's a single unified agent that works across your entire customer experience, from service to sales to success and beyond. Fin is the agent making perfect customer experiences possible for thousands of customers. It's powered by custom models, trained on years of real customer interactions, so it understands the nuance and complexity of customer service better than any other agent. That means faster resolutions, more consistent support, and just better experiences for every customer. It's also designed to be fully self-manageable, so you can easily improve and adapt it as your business evolves. No third parties required. Leading companies like Gamma, Asana, DoorDash, and Crypto.com already use and love Fin to deliver better customer experiences. So for a limited time, you can get $500 a month in Fin credits for your first three months. Learn more at fin.ai/20vc.