# 20VC: Will Jensen Return to NVIDIA | Anthropic Set for Largest IPO of All Time | Kalshi, Mercor & Fractile: Mega-Rounds Continue

20VC · 2026-11-19 · 73 min · https://thetwentyminutevc.libsyn.com/20vc-will-jensen-return-to-nvidia-anthropic-set-for-largest-ipo-of-all-time-kalshi-mercor-fractile-mega-rounds-continue

## Transcript

Speaker 0
If I own 100% of the stock of this company, the person on the planet I would most want to run it is Jensen.

Speaker 1
But the company’s been maimed. The whole ecosystem’s been maimed.

Speaker 0
The winner here, of course, is Anthropic.

Speaker 1
I think the liquidation preference has been weakened by the NVIDIA drama.

Speaker 0
You buy SpaceX stock entirely on the basis of what you think Anthropic will be paying you one year from now for that compute.

Speaker 1
LPs are gonna grade us all on how many $20 billion-and-up portfolio companies we have.

Harry Stebbings
This is 20VC with me, Harry Stebbings. We're joined in the hot seat today by Rory O'Driscoll, Jason Lemkin, breaking down the biggest news in tech. Wow, what a week of news we have. The biggest IPO ever. Dear Lord Dario coming out with Anthropic. $1.4 trillion, 7X oversubscribed. We have Jensen Huang. Will he come back? Will the board give in? We shall see. Microsoft outage. Outlook, Teams, Copilot down. Their users probably smiling. I'm gonna get in trouble for that one. And then SpaceX catches Starship. Incredible. What an innovation. Elon Musk stuns the world again. This and so much more in the show this week. But before we dive into the show today, Base44 is one of those products I used, and I just thought straight away, "Man, I wish I'd had this when I started 20VC with $50 and no contacts in venture." I wanted to see what it could do, so I built a guest research hub for the show. Upcoming guests, company notes, previous interviews, sponsor contacts, follow-ups, all in one amazing location. I described it in plain English. A few minutes later, I had the product open in front of me. Database, login, back-end and hosting already handled, and just all in one. It made it so easy. And that's what I loved. I didn't write a long spec. I didn't stitch five different tools together. I built the thing, and I started using it. For any founder listening, that's the point. The faster you turn an idea into something real, the faster you launch. Apps, dashboards, internal systems, AI agents, whatever you need to build, Base44 gets you there. Go to base44.com/20VC. That's base44.com/20VC. While Base44 helps you build, AlphaSense helps you find answers really, really fast. Quick one for anyone doing diligence in venture right now. AlphaSense just launched SuperAnalyst, and it's one of the more genuinely useful AI products that I've seen come out of this space. It's an always-on agent that works like a power user inside AlphaSense, but on your behalf. So check this out. It pulls from over 500 million premium sources, covering both public and private companies. That's filings, earning calls, broker research, expert transcripts. And it doesn't stop when you close your laptop. Point it at a sector, a company, a thesis you're tracking, and it keeps working in the background. And here's the part that I really care about. It remembers your process. It learns your templates, your firm's logic, and it applies that to every new task. So it compounds. It gets better the more you use it. Isn't that the value? And every single output links back to its source. So there's no black box. There's no trust me. And you can see exactly where something came from, so you have complete traceability. Pick the deal memo or the market map you've been putting off and hand it over to SuperAnalyst instead. Start your free trial at alpha-sense.com/20VC. That's alpha-sense.com/20VC. While AlphaSense finds insights, Intercom helps you serve customers. 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.

Speaker 0
You have now arrived at your destination.

Harry Stebbings
Boys, we have more on this week than I think almost any other week. We’ve got what I believe is the largest IPO in history, and then we also have one of the greatest founders potentially coming back. The news just doesn’t stop. Jensen Huang falls from NVIDIA, and the stock loses up to $601 billion in a morning. Then the biggest investors, Gavin Baker and Altimeter, come out with letters, and 5,000 employees and 200,000 people sign the petition. Let’s just start with this: Do we think it’s right that he was let go?

Speaker 0
What do you mean by “right”?

Harry Stebbings
If you were on the board, would you have let him go?

Speaker 0
If I were on the board with this guy who’d founded and built a company for 30 years, one of the things I’ve learned is that decisions made in haste are often wrong. I’d find some way to say, “Take a leave of absence. We need to figure this out. Get to the bottom of it. Slow the whole thing down so you can make a better decision.” When I look at the dumb decisions I’ve made—and I’ve made plenty—usually it’s because you’re reactive and quick, and you don’t have time to think and let it settle in. In the same situation as a board member, I’d probably have tried to find a way to make a less irrevocable decision and get more time to think, right? So I actually think, “Hey, take a leave of absence, interim CEO,” whatever, and let’s figure out what’s going on.

And then just two other comments. This is a messy thread because you have the personal allegations, both with a customer and with employees, and separately you have this emerging dynamic of compliance with export controls, and all the dynamics around that getting intermeshed with the administration. One of the questions I’ve had here is what the board knew at any particular point in time. Was it just as simple as, “Oh, I can’t believe there’s a second affair. I’m done. I’ve lost trust”? Or were these other dynamics coming into play? I just don’t know. But again, I go back to the idea that more time to figure out what’s going on would’ve been useful.

Harry Stebbings
Can you just temper my hopes and expectations? Do we think he will come back?

Speaker 0
As of this morning, he just made that statement which basically says, “If you want me back, I’d love to come back.” It proves one of the old rules: never resign in anger or even in shame. Take your time. So he’s now sitting there going, “Hmm, maybe I shouldn’t have quit.” But yeah, that’s the statement now. You asked a separate question: Will he come back?

I want to say something about the board. Four or five of these board members have been on for 30 years. They backed Jensen 30 years ago when he was in his early 30s. There are two VCs, an audit chair, and one other person. It’s probably been the single most significant business relationship of all those board members’ lives. I’ve been on boards for 10 years. I’ve never been on a board for 30 years, and I’ve never been on the board of a company for 30 years that went from a startup to the most valuable company on the planet. But I know how much I like my CEOs that I’ve backed for 10 years, and how hard it would be for me to vote to replace them. I can only imagine how hard that must have been.

Speaker 1
Mark Stevens made $4 billion off NVIDIA. You don’t want to fire the guy, and he just sold a billion in cash this year.

Speaker 0
You reminded me that while they technically did not fire him, they also did. Again, you get a lot of leaks here, so I want to acknowledge one thing. When four or five people who are widely experienced and have made a ton of money decide to go against someone who, as I say, has been the most important successful business relationship of their lives, you really think, “There must be something that’s really getting on their goat.” Given that, it’s hard to imagine this board saying a week later, “You know, upon sober reflection, we made a horrible decision. We’re gonna just vote you back in.” That seems like a big lift.

Technically, you could have a shareholder vote to replace the board, and then the new board could do it. So it’s technically possible. In the absence of a board change, my gut is no. The thing you don’t know, and I don’t know, separate from all the relationship questions, is the Department of Commerce and Department of Justice comments on NVIDIA’s actions in terms of selling chips to Indonesia that may well end up in Singapore and then in China. The Department of Justice and the Department of Commerce issued a statement on this just this week, which is interesting. So there’s that pending issue, which obviously wasn’t the crystallizing issue for the resignation—the relationships were—but it’s kind of out there. It’s hard to imagine a board saying, “Oh, let’s come back and fight this one.” My gut would be no, in the absence of a board change. I could be wrong, but I could look like an idiot in two days.

Speaker 1
I’ll tell you why I think there’s a high chance he comes back.

Speaker 0
By this week?

Speaker 1
Maybe, maybe. And this week is fast, right? He shouldn’t come back this week.

Speaker 1
It should take a couple of weeks to happen, but he may. I’ll tell you why. We’re all getting facts, and all these leaks really bother me. This is the leakiest ship I’ve ever observed. It’s horrific. It’s disgusting.

I don’t think there’s going to be a huge issue around the spy who loved him. I think that’s fading. Maybe not completely, because the facts are actually getting worse, but I think it’s going to fade away. I think we’re going to say he didn’t know she was a spy. Let’s move on.

And I think this issue—because it’s already happened, it’s already been banned—of allowing data centers adjacent to China to use top-of-the-line chips, then ship them and transship them, breaking open the servers and driving the chips off the border, has already become a massive issue this week. Massive issue. This will reverberate for a long, long time, and it may be 20% of NVIDIA’s total revenue if you count China.

The problem is, if Jensen’s guilty, so is the CFO who signed the financial statements, so is the head of supply and engagement, and so is the CTO. I think you’ve got to bring him back, because otherwise you’ve got to fire all of them. They’re all guilty. They all set up these data centers.

Speaker 0
And it’s funny you should say that, because the CFO actually being the CEO now is interesting. She’s going to be on the hook. Is she going to reaffirm the guidance she gave a month ago? Is she going to say, “All the accounting that I booked as CFO is still good now that I’m the CEO”? It’s very hard to walk it back and say, “But it was me, and I’ve changed my mind.” Jensen may… So, yes, there is complexity there.

Speaker 1
How can she say it wasn’t me when she signs the financials? And she—

Speaker 0
No, I agree.

Speaker 1
—and it’s more her job to know where every chip goes than his. He’s the guy out eating chicken and schmoozing. Her job is to know where every GPU is. No, seriously, she can’t say it wasn’t me.

Speaker 0
Well, it’s interesting, because the board has now decided it’s their job, too. They formed this board committee on the importation and exportation of chips. Basically, some board-level risk supervision. That tells me they think that’s the big tail risk: that something happened deliberately or accidentally, and there was some kind of breach of whatever ambiguous litigation we have around these issues. So that’s clearly been the dominant issue now.

And just one other comment here: Harry, you asked the question, “Will he come back?” There are 2 questions: Will he come back? Should he come back? Let me make this clear: If I own 100% of the stock of this company, in the absence of the US government trying to put my CEO in prison, the person on the planet I would most want to run this company is Jensen. I’d want to make everything else go away. The question is, can you do that amid the morass of being public?

Harry Stebbings
If he were to come back, what price bump would the stock get?

Speaker 0
It wouldn’t go back to where it was.

Harry Stebbings
Really?

Speaker 0
No.

Harry Stebbings
Oh, I think so.

Speaker 1
But the company’s been maimed. The whole ecosystem’s been maimed. It will never be the same company.

Speaker 0
You’ve got 2 things. The first question would be: What percentage of your revenue really was coming in directly from China? It’s worth pointing out that 10 years ago, back before it mattered, I think China was 15%, plus or minus, of the business. So a good slug of the business goes away. If 15% of your business goes away, do you go down by 15%? Maybe. I don’t know.

The second thing is—and this is a weird phenomenon—you see it. He’s been the master banker of the whole AI boom, and everyone’s been saying, “Yeah, yeah, that’s great. That’s amazing.” Then, the way humans are, people are awful. The minute he’s out of the picture, everyone is dumping on those deals and saying, “Those deals are stupid and they’re all going to blow up.” The question is, if he comes back, can you go back to doing those deals? I don’t think so. I think you wouldn’t be able to use capital as aggressively as he has, and that would have some impact at the margin on revenue.

Harry Stebbings
Boys, is it not fundamentally so bad for the venture ecosystem, not having Jensen and losing his aggression in direct deals and…?

Speaker 0

Yes, it sucks.

Harry Stebbings

Yeah.

Speaker 0

It sucks. Start with the very basic comment: The winner here, of course, is Anthropic. The biggest single thing Jensen was pushing—the thing that was most significant for the ecosystem, I think—was an open-weight US alternative. His big agenda was with his post-ad acquisition and Hugging Face. It takes real money to build these models. You look at the statement in the S-1: It took, I think, $10 billion to train the last Anthropic model. No one else has the dollars to build an open-source product.

So if he isn’t able to come back and that agenda doesn’t get pursued, you don’t end up with a significant US open-source, open-weight product. A second comment: Given the whole China dynamic, you probably end up with more pushback on using Chinese open-weight models, as Jason pointed out 2 weeks ago. The winner is the closed-weight provider in the US, which now has pricing power.

Apparently, there have been leaks and rumors of senior Anthropic people smiling happily about this, because it’s just good for them. They’re not big NVIDIA customers, and OpenAI is, so their competitor is weakened, and the provider of the cheap alternative product has been weakened as well. This is a good day.

Speaker 1

There are a lot of conspiracy theories about who did the leaking. But to Rory’s point, I think what NVIDIA did was perfectly legal. It was okay to set up data centers outside of China that were used essentially remotely by Alibaba and friends. It was completely legal, and as of now it’s not; apparently, it’s no longer legal.

When you’re Anthropic and you’re feeling distilled and ripped off, that seems like the crappiest loophole on planet Earth. How can they get away with this? You’re just stealing our IP, stealing our hard work by building this data center a couple hundred miles away from our biggest competitors. It’s epic. I don’t know that Dario is gloating, but, man, it’s a boost.

Harry Stebbings

Speaking of a good day for Anthropic, it’s about to be their big day, the special day we’ve all been waiting for. Again, we’re slightly ahead of what will be the news item of the day, so to speak. The IPO is 7 times oversubscribed, with some reports saying up to 10 times oversubscribed, at a price of up to $1.4 to $1.5 trillion. The largest IPO ever. How’s this going to go?

Speaker 1

Can I bring up my one big takeaway from it? For what it’s worth, it has nothing to do with the $1.5 trillion we’ve talked about. The oversubscribed book is mildly interesting. Rory can give us his thoughts. I know we knew this, but I didn’t realize it until I wrote it up this morning.

At $100 billion in revenue, which is where Anthropic will be by Q1 or whatever, at run rate, Google had 81,000 employees. At $100 billion in revenue, Meta had 71,000. At $100 billion in revenue, Anthropic will have fewer than 6,000. We’ve talked about this on the show, and many folks have said, “Oh, it’s no big deal. It’s inference and AI costs, and compute is $10 billion instead of humans. That’s why. This is not a SaaS model. The margins are different, blah, blah, blah.”

All true, but let’s think about the human implications of that. The next $100 billion company got there faster than ever. Yes, it’s potentially the greatest venture return of all time. But it got there with one-twelfth the employees of Google and Meta. One-twelfth. Not 30% less. The average go-to-market team, according to ICONIQ data, is 30% smaller with AI. All the Harveys and Schmarveys and Tarveys and Sierras are about 30% smaller. Some are a bit smaller. But one-twelfth the size—that says to me jobs are in even worse shape than we’ve been discussing.

One-twelfth. They got to $100 billion with one-twelfth the employees. Who cares at the end of the day, as long as it’s fundable, whether it’s GPUs, TPUs, or EPUs? That is mind-bogglingly different for tech and our economy: getting there with one-twelfth the humans.

I don’t mean to be one of the people who are anti-AI, but it’s worth reflecting on this, in the moment when the average employee is making $41 million and Ashton Kutcher is making more than—good God—$600 million. I want to quit venture. But this employee thing, to me, for a moment, is the most important thing in all of this, because we’ve already used the products.

Speaker 0

Yes, this is a much more compute-leveraged, human-non-leveraged play. You’re exactly right. That’s why the humans who are there make more money: There are fewer of them, so you can share roughly the same pot among a smaller number of people.

Speaker 1

Yeah, it turned out not to be that crazy what they’re paying, right? $400,000 for an office manager. It’s okay.

Speaker 0

You want a small number of people. They have to be really good. I’m not sure it’s as disastrous for labor as you think. It’s not like every company is going to be 90% more efficient. I think these are the guys who can leverage compute the best.

I don’t want to get lost in the weeds here, but I don’t think most old companies, like an old-school consulting company, are going to end up with $5 or $7 in inference for every dollar in labor. We’ve discussed this a lot. That ratio will vary between old-school companies, software companies, and companies like Anthropic, and they’ll be on the leading edge of it.

Speaker 0
And so I think the companies that are net additive will be compute-intensive. I don't think it'll be as grim for jobs as you think, Jason, but nonetheless, it is wildly noteworthy.

Harry Stebbings
Did you see the rumored 40% layoffs at ServiceNow today?

Speaker 0
I couldn't figure out if it's a rumor. I just thought, is it real?

Speaker 1
I think it's true. I think it's a big deal. I think it's more telling than you think, just to challenge you a little bit. Even segments that appear to be very human-centric are going through a lot of changes. Front offices and back offices are being automated. Food delivery is being automated. I mean, 80,000 to 6,000—I think this is the biggest disruption of our lifetimes, beyond AI itself, right? Which is bigger, but my God.

Harry Stebbings
Guys, does the Anthropic IPO change venture and our world in significant ways? For me, yeah. I see LPs much more willing to redeploy the cash they're going to get from it. They're feeling flush in some cases. I see a lot of Anthropic early employees becoming angels in U.K.-based companies. Any other ways where you're like, “This has a meaningful knock-on effect”?

Speaker 0
Well, of course it will. Depending on where it ranks versus SpaceX, it's the biggest or second-biggest IPO by dollars raised. More importantly, it happened in 5 years, not 20. Founders Fund invested in SpaceX in ’07 or ’08. It was founded in ’02, so a good 26-plus years ago. This was founded 5 years ago and got to a trillion dollars. It just massively raises the bar of what is even vaguely possible.

The people who did it early are going to make out like bandits. All credit to Spark in particular, and then Menlo in the following round. As Venky said in that great post of his, they’re going to be playing with the house money for a long time to come. Congratulations to them. Well done: great decision, well executed, well written. They’re going to make out like bandits, and the rest of us are going to go, “Wow, I’m working really hard for 1/10 or 1/20 of what you’re going to make.” That’s just the nature of an extreme power law of business. It can get in your head, because you can work really hard and think, “Wow, by anyone else’s standards, we’re doing great.” But that’s the single biggest win in history, and one of the fastest. So, yeah, it’s going to raise the bar of what’s possible. You can see it even in the market right now. I think what it does is increase LP willingness to deploy and GP willingness to take bets. Everyone’s like, “That bet worked. We should do more of them.”

Speaker 1
I was just looking over Carta’s Q3 data over the weekend. Since Peter went to OpenRoad, it doesn’t get as much attention as it used to. Peter, right? I didn’t even see anyone talk about it, but I looked at it over the weekend. To answer Harry’s question, in Q3 2026, the most recent data they have, deals were down 21%, while dollars were up 2.5x.

So deals are going to keep going down or plateau, and the dollars will keep going up. That is a radically different world from when we met, and maybe even radically different from when this podcast started. Everything else is a derivative of what you talked about. It’s a derivative of tripling and quadrupling down on winners. It’s a derivative of AI spend. Fewer rounds, 2.5x the cash—that’s the world today, and that’s the risk GPs are being paid to take and are happy to take.

Harry Stebbings
Weird new world of venture.

Speaker 1
It’s so different, right? I know you’re going to be graded. The LPs are going to grade us all on how many $20 billion-and-up portfolio companies we have. That’s how they’re going to look at our portfolios: how many $20 billion and up. When I started, it was unicorns. I don’t know what it was when Rory started, but if you had one unicorn in your portfolio, LPs would seek you out. Not the best ones, but folks would come out of the woodwork: “I’m doing a unicorn analysis on a Google search. I found you have 2. Are you raising?”

Now it’s going to be $20 billion and up. How many $20 billion-and-up startups do you have, ideally ones you invested in before a $20 billion valuation? That would help, right? I think that’s how LPs are going to grade GPs: how many $20 billion and up. It’s a new world.

Speaker 0
The funds that got in early on Anthropic are going to do amazing. They’re probably going to have 8x or 9x funds on big dollars. The amazing thing is the median fund in 1996 had a 6x DPI, because they were all just doing seed and Series A, and then the companies went public. It wasn’t even a unicorn. You put something in at a $20 million pre-money valuation, it went public at $500 million, you had a 20x, and life was great.

The odd thing is the multiple isn’t that different from the last 2 bubbles, but obviously the dollars involved are huge. If there are only $10 billion or $20 billion outcomes, then the earlier you go—Series B, Series A, seed—the probability of landing on one of those 10 or 20 outcomes logically has to go down, especially with lots of people there, right? It’s why you could argue that having early and growth funds is a very good structure for this environment. It’s noteworthy that no one got into Anthropic from the venture side before $5 billion. No one meaningful got in before $5 billion, when Spark did it, still pre-revenue, from the growth fund.

If you’re going to be graded on getting $20 billion outcomes, and Andreessen have been very clear on this, then you’ve got to do whatever it takes to get those $20 billion outcomes. If you can get some of them at the Series A, that’s fricking amazing. But if the only way you get some of them is at the Series C, that’s fine too, especially if the non-$20 billion outcomes can’t get liquidity in the same fashion.

Harry Stebbings  
I think one shift we’re seeing, with people playing with the house money, is all of the mega funds raised in Q1 ’26. I thought, “Great, we’ll raise in ’27. We’ll be out of their cycle. Fine.” But every single fund is queuing up out the door for January 1st. We have to get the bag. Every mega platform is raising in Q1 for what will be the biggest funds they raise. It’s a 12-month deployment cycle before they raise from last time. How do we think about this?

Speaker 0  
Generally, when you raise money quickly, you tend not to make as much money. This is that stage in the cycle. It’s worth remembering that the cleverest checks in Anthropic were written in early ’23 and early ’24, when the world was very different. If you go back to where you were then, people were much more scared. In general, I think now people are going to put money out with high velocity. It’s not clear to me you’ll end up with anything like the same return, because I think that’s just always the case. But, yes, that’s what’s going to happen, because people are going to people.

Speaker 1  
I think deploying it that quickly also inherently fuels the rapid markups we’re seeing. The best AI companies grow so quickly. Does it justify going from $10 billion to $20 billion to $40 billion in 6 months? Sure, on a spreadsheet, of course it does. But maybe the pace of deployment also self-justifies it, and if it does, to Rory’s point, it will compress returns. Maybe that’s fine. Maybe 8x compressing to 4x or 3x is fine, but it seems to be potentially partially fueling that. It can’t all be the revenue growth since last month.

Harry Stebbings  
So we have investment committee time, Jason, in the show-me-the-money vein. We have 3 deals on the table today. We have Kalshi, the prediction marketplace that’s taken the U.S. market by storm, at a $65 billion price. Not cheap. We have Mercor raising at $20 billion, one of the leaders in the data space. And then we have Fractile at $720 million, at a $7.5 billion price, producing inference chips that compete against the likes of Groq, Cerebras, and others.

Speaker 1  
Yeah.

Harry Stebbings  
Where are we going?

Speaker 1  
Well, look, I’m disappointed at what you’ve brought to IC this week. We’re going to do a deal because we’re going back to market in January. We have no choice. We’re going to do a deal. But last week I loved. Last week we had Zipline, a little expensive, but Rory and I love Zipline. Who else did we have last week? We had fal, right? You can’t argue with its success. Who was the third one we had last week? I think you picked the third one, Harry. It was a great group of 3.

I know we’re only able to deploy a few hundred million into them—

Harry Stebbings  
Revolut.

Speaker 1  
—it was a good one. Yeah, Revolut. Come on. Revolut is just a question of scale, the outcome. That was a great one. That was a fun one.

This week, I don’t like this one. Kalshi, I don’t like the regulatory risk at $65 billion. I think this is hubristic. We may make money, but I’m still uncomfortable with the lawsuits and the other issues. We’ll see. I’d rather stay low. Fractile: no revenue, the 11,000th inference chip startup. When is it going to end, boys? When is it going to end, at $7 billion? Who the hell knows what this thing is?

And Mercor, listen, I like Brandon. He’s an old soul at 20, and this is a great space because Anthropic and everybody else don’t want to do this work, but they’re still doing the dirty work. I just can’t predict the future of dirty work. Mercor is one of those ones where I’m jealous of the guys who did the last round. I’m jealous of the guys who did it at $10 billion, $5 billion, or $2 billion. I’m going to pick it because we’ve got to put money to work. It’s November 17th, but I like last week a lot better.

Harry Stebbings

Can I explain to you why you’re the associate and I’m the GP at this firm?

Speaker 1

Well, I think that’s fair enough.

Harry Stebbings

Because Kalshi is the unwavering market leader in what is a monstrous sports betting market in the US. You have states that are going to get so addicted to this revenue that they won’t allow it to be banned by a new administration. So that’s a pretty good place to be. We can see a 3x here.

Mercor: it’s not impossible to see a $10 billion spend from the LLM providers on the data Mercor sells. If you believe everyone is going to own their own intelligence, they’re going to buy more data for it. That’s an even bigger market. There’s another $10 billion unlocked. That’s a $100 billion business, so at $20 billion, it looks pretty good. I agree with you.

And Fractile, dude, they’ve got a $250 million order from Anthropic. NVIDIA is being weakened. You have Cerebras, which some say is in trouble and some say is booming. TBD. Andrew’s amazing. I love him, and I’m buying him and Cerebras, so hopefully it’s great. But we’re going to see the specialization of chips, and it’s a $7.5 billion entry. Come on. Which side are you on, Rory?

Speaker 0

As you’d expect in the late-stage business, you have to assume that every company is good, because if they’re not good, they wouldn’t be raising late-stage money. The universe is not totally idiotic. As I say, at the stage we invest at, there’s business risk up the wazoo with Series A and B, right? By the time you get to these companies, you should be done with business risk. You’re dealing with typically a market leader, a number one or number two, in a big-ass market. Then the only question is value relative to TAM.

So it’s not surprising all 3 companies are great, because they’ve all priced recently at good prices. You’ve got to drill down one level. It’s not enough to say they’re amazing companies, Harry. In each case, you’ve got to get to grips with value.

The thing with Fractile is, to state the obvious, it has the most defensible technology. On the other hand, Cerebras is at a mid-$20 billions market cap. They had a great quarter. They’re going to do just under $1 billion this year, and probably triple next year to $3 billion. So there’s the fast-inference public company that ships the chip, that’s in production, that’s booked $1 billion in revenue this year, and it’s valued at $28 billion.

Would I buy that over Fractile if I could do public or private? I don’t know. I don’t have a good enough sense of Fractile’s edge. You hear amazing things, but I just look at value relative to TAM and relative to comps, and I go, “I don’t know.”

Harry Stebbings

I mean, Fractile’s not the value play. Let’s just be clear.

Speaker 0

No, no, it’s not the value play.

Harry Stebbings

It’s the hope play.

Speaker 0

No. Yeah, yeah, yeah. I mean, Mercor—I think Jason said it right. This is one of those markets where you continually go, “Oh, I don’t know, I don’t know,” and then they just spend more money.

The amazing thing, if you read Anthropic’s S-1, is that 75% of their compute goes toward making the next model. They also disclose some of their training budget, I think. It’s freaking huge. Seventy to 75% of their compute goes to making the next model, right? It’s something like $10 billion. So they’re each going to spend a billion, a billion-something on training data. You’re right: the bad news is you have 5 or 10 customers. The good news is they each spend a billion dollars. It’s good for TAM. I don’t know what it means for value.

And Kalshi, yeah, reluctantly. The great thing there is that sports gambling is wildly addictive. They’re pushing more of the predictive market stuff now, but that’s one where it could be huge. I’d really want to sit down and understand it. You say the market is huge, and it is very big. But you’re already at $65 billion. They’re doing $5 billion in revenue, so they’re growing like a weed.

I suppose my question there would be: What’s a sober assessment of the real TAM? How many bored 25-year-old dudes are there betting on sports? How big can it be, assuming full legalization? And then can you get prediction markets on top? So I’d probably do the work on that.

Harry Stebbings

I think we’re all aligned that we’re due Mercor, actually.

Speaker 0

No, I think Jason said it right. It was cute: I’d always want to do the last round, because you’re like, “Oh, wow, they pulled it off again.” But yes, great business, and they appear to be the winner. I mean, it’s not really a winner-take-all business. You’ve got Surge, you’ve got Scale. But these guys appear to have pushed their way in in a very impressive fashion.

Harry Stebbings

They’ve done amazingly well. Their proximity to the labs is really core to their success, and Brendan’s amazing.

Speaker 1

I think we just put in $200 million and move on to next week, because I like these, but we’re going to have an even more banger group next week. So let’s do $200 million and move on. We’ll own 1%, and even if we don’t make that much, putting $200 million into Mercor at $20 billion, we’ll learn. Even if we don’t make that much, it’ll be good to learn.

And you know what we missed on the NVIDIA thing? You asked about it in relation to IC. I think this whole Jensen thing damaged a lot of things in venture. I think it damaged Harry’s 100% confidence in the liquidation preference. What was NVIDIA going to buy, Reflection? Is that who they were going to buy for something billion? That deal’s dead now. There’s some chance Hugging Face won’t close.

Harry’s running around doing these deals at IC left and right for 12 months, always telling us our liquidation preference—I’m joking, but not. I think the liquidation preference has been weakened by the NVIDIA drama. Because NVIDIA or its friends would always buy it, backstop it, come in and bail out Endscale, or buy out the team. Buy out the Reflection team, the Poolside team, the Hugging Face team. I can’t lose on these deals because of my liquidation preference.

Well, with Jensen gone, maybe you can lose on these deals. I don’t think the CFO is going to buy all these kids and bail out their Poolsides and Toolsides. I don’t think she’s going to do it. And I don’t think when they hire some retired guy from wherever, from AMSL or wherever, he’s going to buy all these kids for $17 billion. I don’t think he’s going to buy Clem for $15 billion to win open source. I think these liquidation preferences are not going to be worth the paper they’re e-signed on, Harry.

Harry Stebbings

Dude, I agree. But the Hugging Face acquisition was perfect timing.

Speaker 1

Assuming it closes. Close to perfect timing. Perfect timing would be closed.

Speaker 0

And the biggest risk is probably regulatory, if we all get caught up in stuff. Okay.

Speaker 1

Yeah. I mean, Elon and Twitter proved it. Sometimes it can be hard to get out of a deal, but there are also many outs. Do you really think the CFO wants to do Hugging Face right now? I don’t know if it’s at the top of the list right now. I don’t think the whole team is super excited about this right now.

Speaker 0

I would say, though, that the cost is small relative to market cap. All other things being equal, they’d want to do it. But I think the issue is more if NVIDIA is going to be subsumed in regulatory morass.

As a random comment on the regulatory side, one of the scariest things if you’re NVIDIA is that it would appear on the other side of the table in Washington there are 2 different sides on the issue. There are the “sell shit to China” people—Luttig and Commerce are doing that—and then you have the “don’t sell shit to China” people, and it seems Marco Rubio and some of the others are on that side. One of the real complexities is that you’re not just dealing with a government that might be mad at you; you’re dealing with a government that might be mad at itself.

So you could easily see getting caught up in a bureaucratic swirl for the next 2 years. If that happens, and the FTC is being asked to approve some Hugging Face acquisition, maybe they kick it back. They don’t want to piss off Marco. They don’t want to piss off Howard. You’re right: the regulatory risk here definitely went way up. I hope they got a good breakup fee, because they deserve not to be screwed over by this. If you don’t think Hugging Face are the good guys, you’ve run out of good guys.

Speaker 1

Sorry, just briefly, we can move on, but you could imagine 2027 being a banger year in many ways. If Jensen isn’t there, and Salesforce doesn’t buy anybody because they’re digesting these deals this year, the world would look very different if NVIDIA isn’t doing deals and Salesforce aren’t doing deals next year, okay? It’d be a very different year from our “you can’t lose with our liq pref” year of 2026.

Poolside could be bankrupt. They didn’t have the capital. That’s why they sold, right? What was plan B?

Speaker 0

Oh, don’t be negative, Jason, please.

Speaker 1

No, no, I’m not.

Speaker 0

Come on. I mean—

Speaker 1

I mean, amazing outcome, but what was great about Poolside was an element of transparency we don’t usually see, right? Everyone’s full of it. This is them saying, “Listen, we cannot afford to keep this business model going. We’ve done a great job, guys, but we can’t raise the $10 billion, $20 billion to go.”

Will Thinking Machines survive? I just don’t know.

It could be a very different world.

Harry Stebbings

We have a friend on the show, Mr. Brian Halligan, back in the hot seat, baby. HubSpot has had a challenging 18 months, I think everyone would agree. Brian and Dharmesh have called it a refounding moment. It’s been awesome to see them get back together. What’s the take on this, Jason? I’ve got to start with you. Did they have to do this?

Speaker 1

Well, look, HubSpot has been under pressure for 2 years, as many have. Its stock price fell 50% this year, but let’s ignore the ups and downs of the stock price. I think Rory will be with me on this one: for years, even during the start of the SaaSpocalypse and the 2024 dramas, they reliably closed 10,000 net new customers a quarter, like clockwork. Those customers had different values and were different sizes, but it was 10,000, 10,000, 10,000. Two quarters ago, it was 6,000. One quarter ago, 5,000. Next quarter, they’re guiding below 5,000. monday.com, which is a peer, went from around 1,200 net new customers to under 200. It fell by 80%.

Something had to change, and it had to change urgently. The tough question, with love, is: will it work? We can talk about why it will always work to have founder energy back. The founder’s ability to make decisions others can’t, and being more connected to the product, always works. But I’m not saying it’s too late, but it’s tough. That decline is already in place. I don’t really care about revenue going up and down or the stock price. The thing is, net new customers. When I started investing, that was my lighthouse metric, and we get distracted. It is the only thing that matters at scale.

People hid this for 2 years with endless price increases. If you have decent NRR and high price increases, you can mask 20% growth for several years, and people who don’t look closely won’t know. But if you’re not attracting net new customers, your business—if it’s not terminal, you’re in deep value mode. There are no new customers. He and Dharmesh have to turn this around almost instantly, or you risk falling below the hard deck of net new customers. I don’t envy them. So it will work. The question is, will it work enough?

Everyone’s in founder mode. Dev’s back at Mongo, who was on the show, right? Back at Workday, back at HubSpot, back at UiPath. It’s all working. The question is, is it enough? The last thing I’ll say is every VC in Yahoo points to Fin and Intercom. Rory was a large investor. I was a tiny investor from the early days. I’m not sure Intercom worked, and I say this with love. I don’t mean that literally, but let me finish my thought. It definitely worked. Owen came back in 2024 and immediately fixed pricing and other issues. That worked before AI. In his first week, he fixed the endless price increases and customer rip-offs that existed before him. Then he went all in on agents and AI at the very beginning. It worked.

However, even with all that, and with the incredible blood, sweat, and tears of the founders staying for the whole journey, Owen coming back only got it to $100 million-plus in agentic revenue. Did it really work? We’ll never know, as part of Salesforce. We’ll never get the next chapter of the story. Every VC who says this is a case study—I think it’s an example, but I don’t think it proves the case that this works all the way through. We didn’t get to see Intercom go to a billion in agentic revenue. We never saw the next act of Eoghan and friends, and that’s okay. People made money. But this was never truly proven, that coming back works. We just know it works to some extent, and it’s really hard.

Harry Stebbings

Well, that was going to be my question: what is enough? You’ve said “enough” several times. I love Daniel and UiPath. I love Dave. But I think it’s all TBD.

Speaker 1

Can you become a growth stock again, or do you become Zoom or Dropbox?

Harry Stebbings

Can you tell me one that has become a growth stock again?

Speaker 1

No. There are some that look like it, like Okta and Twilio. The growth isn’t astronomical, but the stock prices show so much benefit from agentic spend.

Speaker 0

Agreed. There’s no one above 30% in the public markets in that category. Obviously, Palantir is doubling, but to your point, there’s no one from the SaaS era who has reaccelerated back above 30%. I kind of agree with your comment, Jason. I think it’s really hard to dramatically change the trajectory of a billion-dollar revenue company. You’re probably playing for some version of this: instead of having only 5,000 customers going down to 2,000, I can get it back up to 8,000. I can do better. I can point the growth rate up to 25% and survive as a successful, viable, interesting public company that matters.

But there’ll be a 5-year period where the startup, the Sierra growing 4X year on year, will have the growth, sizzle, and cachet that are just very hard to compete with. So yes, I agree. It’s worth one random comment on this, going back to the Jensen thing: it’s kind of funny. It’s worth noting that when 3 or 4 boards felt their backs were to the wall and they had to save the company, they broke the glass and got the founder back. Just saying to the board, “Hold that thought,” because it’s noticeable. For all the faults founders can bring to the table, it’s worth noting that in every case, it was the right move. Dines, I think it’ll be right with Brian. I think it was definitely right with Owen. Dev makes a ton of sense. It’s noteworthy that the person you need is that early-on person who knows what to do here.

But I agree, it’s hard, especially at the app level. I don’t know what Mongo’s current growth rate is. With some of the companies, like Datadog, if you can co-attach to the infrastructure boom, you can get your growth rate up. But fundamentally, going back to your comment, Jason, there are 2 different things. At the app level, if you’re selling a SaaS-based app, the problem you’re up against is that the new thing a customer buys in the AI-first world could be so different that, to a rounding error, all you’re doing is building a slightly better version of yesterday. That’s why I thought Brian Halligan’s comment—I don’t know if I love the positioning—about being the “customer relationship harness” was a clear attempt to position themselves in the new world, which is smart.

I’m not sure the average HubSpot customer—a 50-, 100-, or 250-person company—really has an opinion on harnesses. Remember, we were investors in the Series C, I think, back in 2010, so we did very well. Clearly, what they’re doing is trying to position themselves in that new product world, but it’s hard. It’s really hard because even if your new business goes great, and I think Intercom’s new business was going really well, you’re still trapped in this bigger, slower-growing thing. See Wix and Base44 for details. It’s very hard to dramatically change those trajectories.

Harry Stebbings

I always say to Daniel, “Dude, why would you go back?” It’s so brutal. You have a nice life now. It’s just—oh.

Speaker 0

Man loves to struggle.

Harry Stebbings

Man loves to struggle. Actually, do you know what I see? I see a level of aggression in all of them. They’ve already made so much money, so it doesn’t matter like it used to. They go with unlimited aggression and confidence that they didn’t have before.

Speaker 1

I think so. I will say that’s true. It’s very hard for an outside CEO to do that, and certainly hard for the CFO who was promoted to CEO 2 days ago. Of course it’s a level of aggression that you need, and that’s why I think it will work. I think Brian will work. I think they will all work, as Rory and I said. But I do see a hint of hesitation. Maybe hesitation’s the wrong word. When I talk with a lot of founder CEOs—not necessarily ones who went back, but ones who are there—I hear a hint that they also recognize there may be a limit, to Rory’s point.

Thirty percent may be the best I can do. I already have 320,000 customers at HubSpot. Salesforce hasn’t even disclosed its customer count for about 4 years. That’s telling, despite being a $50 billion-revenue company. They stopped disclosing it. Whenever someone stops disclosing a metric, it’s telling one way or the other. Salesforce is clearly a master of selling you 5 to 10 expensive products, but not so much of bringing in net new logos.

I think people who are off-the-charts smart, and I’m sure you can ask Daniel, probably also have a limit: “Okay, I can only do so much with agents and LLMs. No matter what I do with UiPath, it’s never going to grow like it did in the past.” There’s a certain logical way to think about this: maximize the realistic outcome, upgrade the team, do the layoffs, do whatever. But there’s a hint of hesitation that maybe whatever limit there is, that’s as good as it can be after 15, 20, or 25 years. So I don’t know.

Speaker 0

To some extent, that’s not surprising. One of the things we said back, I think in 2018, was that we foresaw this SaaS slowdown. It’s not because AI is going to replace it as much.

That was the second issue we said at the time, but the more important thing is that the TAMs get saturated. People have a CRM. People have a HubSpot. Some of these markets are more finite than one thinks. It’s hard to grow a customer count infinitely, especially on the enterprise side, so I’m not surprised by the Salesforce comments.

Speaker 1

Well, look, that’s true, but for what it’s worth, there’s a mystery to me here. The fact that Monday and HubSpot have seen their net new customer count decline so radically over the last 3 quarters—that is a fundamental change in the way SMBs are buying software today.

This is a radical change, and I don’t think it’s vibe coding your own software. That’s ridiculous.

Harry Stebbings

So, Jason, who are the winners here? Is this like Attio and Zero and the new CRM companies?

Speaker 1

If you look at the Ramp data, I guess Attio’s benefited, but they’re so small in the grand scheme of things that it’s a rounding error.

Speaker 0

I don’t think so. I think the winner, candidly, is probably Claude, in the sense that for small SMBs, where the workflow and the auth and all the corporate legacy stuff isn’t that important, you can crank something out now that will suffice as a CRM app.

Speaker 1

Right?

Harry Stebbings

No.

Speaker 1

Using Lovable.

Harry Stebbings

No. The people who use Monday aren’t going and vibe coding their own CRM with Claude Code.

Speaker 1

You can do it yourself, but, look, I would’ve said—I said it on the show 6–9 months ago, and every single associate and principal here has cranked out their own front end using Claude for our Salesforce instance. No one logs into Salesforce anymore. They’ve built their own thing. They write—

Harry Stebbings

I’m not surprised. They went to HBS and work in San Francisco. They’re not Caesar the pizza maker.

Speaker 1

I think the answer, if you want my opinion, is in the middle, okay? To Harry’s point, I don’t think your average SMB that isn’t tech-savvy is vibe coding its own front end to Postgres or Salesforce or whatever. I just don’t; I see no evidence that’s happening. I’m pretty close to the Replit team. I can tell you CRM isn’t one of their core verticals, at $1 billion in revenue. It’s just not. I briefly asked Anton. He never brought it up as something they’re doing. I don’t think it’s a big deal.

But here’s what I do think is happening, to Rory’s point. I just think folks are trying not to buy the software for SMBs. They’re just saying, “It’s not so much that I’m buying it.” They’re sitting around, and in the old days, when we started, you’d bring in a VP of sales. What’s the thing he or she would do in their first week? Bring in Salesforce. “First week, I gotta roll it out, and I probably have to hire a consulting firm to help us,” small, medium, and large. That was a day-one kind of thing.

You can’t argue with the data. Is Attio benefiting? Yeah, Ramp says a little bit—another $10 million in revenue. That’s not enough to explain what’s happening. SMBs are saying, “I don’t want to buy this crap. I just don’t want to buy it.” Whether it’s doing a little bit of data analysis in ChatGPT without knowing I’m doing data analysis, or running it in Google Sheets for another 6 months because, with ChatGPT, it’s good enough, it’s just, “In my small business, I don’t want to spend $40,000 a year for HubSpot right now. I just don’t want to do it.” I think that’s what’s happening, or with Monday. Monday’s not even expensive, but—

This is something I didn’t anticipate: SMBs would just say, “I don’t want to buy this stuff.” On the other hand, in my portfolio, selling to SMBs, anyone that’s automating away front-office or back-office humans is exploding. My God, it is the easiest thing. Nothing’s easy to sell, but I have 3 portfolio companies selling front-office and back-office agentic solutions for $500 to $1,000 a month. As long as they can get a prospect on the phone, they can close it that day.

But they don’t want to buy these software products. It’s not enough. They’re just gonna bumble along with what they do. It’s not a P1 or a P0. You know what my P0 is? My receptionist quits every Wednesday. “Oh, you have an agent that will do that 24/7, reschedule all the appointments, handle billing, handle everything for me, and it actually works? $800? I’ll sign this afternoon, kiddo.” That’s what people want to buy, not Kanban crap.

I didn’t anticipate this because we grew up in a world where you couldn’t survive without the software, right? You couldn’t survive without it. Clearly, people can, and they want to put that energy into things that are different. And that’s what’s happening. They may buy HubSpot or Monday or someone, or Attio, but they may put it off a year or two, and that may end up being never. We don’t know.

That’s my theory because I looked at the Ramp data and others. It’s not substitution; that’s not sufficient to explain this. They’re just done. They want to do stuff that has high ROI. This used to be high-ROI software. It isn’t today.

Harry Stebbings

I feel we’ve gotta talk about Microsoft going down for 2 days. I mean, seriously, Rory, you’ll be surprised: I’m still dating the same girlfriend. She stayed with me.

Speaker 1

Good for her.

Harry Stebbings

Short horror story: she wakes up and she’s like, “I got no emails. I don’t know what to do today. Teams is down—” Copilot’s down, Outlook’s down. Down for 36 hours or so. And then we had flights grounded in the US, FAA problems involving L3Harris and one of the contractors.

Speaker 1

And it’s worth pointing out, based on the initial comments, that in neither case is there an allegation of a severe cyber issue. As I understand it, it’s not a cyber issue. It’s not AI cyber. It’s not even, directly speaking, AI. These are old-school problems that people just missed.

In fact, in the case of the FAA, it was a router issue that kicked off the whole thing. In the case of Microsoft, I don’t fully know, but it appears to be some kind of network configuration thing. I’ve seen a post that basically says, “It turns out if you whack all the senior engineers and think the AI’s gonna figure it out, this is the kind of thing you miss.” I can’t comment on that. I don’t know. But they appear to be fundamental, classic IT screw-ups that have reverberated quite a lot.

Harry Stebbings

But the NSA went down today as well for their internal comms, and it’s down now. I’m just like, hang on a minute. The core systems that run our society—air traffic control, Microsoft. We’ve never had this fragility of systems in my lifetime, it feels like.

Speaker 1

I think they’ve been more fragile, but here’s how I think about it. I think it’s DDoSing 2.0, or DDoSing for the AI era, okay? You could always take down a website back in the day. Websites used to be so fragile you could DDoS anything. You could hire some nefarious agents offshore, and I could take down Scale’s website if I was pissed off, right? It was pretty easy to do because we could bring down those Pentium servers pretty darn quickly. We could find a way to DDoS it.

This Microsoft thing apparently was a load-related bug that existed for 4 months, and they hit some sort of maximum load they hadn’t modeled for, and it went down. I completely agree with Rory: as our DevOps and SecOps teams decline in quality through optimization, this is gonna happen more. But also, not only is everyone gonna get hacked, like we’ve talked about for months, and now we’re seeing it happen constantly, right? Google said we saw the first truly autonomous agentic hack this week.

Everyone’s gonna be DDoSed. Everyone’s servers are gonna be overloaded. It’s too easy to spin up agents. It’s too easy to break Microsoft. It’s too easy to break things, and every system can be broken and is fragile. So it’s just another thing we didn’t anticipate being able to do, which was attack anything. Was it really just an old Level 3 Linksys router from 1998 that took it out, or was it a load issue, or was it someone probing the edges of a network?

I think every network’s gonna be probed. It’s gonna be probed with traffic, with load, with hacking. All the vectors are gonna be hit. Everything’s gonna be hit, and it’s not all gonna be about security and stealing credentials and API keys. It’s gonna be load-based, too. Anything that can be taken out with massive load, you’re gonna spool up agents and see if you can take it down. Let’s take down everything. Iran will spool up 10,000 agents on Quinn or whatever and see who they can take down. It’s not that hard.

And this is a tough one: cascading failures. If you’ve ever run a B2B service, few things are more stressful than cascading web servers. All of a sudden, you have 1,000 servers, 10 go down, and it’s really stressful. Then you sort of fix those, and another 10 go down. Now you spread the load—1,000 loads on 900, 800, 700, 600, 500—and then you wanna jump off the roof because it gets worse and worse. Even as you find the solution, it gets worse and worse. It’s gonna happen to everybody.

Speaker 0

And I think even though those 2 examples weren’t obviously AI cyber-related, we’ve definitely seen a plethora of other companies wrestling with smaller cyber breaches in the last 3–6 months that are probably related to the ability to hack almost anything more quickly. So there’s a lot going on.

Speaker 0
This is not perhaps P. Doom or any of that stuff. This is one of the big risks in the AI boom: when you see what it’s like to be without something. Frankly, I could live without Outlook. I didn’t even notice, and I don’t love Teams. But when you see the FAA go down for a day—or can you imagine a week?—that would start to get economically very painful. It highlights the issue.

Harry Stebbings
Dude, the weekend was economically very painful. By the way, I’m with you completely on Teams and Outlook. The devastation that causes for four hours of downtime.

Speaker 0
It’s probably going to mean all these companies really have to think about what level of competence is required to deliver these systems. There’s probably going to have to be a substantial uptiering in the seriousness, professionalism, and capabilities of any company offering a mission-critical product in today’s world.

I remember when DocuSign suddenly started to become huge, and I had one really good board member with a cyber background who basically sat everyone down and said, “Look, now you’re going to get hacked. This is what you have to do.” And they did it. We spent the money to be as reliable as it takes. You can’t continue to have the FAA run on a shoestring. Did you see the thing? They just announced they’re getting rid of floppy disks. Ironically, floppy disks are probably impervious to many of these issues.

Speaker 1
They are air-gapped.

Speaker 0
There is going to have to be a lot of spending on making these systems work and making them durable, because we can’t be doing this two weeks in a row. If the FAA went down again this weekend, I think people would lose their collective shit and probably put Sean Duffy’s head on a spike.

Harry Stebbings
It’s fucking wild. Have you seen Die Hard 2, where they’re unable to connect air traffic control to planes? That’s what was happening.

Speaker 0
Yeah, that’s terrifying.

Harry Stebbings
Fucking terrifying.

Speaker 0
No, totally. I read the article. There are 4,000 planes in the sky, and you have a nationwide ground stop.

Speaker 1
I know that one’s scary. I just think, more tactically, in relation to venture and startups, the Optimizely thing is worse. I don’t mean in terms of global impact, because Rory made a really important point. On the one hand, we’re in the era of “Can one person build a billion-dollar startup?” On the other hand, we’re in an era where I don’t think we can trust anything except companies with S-tier security teams.

Optimizely is a moldy oldie of B2B, and it gets prompt-injected or whatever happens. CNN, the BBC, Zoom—everyone’s using it. Credit cards and PII are being stolen. Every mediocre third-party plug-in and app is a security issue. That’s what happened to Fireworks, too, right? Then Notion and who else blamed Fireworks by name, which you never see before. Before, you’d always see, “One of our vendors had an issue. It impacted less than 1% of them.” Notion sent an email to its entire 1 million-customer base attacking Fireworks. I’m not picking on them—I am a little bit—but this is where it plays out.

On the one hand, we want one-person, billion-dollar startups with no security. Nothing. Have you ever built a real production app yourself? You have no security. You can do security audits, on the one hand, right? We want this massive innovation, but these products are going to be totally insecure, and I don’t think you can trust any old product that’s a plug-in. I just don’t think you should use any web analytics. I don’t know if you should use any new payment startup besides Stripe or Adyen. The risk here is so high with automated, agentic hacking, and how that plays out for startups is confusing to me. On the one hand, we can scale like never before. On the other hand, your security risk is like never before, unless you’re the best of the best of the best.

Harry Stebbings
It goes to the point you made, Jason. The scary thing is that the Fireworks team is exceptional. The engineering team is truly best-in-class. If their security team and SecOps team are 10,000 times better than 99.9%, then I really am quite worried.

Speaker 1
Yeah, I’m sure they’re great. It’s their job. And it happened in a corner of their application, too. That’s where it often happens. Whether it’s a human or AI, you’re looking for the edges and the corners. It’s not like all of Fireworks’ data was exposed, right? It was this corner for testing where they stored some keys, and that’s where they found their way in. That’s with probably 50 great people in SecOps and DevOps.

Oh yeah, Cursor called them out, too. It’s pretty bad. It just shows the stakes, right? Notion, Cursor, and Harvey would call out a vendor. I remember when Harry and I were in London about a year ago, at SaaStr AI London, and OpenAI had a breach with a third-party vendor while we were there. I can’t remember which one it was. I remember they didn’t name them. They canceled the contract, but I forget who it was. It was a classic, “We’ve had a small issue. It was only with our CRM and external systems. It didn’t access any of our data.” But now it’s Fireworks—Notion, Cursor, and Harvey: Fireworks, Fireworks, Fireworks.

Harry Stebbings
Dude, I think we’ve just lost all elegance in this business. Before, VCs didn’t bash each other as much, and companies didn’t bash each other. Now it’s an all-out brawl. I do want to touch on one M&A deal that seemingly flies under the radar, but I actually think it’s quite interesting. It’s OpenAI buying a company called Edesia. It’s like Instinct, but for travel specifically, and it’s the founder of Sonder, Francis Davidson. My question off the back of that is: Does this signal the end of verticalized personal assistants—your financial personal assistant, your travel personal assistant—and is it just a horizontal play? Any thoughts on this?

Speaker 1
Well, he founded Sonder?

Harry Stebbings
Yeah.

Speaker 1
I love Sonder. It just ended up being a terrible business model, so I’m happy for him because it was one of my favorite products. I stayed in a three-bedroom Sonder in Covent Garden every time I was there. It was a gem, so I’m glad he had a win in the end. Seriously, it makes my heart warm as a founder.

I’ll tell you my one learning. Everyone comes on the podcast—we had Jack Altman and others—and asks, “Why aren’t you using Instinct, Jason?” I’m so deep in Muse now. You don’t know all the workflows we’re running across SaaStr and Muse. We’re so deep on Muse.

Harry Stebbings
Muse?

Speaker 1
Our horizontal agent. Muse.

Harry Stebbings
Wow.

Speaker 0
Wow.

Harry Stebbings
What do you use it for? Tell me.

Speaker 1
Everything. We use it for outbound and competitive reviews. I use it to edit all my content, create clips, and find all the gaps in everything we’re doing. We use it to analyze our web traffic and our SaaStr Annual attendees. It does everything constantly, around the clock.

I have so many workflows in it that the burden of using another one is pretty high. It is a shallow moat, but it’s become wide, with 20 or 30 workflows. Any single gap we have—and we have so many agents at SaaStr, 21 agents and 3 humans—we spin up a Muse. Our Muses talk to each other. Amelia’s and mine talk to each other. They have their own web chat channel and email address. They don’t tell us; they talk to each other. Whatever I’m working on and she’s working on, our two Muses can collaborate. Carson and Friday are their names.

It also reviews every single inbound venture deal I get. It re-ranks them automatically and sends me an update. It reads the ones I didn’t read, and it reminds me if there are any good ones I didn’t get to give attention to. It does that every single night.

Harry Stebbings
So you’ve given it full access to your email?

Speaker 1
Yeah, full access, full-time.

Harry Stebbings
Credit cards?

Speaker 1
No. It did order flowers for a big deal closing on the team without telling me, so maybe that’s a thoughtful reminder about the flowers. No credit card yet. But it’s so useful that I’d probably go to the effort of locking down a Mercury account with a $1,000 cap if I needed it to buy stuff right now. We don’t need it. We have so many other systems for that. We have Bill and Brex and others, but I would if we didn’t have them.

Harry Stebbings
Are you more bullish on Matter as a result of your exposure to Muse?

Speaker 1
What I’m worried about for startups and others is that, as humans, we go out to buy a vendor. If we’re early adopters, we see what the hell’s out there. Bless their souls—the CIOs with emerging-vendor groups and stuff like that. Most of us, as humans, look at the two leading vendors and do our best to evaluate them when we buy one. Maybe over two years, we turn over.

Agents have a different way of doing it, but once one really works, you’re not going to change unless there’s a reason to change.

Speaker 1

It gets locked in. I think Mark was right to push this when he did, because January would be much harder than now. He’s going to have a 20-million-person install base to build on, and I don’t think people want to go from Muse to Instinct to Groq to Dots and back and forth. There’ll be people in tech who do it, but once you have these workflows running and they’re really, really good, you don’t want to move them. You can, but you don’t want to.

Harry Stebbings

I’m with you. I have 50 automations running on Instinct, and it’d be a real ball ache to have to change them all. Ugh.

Speaker 1

You would do it if there’s something important that you can’t do in Instinct. For example, Muse is very, very good at headless browser interactions. You might find there’s something it’s so good at with a headless browser. I’m not saying this is true of the Instinct team, but you might find something that’s really good at headless browsing. Muse is the best at it. We still get 1 million views a month on our WordPress content. Nothing else I’ve found can interact with WordPress as an agent. Muse does it perfectly.

Let’s say you had a WordPress workflow. You might use it just for that one thing, and then give it another. But it’s because it can do one thing that’s at least a P2 that you couldn’t get otherwise. Otherwise, you’re not going to move off Instinct.

Harry Stebbings

Jason, how much do you think you cost—

Speaker 0

Exactly.

Harry Stebbings

—Mr. Zuckerberg? You said 4 or 5 bucks last week, and I was like, “Now?”

Speaker 1

No, we spend $20, and we get whatever the Pro is. How much? I don’t know, $200, $500 a month, $1,000 a month for $20. I’m sure with Claude, with my $200, I’m spending $10,000 a month. It’s a lot.

Speaker 0

The good news for both companies is you’re not the typical user. Their business model will survive, and the poor dude who just—

Speaker 1

No, maybe. Or maybe I’m the future, just in the present.

Speaker 0

It could well be that you think of yourself as the future.

Harry Stebbings

Now, that’s the caption: “Maybe I’m the future.”

Speaker 1

Honestly, Rory, I don’t mean that with hubris. I mean it—

Speaker 0

No, I know. I understood what you were saying. I thought it was a—

Speaker 1

I mean it as something to learn from. I mean, to learn from, right?

Speaker 0

No, exactly. What percentage of people will have that kind of set of interactions on an ongoing basis is the question.

Speaker 1

I think what’ll happen in B2B is you’ll wake up, get the quote from the HubSpot rep—it’s $30,000—and realize Muse can do it. Not all of it, but more than you thought. You might put off that purchase for a month because it turns out Muse can go to your email list and email all the 20VC listeners.

You know what Muse is really good at? Going into the Scale database in Salesforce, finding the top 200 founders we want to invite to a VIP dinner and cocktails, and inviting them. And it does it. You’re like, “Why do I need marketing automation software?” You might do that without realizing you’re replacing marketing automation software, because you just asked Muse. I guarantee you, if your Muse is hooked up to any CRM database, it can do just what I said, because we’ve already done it: invite the top 200 CEOs to the 20VC office tomorrow night.

Probably Instinct can do it, too. That’s what you pay $100,000 a year for marketing automation software to do, except humans had to figure out the list, pick it, get it wrong, and accidentally invite the failed CEOs.

Speaker 0

The interesting question is—look, we don’t want to get into it—it’s so funny: Muse, Facebook, the ultimate consumer company. What you’re basically saying is it’s powerful enough to effectively be a small-business agent, right? And the question is, will there be a better small-business agent you’ll switch to, or is this in fact the right answer for you?

Speaker 1

I think for me, technically, the interesting thing people don’t get about Muse is that it is so overengineered.

Speaker 0

Agreed.

Speaker 1

It is so overengineered because of the team building it, because of Alex and all these great guys. The best of the best built this. They talk about it being a little consumer app to make reservations, but it is the ultimate expression of OpenClaw, and it is amazing software. It’s the best software I’ve used since Anthropic 3.x got good. It is so overengineered for the use cases in the popular media. It’s epic.

Harry Stebbings

Jason, can I ask you a really hard one? For someone today who doesn’t have Muse, what would you tell them is a great first use to see its power and get started? What would you say?

Speaker 1

It’s easy. First of all, Muse is free for a lot of use. It’s the most generous free tier I know of. So sign up; it’s free. If you don’t know, ask. If you have any friends who have Facebook, ask them for an invite. It’s pretty easy to get.

Here’s the magic: Look around your little business, your little office. What isn’t getting done, Harry? In your general business workflows, what isn’t getting done? Just ask Muse if it can do it. It probably can. This post didn’t get done. This email didn’t get done. We’re not following up with customers.

We had a dinner last night at Scale. Did you follow up with the founders and ask them how they did? No, I asked my secretary, but she didn’t want to do it because she didn’t want to figure it out. Ask Muse. Muse will figure out everyone who came to the Scale dinner last night, craft a really good email, and ask them how it was.

Think about something you just didn’t get done and ask Manus. You’ll be shocked: I think 5 times out of 10, it’ll do a better job than whatever humans you have left in your office. Try this. Your jaw will drop. That’s because of the memory, the browser agent, how it handles everything, the quality of the LLMs, and the way the LLM is tailored around these workflows rather than arguing with you or being developer-focused. It’s very task-focused. You’ll be shocked by what it can do.

Want to know a little more about AEO or GEO? Just ask. Ask Muse, “How’s my AEO?”

Harry Stebbings

Jason, does it make you more or less bullish on Instinct?

Speaker 1

As I said in our IC, there are some vendors where I worry about runaway infrastructure costs. There have to be some limits. It can’t just be Rory’s endless hand-waving about great founders figuring it out. I’m being slightly facetious, but I’m not bullish on it today. I haven’t seen all the justification for how simple this will get.

I’m not bullish because I do think this product gets maimed if I can’t run this thing, this gift from Meta, which lets me run these agents 24/7.

Harry Stebbings

It’s got a cash cow that powers the business—

Speaker 1

Yeah.

Harry Stebbings

—and if you are n8n—

Speaker 0

I mean, just a comment: If the typical user of these products is a Jason who wants a whole bunch of B2B functionality, then by definition he’s going to gravitate to the company that can afford to give him all that functionality for one-tenth of what it costs them to deliver it, which is probably going to be Meta rather than Instinct. But the other question is, if the vast majority of users today—and I accept your point, Jason—don’t need that level of compute and aren’t running their entire SMB on the infrastructure, then at least Instinct has a chance to compete and maybe can win by moving faster and being standalone.

But it is scary when a large company with 5 billion, 4 billion of human reach is producing a product that’s arguably more overpowered technically than your product and has distribution up the wazoo. That didn’t work for Threads, but there you go. This came a long way from that little question about the one-person travel agency at the top of the meeting—

Harry Stebbings

Yeah, and I freaking—

Speaker 0

—but what a lovely diversion. That was awesome.

Speaker 1

Oh, sorry. What happened with Edesia [?]?

Harry Stebbings

Oh, he got bought—

Speaker 0

They bought the company. They bought Edesia [?].

Speaker 1

Oh, as agents. He bought it for the agent, for agents, right?

Harry Stebbings

Yeah.

Speaker 1

I guess my comment is, I know this is Captain Obvious, but I think we’re past the era of agentic apps. We’re past the era of AI, CX, and crap. We’re truly in the agentic era next year, with these horizontal, autonomous agents. This is the big win for most of us at a lay level.

Most of us are not—Claude Code is the greatest app of all time, but most of us as humans are not Claude coding all the time. This gets good next year. It gets good with agents. We’re going to see 20 of these Edesia [?] deals to try to win.

Harry Stebbings

The biggest challenge I find is actually workforce mindset. We have big teams here now, and it’s really about getting people to ask, “Could an agent do this? What can I replace with agents? What are we not doing that agents could do?” It’s not a primary mentality that a human brain that’s worked for 10 years can absorb.

Speaker 1

Yeah, and those people will disappear. This is a transition phase where you’re trying to get humans to be agents. When you do what I just did and ask Manus to do it first, you just won’t hire that person next.

Speaker 1
Our agents do all of our collections. Our previous collections contractor billed us $3,000 or $4,000 a month; now we get billed for 1 hour. It will just get inverted. We’ll stop trying to get humans to work with agents and start asking agents to do it, plugging humans into the gaps. It’s a transition phase, and what’ll happen is you won’t even fire them. They’ll think they get a better job and they’ll quit, and you don’t—you’ll stop backfilling people, is what will happen.

Speaker 0
Yes, it’ll just—

Speaker 1
You just won’t backfill them.

Speaker 0
…ebb slowly away.

Harry Stebbings
I sat down with one of our guys and said, “Just to let you know, I agree that right now, the clips you make can’t be done with AI. In a year, they will be. You need to get better, or you will be replaced.”

Speaker 1
I think that’s right. I tell you, we went from 20 people to 3 humans and 21 agents. You say that, and it’s true. You know what it requires today? It basically requires managing your agents for even more time: 6 hours a day, 8 hours a day, 10 hours a day. Do you want to be sitting in front of Muse for 8 hours a day? It doesn’t have to be you, but it’ll be someone on your team, right? So it’s not—

Harry Stebbings
I just wish I was Menlo with the carry. Fuck off.

Speaker 0
It’s good to see you’re doing it for the love of the game, Harry.

Speaker 1
Oh, come on. Yeah, I’d rather be Sound Ventures or Menlo. Who would you rather be?

Speaker 0
Yeah.

Speaker 1
I’d rather be Sound Ventures.

Harry Stebbings
I would too. $435 million to $7 billion. Ashton Kutcher, well done.

Speaker 1
Yeah, 2 GPs. 2 GPs.

Harry Stebbings
Yeah.

Speaker 1
Call it a day.

Harry Stebbings
Yeah.

Speaker 1
Don’t have to work for anybody you don’t want to.

Harry Stebbings
Amen. Norwest raises a $4 billion fund from one LP, Wells Fargo. Jesus, I wish we all had a Wells Fargo in our lives. SpaceX guts its Grok video team.

Speaker 0
No, if you’re gonna do SpaceX, I’m gonna stop you there. You’re at the tail end of the agenda. If you’re gonna do SpaceX, you can’t talk about Grok; you’ve just gotta give them credit for Starship. Just wanted to say a big-picture thing.

Harry Stebbings
Oh, no, yeah, we can do Starship. Starship, oof—

Speaker 0
Yeah, I mean, there’s not much to say except that is a freaking amazing technical achievement. Honestly, I personally find the interpersonal stuff on the guy to be a pain, and the politics to be a pain, but that’s a stunning, stunning, stunning achievement. When you look back on the top 5 things of the year that matter most, that’s gonna be one of them. All the orbital costs plummeting by virtue of that. I thought it was just cool, and it’s a sustained effort over half a decade to make it happen as well.

Harry Stebbings
I get you, dude, but it’s not SMB CRMs, is it? I mean, come on.

Speaker 0
No, it’s not gonna sell any HubSpot, but it is stunning to realize that now the whole rocket launch business is less than 10% of SpaceX’s market cap.

Harry Stebbings
Rory, would you buy SpaceX?

Speaker 0
Not at this price, because it’s not about the rockets. The truth is, the rockets, which are amazing, unique, and differentiated, make Starlink, which is very amazing and reasonably differentiated, especially for the non-crowded-area stuff. But then most of the business, most of the market cap, and most of the TAM, according to their own S-1, is all about compute. So really, you’re asking yourself: Would you buy compute? That gets into the whole dynamic of forecasting compute for the next 5 years, which is just a different question. I might buy it, I might not buy it, but I wouldn’t buy it because of this amazing technical achievement. You buy SpaceX stock entirely on the basis of what you think Anthropic will be paying you 1 year from now for that compute. Will they be canceling it, or will they be relying more? It’s the only question.

Harry Stebbings
Well, that sounds like a long-only investor right there. No, I think you buy SpaceX because you read Sean Maguire’s Excel model and you’re like—

Speaker 0
People gave him shit for that. I thought it was great. I saw the comments on that. Everyone was like, “Oh, you didn’t take into account tax.” Dude, it’s the big picture. How much money, 5 years from now, from each of the 3 markets? You don’t have to believe in it to say, “Yeah, that’s a perspective.”

Harry Stebbings
All right, boys, anything I’ve missed?

Speaker 1
Should we record a little bit at the end in case Jensen’s back by the time we do it? Boys, I wasn’t surprised at all that Jensen would come back. I told you it would happen by Thursday. Altman was back in, what, 80 hours? How long was it for Jensen, 100?

Harry Stebbings
Do you know what, Jason? That’s a really interesting question. So I predict that Jensen will come back.

Speaker 0
In the next week or over a period of time?

Harry Stebbings
Oh, no, I think Jensen will come back in the next 3 days.

Speaker 1
So let’s record it now so that you’re right either way.

Harry Stebbings
He’s gotta come back in the next 3 days. Otherwise, the chorus dies, the Anthropic IPO happens, Trump does something else, and it fizzles. This is the mother lode.

Speaker 0
I don’t think that’ll happen. I mean, it’d be great, but—

Harry Stebbings
But then if it doesn’t happen, it won’t happen.

Speaker 0
Well, yeah, some things don’t happen, Harry.

Harry Stebbings
And that’s Jensen and the board’s mistake: news cycles have never been faster. I think if they’d just put it on chill, then the Anthropic IPO would’ve happened, Dario’d be outside the stock exchange, woo-hoo, woo-hoo, and it would’ve fizzled.

Speaker 1
Okay.

Harry Stebbings
Boys, what a joy as always. It's a wrap. But before we leave you today, Base44 is one of those products I used, and I just thought straight away, "Man, I wish I'd had this when I started 20VC with $50 and no contacts in venture." I wanted to see what it could do, so I built a guest research hub for the show. Upcoming guests, company notes, previous interviews, sponsored contacts, follow-ups, all in one amazing location. I described it in plain English. A few minutes later, I had the product open in front of me. Database, login, back-end and hosting already handled and just all in one. It made it so easy, and that's what I loved. I didn't write a long spec. I didn't stitch five different tools together. I built the thing, and I started using it. For any founder listening, that's the point. The faster you turn an idea into something real, the faster you launch. Apps, dashboards, internal systems, AI agents, whatever you need to build, Base44 gets you there. Go to base44.com/20VC. That's base44.com/20VC. While Base44 helps you build, AlphaSense helps you find answers really, really fast. Quick one for anyone doing diligence in venture right now. AlphaSense just launched SuperAnalyst, and it's one of the more genuinely useful AI products that I've seen come out of this space. It's an always-on agent that works like a power user inside AlphaSense but on your behalf. So check this out. It pulls from over 500 million premium sources, covering both public and private companies. That's filings, earning calls, broker research, expert transcripts. And it doesn't stop when you close your laptop. Point it at a sector, a company, a thesis you're tracking, and it keeps working in the background. And here's the part that I really care about. It remembers your process. It learns your templates, your firm's logic, and it applies that to every new task. So it compounds. It gets better the more you use it. Isn't that the value? And every single output links back to its source. So there's no black box. There's no trust me, and you can see exactly where something came from, so you have complete traceability. Pick the deal memo or the market map you've been putting off and hand it over to SuperAnalyst instead. Start your free trial at alpha-sense.com/20VC. That's alpha-sense.com/20VC. While AlphaSense finds insights, Intercom helps you serve customers. 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.
