Ep19. State of Venture, AI Scaling, Elections | BG2 w/ Bill Gurley, Brad Gerstner, & Jamin Ball

00:00:00.000 | if you're on a run rate where you're raising 5 billion bucks every two years, okay? So give or

00:00:05.360 | take that's, you know, $50 billion over a 10-year period of time, a 2% on 50 billion, it's a billion.

00:00:11.920 | Okay? A billion a year.

00:00:14.640 | Right. Let's just be clear.

00:00:16.640 | Right. So underscore that for everybody, Bill. Great to have you here. Great to see you in

00:00:34.080 | Austin this weekend. Congrats on 25 years of marriage.

00:00:36.960 | Thank you very much.

00:00:37.680 | That was, it was a really fun time. And you know, I went by J Cal's, saw his new place.

00:00:44.000 | Yeah.

00:00:44.480 | Saw so many friends when I was there. Austin's really, really bumping right now.

00:00:48.320 | I agree.

00:00:49.040 | Seriously, just an amazing number of people now calling at home. And I know that,

00:00:54.240 | you know, you recently took your dad who worked at NASA down to see the launch of Starship 5.

00:01:01.040 | Yeah. And I, look, I'm so grateful that SpaceX was able to make this happen for my dad's 88.

00:01:08.240 | And we'd been talking about getting him down there to see a launch. And they

00:01:12.080 | put us in this VIP section, which made it easy for him to move around. And he was just in awe.

00:01:18.400 | You know, I'd maybe start this discussion by telling you, on the way home from the launch,

00:01:23.760 | I did a search for my dad's name and NASA. And I found this document. It says, "A sketch prepared

00:01:30.080 | by John Gurley demonstrates the spacecraft's skip when entering the Earth's atmosphere."

00:01:35.200 | So this was hand drawn by my dad.

00:01:37.200 | Come on.

00:01:38.080 | And if you look here, "Memorandum John R. Gurley to Chief Flight Operations." That's Chris Kraft.

00:01:45.040 | So my dad wrote this to Chris Kraft, "A study of skips range sensitivities and allowable errors

00:01:52.320 | in exit conditions applicable to Apollo missions."

00:01:55.280 | How cool is that?

00:01:57.040 | June 12th, 1963.

00:01:59.280 | And you just took him down to Boca Chica?

00:02:01.360 | He was 26 years old when he wrote that.

00:02:04.160 | What's the average age of the guys down at Boca Chica?

00:02:07.200 | 26.

00:02:08.080 | Oh, 29.

00:02:09.760 | Yeah.

00:02:10.320 | Yeah.

00:02:10.800 | Yeah. What was it like being down there with your pops?

00:02:13.040 | It was awesome. It was awesome. So I grew up in a town called Dickinson, which was near Clear Lake.

00:02:20.000 | And my father worked for NASA. And he worked there from very early. When they commissioned NASA,

00:02:26.880 | they took 50 people from NACA, I think it was called, which was on the Langley Air Force Base.

00:02:33.840 | And that's where he had started his career. And they moved him to Houston. So I'm a Texan,

00:02:39.200 | probably because of LBJ's pork barrel politics. But we grew up on a space street. Half the fathers

00:02:45.920 | on the street were NASA. And the Apollo launches, it's just like you see in the Apollo 13 movie,

00:02:52.160 | you go over to someone's house, if you didn't drive 16 hours to the launch.

00:02:56.320 | Yeah.

00:02:57.120 | And yeah, it was just part of our lives. We had a wall in the house with all the Apollo launch

00:03:03.280 | things. And it was really like until he took early retirement in the late '80s, it was just

00:03:10.480 | part of our lives. And so to have him down there, and he's-

00:03:13.920 | So you went for Starship 5, the launch.

00:03:16.240 | Yeah. And I'd been trying to do it for a while because he's 88. And all of his former colleagues

00:03:22.560 | that are still alive are kind of in awe of what Elon's doing and super appreciative. Because

00:03:30.160 | for them, the space mission was something that meant a lot to them. And they kind of felt like

00:03:36.960 | the government quit trying. And so to see it rebirth just gets them super excited. And of

00:03:44.400 | course, we were there to witness. It was very lucky that this was the one I took him to because

00:03:49.280 | we witnessed the booster coming back, which was-

00:03:52.960 | How was that? Was it emotional for your dad?

00:03:55.840 | Yeah. Oh, definitely. Definitely.

00:03:57.600 | Yeah. And you look at it today, the average age at SpaceX, probably about the same age

00:04:03.360 | as your dad was when he- Yeah, he talked about that. We

00:04:05.680 | went to an after party and all the SpaceX, not all of them, some of them were SpaceX employees

00:04:11.120 | were there. And it was a very young crowd. Yeah. But that to me is part of the magic,

00:04:16.800 | right? That believing you can see how mission driven they are. And we have lots of friends

00:04:24.640 | who work there, work there. And it's great to see America cheering for it.

00:04:28.480 | I've even heard stories that people, just because you start looking around, I've heard stories,

00:04:35.200 | people leave SpaceX and go work somewhere else and just inevitably come back.

00:04:40.480 | Elon has said that we need aspirations. We need dreams. We need to think about the stars in order

00:04:49.440 | to inspire us to invent. And that's part of the purpose of being human. And I certainly know with

00:04:56.480 | my kids and my family, we feel that way. So, super cool you're able to take your dad down there.

00:05:02.320 | Yeah, it was great. It was great. Great.

00:05:03.760 | On Sunday morning, I went on a walk with our friend, Michael Dell. I always learn something

00:05:09.600 | from him. And the thing that dawned on me, he's not one to get hyperbolic about things. He's been

00:05:17.600 | at this a really long time. And he's like, we're in that grindy phase now where he's really starting

00:05:23.520 | to see the benefits of AI, right? Inside Dell, inside his customers, other enterprises, finding

00:05:30.160 | ways to take, to find efficiencies in their business. I think we're entering that, what

00:05:36.480 | could be a really golden era where there's market leaders extend their lead, where their margins

00:05:42.000 | expand. I just saw a headline today that Visa is finding more efficiencies from AI, is letting

00:05:48.720 | 1400 contractors and workers go. And I just think this is the beginning of a drumbeat that we're

00:05:55.680 | going to hear for a decade. He said, just at the very beginning of companies really starting to

00:06:02.240 | think about ways they can improve their business. And it's not the sexy chat GPT stuff. This is the

00:06:08.480 | real pedestrian stuff by using AI to basically translate documents into a hundred different

00:06:15.680 | countries. One of my observations and a fun thing to come out of that walk. But there are a few

00:06:22.160 | topics I want to hit on today, including the latest on AI and GPU scaling. And I want to hit on the

00:06:28.560 | topic of how we're thinking about the markets heading into the election. But I want to start

00:06:33.520 | with a topic that I know is near and dear to your heart, which is the state of venture capital.

00:06:37.920 | Yeah.

00:06:38.320 | Okay. So my partner, Jam and Ball, who we're going to bring in here in just a minute,

00:06:41.760 | wrote this week on his blog that VC funding sizes have ballooned, that the incentives for VCs have

00:06:51.200 | shifted perhaps a bit, maybe now focused a little bit more on this cadence of raising and deploying

00:06:57.120 | every couple of years, as much money as possible, which may create some incentive alignment problems

00:07:03.760 | with founders. The key question he told founders they need to ask is, are you partnering with a

00:07:09.280 | 2% or a 20% venture firm, which I thought was clever. So you then retweeted his post and said,

00:07:17.120 | it's a must read for all LPs and money managers and called it the single most important issue in

00:07:23.040 | the VC landscape today. So first welcome Jammin. You know, I thought maybe just start with you,

00:07:28.960 | summarize for us the blog post and kind of maybe why you wrote it.

00:07:35.360 | Yeah. Well, so I write this blog Clouded Judgment and it has many goals. One goal

00:07:40.800 | is to increase the transparency to founders of what goes on in the venture markets. The

00:07:46.880 | venture markets have changed a lot over the last 10, 15 years. I think Bill, it was maybe you or

00:07:50.880 | someone else, many people have said this, right? The venture markets have transitioned from

00:07:55.840 | a high margin cottage industry to a institutionalized lower margin industry.

00:08:01.120 | And that has a lot of implications, implications for LPs, implications for GPs

00:08:05.680 | and implications for founders. This post focused more on the founder GP leg of the stool.

00:08:13.360 | And I wanted to focus on that leg because I do think that the implications of this transition

00:08:19.520 | to an institutionalized asset class, it changes the game for founders that I don't think they're

00:08:24.720 | fully aware of what, of all the implications. And I had a lot of, I've had a lot of conversation

00:08:30.240 | with founders over the last few months that really highlighted that to me.

00:08:33.280 | So before diving into how have the incentives changed and how are they diverging a bit more

00:08:38.960 | amongst founders and GPs, you know, first we'll just describe what the incentives are for the

00:08:45.040 | audience, just to make it quite clear. So in venture funds, I'm sure folks have heard of the

00:08:49.200 | two and 20 model that is a 2% management fee and 20% carry. There's a guaranteed portion of

00:08:56.960 | funds comp, which is the management fee. It's a percent of the fund size that you'll collect

00:09:01.920 | every year. And then there's the carry, which is the funds share of profits. So take a hundred

00:09:07.920 | million dollar fund that you 3X, you'll take 2 million a year in management fees. You generate

00:09:13.760 | 200 million of profits, which is 40 million of carry. 10, 15 years ago, the funds were smaller.

00:09:20.800 | And as a GP, as an investor, you'll make money on that guaranteed portion, but not necessarily,

00:09:26.800 | right, as I call it, get rich money. Bill, what were you making when you

00:09:29.680 | started per year when you started in the business? No, but I would say on average,

00:09:34.960 | people are making a couple hundred thousand bucks a year working in the venture business.

00:09:39.680 | Nobody was getting rich on that management fee. Keep going, Jeremy.

00:09:42.800 | That's right. So if you wanted to get rich, you had to maximize your carry. The way to maximize

00:09:47.760 | your carry was to maximize the value of the underlying investments that you made, which

00:09:52.880 | resulted in a path of getting rich for founders, which also resulted in getting rich for investors,

00:09:58.720 | right? The incentives are aligned. Let's fast forward to today. Funds are much bigger. That

00:10:04.240 | $400 million fund is now 4 billion. That guaranteed portion of comp, the 2%, the management fee,

00:10:10.400 | you can now get rich on. And the reality is, is a path exists today-

00:10:15.360 | In that case, roughly 80 million a year. That's right.

00:10:18.640 | For a $4 billion fund. That's right. And so they're now-

00:10:21.680 | And you could imagine the managing partner taking a quarter of that, a third of that.

00:10:27.040 | 10, 15 million bucks a year. Yeah.

00:10:30.400 | Right. And so now there's a path where investors, GPs get rich, where the outcome of the founders

00:10:36.560 | and their companies is irrelevant. Not to say they are aligned in opposite directions,

00:10:41.200 | but they're no longer aligned. And now, as a rational actor, you could say, "Why not maximize

00:10:48.240 | the guaranteed portion of the comp, the 2% versus caring about the 20%?" And that leads to some

00:10:53.440 | pervasive incentives where funds incentives are more about deploying dollars quickly as opposed

00:11:01.280 | to maximizing the value of those deployed dollars. There's a lot of downstream implications of that.

00:11:07.520 | It leads to companies, early companies that show early traction, get flooded with offers and

00:11:13.440 | investment dollars. And we can talk about why overcapitalizing and overvaluing your business

00:11:18.480 | actually can pose a risk. I think one of the reasons I wanted to write this post was I thought

00:11:23.440 | we would have learned a lot of lessons coming out of 2021 that it's very clear we haven't,

00:11:27.760 | or we just choose to ignore because the incentives are no longer there.

00:11:30.720 | But for founders, I think understanding and appreciating, are you partnering? Are you

00:11:36.400 | marrying a firm that is more focused and caring about maximizing the value of your company

00:11:42.160 | together? Or are you partnering with a firm that just wants to deploy dollars as quickly as

00:11:47.360 | possible? And there's lots of things we can double click on, but that was kind of the broad

00:11:51.120 | rationale for the post or the quick summary of it.

00:11:54.640 | - And by the way, I mean, rather than say they only care about the management fee,

00:11:58.880 | they may view the other one, there's a phrase in poker called a free roll. They may view it

00:12:04.000 | as a free roll. I'm getting rich no matter what. - It's a lottery ticket.

00:12:07.200 | - Right. - It's a call option.

00:12:09.200 | - But this other, it's a call option. And if this company happens to be the next Google,

00:12:15.680 | the next Meta, we've seen how those compound over 20 years and it'll work out.

00:12:20.800 | - Right. - It'll work out in that case,

00:12:22.480 | or maybe it'll work out for one of the 10 that I'm putting 400 million in and then I'll be okay.

00:12:27.200 | - Right. As investors, we get to make many bets. As founders, you have one.

00:12:30.560 | - Right, exactly. - And one of the things we were

00:12:33.200 | talking about just earlier today is for founders, there are implicit things you are signing up for

00:12:38.560 | by taking a round that is too much money at too high of a valuation. I think the preference stack

00:12:44.160 | and preference in general is something that is sometimes understood by founders in the founding

00:12:48.560 | community, but not always understood. There are companies that can be sold for $100, $200 million

00:12:54.240 | that results in life-changing outcomes for founders and early employees. If you raised $100

00:12:59.280 | million too early, that exit path is no longer on the table. You can't sell the business for $100

00:13:05.920 | million and make any money because that just returns the preference stack. As the call option

00:13:10.560 | investor, your call option didn't hit, but you get your money back.

00:13:13.840 | - Well, let's drill down. Let's split this into two sections because I think

00:13:18.400 | that this thing that's happening and the reason I retweeted it has implications for both founders

00:13:27.040 | and companies, but the asset class also. - That's right.

00:13:30.320 | - So, we started down this path. Let's do it first. What does it mean for a company to

00:13:36.880 | have maybe too much money crammed into it at too high a price?

00:13:41.120 | - Yeah. There's many deep implications of this. One thing that I've learned

00:13:47.360 | that I hold, that I will argue with anyone on is every founder and every company and every board

00:13:53.840 | will tell you, "If we raise money, we're not going to spend it. We're going to stay frugal."

00:13:57.200 | That never happens. - Yeah, never.

00:13:59.120 | - And instead of focusing on, you might ask a founder, "What are the three things that matter

00:14:02.960 | most this year that are critical for your success?" If you only focused on those three things,

00:14:07.680 | you might maximize your chance of success. When you have lots of money, you're going to do six

00:14:12.480 | things at once, which means you are diluting the three things that matter.

00:14:15.600 | - Let's stay on this for a long time. So, many, many, many great people have said that

00:14:23.280 | constraints drive creativity. I've heard famous stories about Steve Jobs where he felt his job

00:14:31.680 | was just to say, "I want it this thin. Go figure it out." And by saying that, he limited the space.

00:14:37.360 | "We're not going to build a phone." And then people can focus and be creative.

00:14:42.800 | You've constantly heard, "We're going to make a better decision because we're going to have

00:14:49.120 | constraints, and we're only going to be in two markets, not five." If you're in five,

00:14:55.360 | guess what happens when revenue starts to slow? You've built some of your revenue in these places

00:15:02.720 | where you don't have a strong competitive advantage. Those churn out. And so, you've

00:15:11.600 | grown revenue faster, but it's less sustainable revenue, is another thing that can happen.

00:15:17.760 | - Well, when I look at this, if we look at the poster child or children for what went wrong

00:15:26.320 | during peak Zerp, okay? So, if your incentive is to deploy so that you can raise again in two years,

00:15:33.600 | then you're going to want to get maximum dollars into the business. Well, there may be only so

00:15:38.960 | many primary dollars you can get into the business. And then we started to see this desire to buy

00:15:44.480 | secondary shares because I wanted to get more dollars into the business. I wanted to deploy

00:15:49.120 | more in the company, even though you were buying common shares that were much riskier. And then,

00:15:53.600 | oftentimes, Bill, something I know you love, the founders or founding team were taking big dollars

00:15:59.120 | off the table, oftentimes before product market fit. And I think, you know, Jamin and I have

00:16:05.120 | discussed that we came out of this 20 to 22 period, and we would have thought that we would

00:16:11.920 | have seen less of that, right? Because we have, as you've discussed, 1,400 companies, right? 90%

00:16:20.640 | of which are going to have to do a down round, IPO or otherwise, in order to get through the system.

00:16:25.840 | But somehow the echoes of that period have not reverberated that loud, right? The incentives

00:16:32.560 | to deploy much bigger rounds into these companies continue to persist. And so, you know,

00:16:39.760 | gets back to your point about scarcity. Not only are you putting so much money in the business that

00:16:45.200 | perhaps they're not being as focused as they would otherwise be, but you really question

00:16:49.520 | whether or not the incentives for the founder are there. If you're taking 10, 20, 50, 100 million

00:16:55.360 | dollars off the table before the company has achieved profitability, then do you really have

00:17:00.400 | the fire in the belly needed to get, you know, both parties? And without picking on anybody,

00:17:06.000 | there are numerous examples from previous waves of even the founders like not being in the building

00:17:12.640 | or not showing up. So that can definitely happen. But there's so much more that can happen. I mean,

00:17:17.920 | one thing that happens is you create a prisoner's dilemma with your competitors where they feel

00:17:25.920 | forced to raise the same amount of money. And now you have multiple overfunded companies in

00:17:31.680 | a single category. Right. Just taking shots. So it's not only too much money within the firm

00:17:37.920 | that's causing the firm to be less fit, less efficient, less, you know, inventive, but it's

00:17:44.480 | also creating this dynamic where you used to talk about, you know, capital is a weapon of economic

00:17:50.080 | destruction. You have excess capital causing excess competition. So the natural market winner

00:17:55.280 | does not emerge as fast as they otherwise would. Well, maybe to level set the conversation.

00:18:01.120 | Can I mention one more thing? Sure, sure. That I think becomes a reality. And we've talked about

00:18:05.920 | this before, but I find far too often that founders think about raising money at a particular

00:18:14.240 | valuation as if they've earned an award. They've won a prize. They've proven their company is worth

00:18:21.600 | this thing. And they can then show the world the trophy and say, we've achieved X. And I always

00:18:28.080 | try and remind them that valuations represent discounted future expectations. And so you may

00:18:36.400 | feel like you've won a prize, but you've really increased everyone's expectation for what this

00:18:42.800 | company can achieve. And in order to raise that up round from here, it's way harder than it would

00:18:50.240 | have been otherwise. Right. And perhaps way more dilutive. I mean, the reality is even if you're

00:18:55.680 | getting what appears to be a high valuation, if you truly think you're one of those seminal

00:19:02.320 | businesses, you look at the ownership today that Zuckerberg has or Bezos had, because they raised

00:19:09.840 | so little capital in those businesses. And they compounded over a much longer period of time in

00:19:15.920 | the public market. And let me drive home this point, I think, in a very, very stark way, which

00:19:21.520 | is we're borrowing a number from the Code 2 presentation, but they said there were, what,

00:19:25.600 | 1,400 unicorns that are still private from what I like to call the pre-LLM period.

00:19:32.000 | What percentage of those could raise enough round right now? What is the number?

00:19:37.440 | I think it's less than 10%. It might be less than 5%.

00:19:40.320 | Today, yes. So that means there's over a thousand private

00:19:45.040 | unicorns that are somewhat stuck right now and could not raise enough round. So this risk I'm

00:19:51.360 | talking about, about putting too high a valuation on your company, we've just had this experience

00:19:56.880 | in this massive fact pattern. But you're right, we're still doing it. AI came along,

00:20:02.400 | and it has so much potential and all the things you're excited about that brought the money in.

00:20:09.200 | And the money's back in. And I'm seeing activity that is at least no different than what we saw

00:20:18.400 | in the past, and maybe even more frothy. Let's come back to that. But I do want to

00:20:24.400 | level set to the total amount of VC that's actually being raised and deployed. And this

00:20:30.720 | first slide we have here just shows that we peaked in Zurb in 2021 at over 700 billion deployed.

00:20:39.600 | We've come back down to about 300 billion deployed, which is still meaningfully above where

00:20:44.400 | we were in 2014, 2015, but we're back to about 2017, 2018 levels. But it kind of hides a couple

00:20:53.920 | of things. Number one, the number of first-time funds raising a second-time fund has plummeted.

00:20:59.840 | The number of first-time funds have plummeted. And so really what's happening is you have the

00:21:04.560 | consolidation among a few big platforms. Just this week, we saw that General Catalyst has raised

00:21:12.080 | $8 billion in new capital. Now, of course, I love my friends at General Catalyst, but

00:21:16.480 | just 24 months ago, they raised $4.5 billion. So that's $13 billion raised in that 24-month period.

00:21:25.120 | Similarly, Lightspeed recently announced that they raised $7 billion after raising about $6.5

00:21:32.320 | billion, so another $13 or $14 billion over kind of that 28, 30-month period of time.

00:21:38.880 | Clearly, LPs are raising their hands and saying, "We don't mind these larger funds. The capital is

00:21:46.160 | available for them." What do you think that they're concluding? If you believe that this is

00:21:54.320 | going to fundamentally reduce the returns of the entire asset class, then what must be going

00:21:59.440 | through the minds of those LPs to want to back those much larger businesses?

00:22:04.720 | That raises so many different questions. One thing I would highlight is that when,

00:22:08.720 | you know, my whole career in the venture industry and prior, because I read about the history before

00:22:15.520 | I joined, the industry was inherently cyclical and there would be these boom-bust periods.

00:22:21.280 | And they're very long waves, partially because of the construct of the agreement between the

00:22:28.080 | GPs and LPs, which used to be a 10-year thing, and now it's like a 15-, 16-year thing. And so,

00:22:35.120 | the window for which you would evaluate whether someone can accurately or successfully deploy

00:22:44.400 | a $4 billion fund might be 20 years. And we just started raising them that big.

00:22:51.920 | So, no one knows. Like, no one knows. They're raising money off of paper marks, you know,

00:22:59.360 | from this period. And it's just, the cycle's forever. It's beyond the lifetime of most

00:23:06.000 | GPs' career left to be done, which gets back to the 2%, 20%.

00:23:12.800 | And I think it kind of depends. I think one way to, one argument or one lens to look through when

00:23:19.120 | trying to predict will it work is everyone will look at venture funds and there's kind of this

00:23:23.680 | view and belief that the bigger you get, the later stage you invest, you can more evenly

00:23:30.400 | distribute your returns in a fund to achieve a top quartile return. That, I would say,

00:23:36.720 | I strongly disagree with. All of the numbers that we've seen suggest that to have a top quartile

00:23:42.880 | fund, you need to have outlier power law outcomes within the fund. You won't evenly distribute them.

00:23:49.120 | So, I think one way to look at this is to say, and this is the way we talk internally,

00:23:53.200 | what does it take to have a power law outcome at a certain stage? At the seed stage, you might have

00:23:57.360 | this many companies that could get you a power law outcome and return your fund. Series A, Series B,

00:24:02.560 | Series C. And it's just, it's a downward funnel of the number of companies that could give you

00:24:06.880 | power law outcomes. I think one of the challenges is if you're investing just in growth stage,

00:24:13.040 | there's only so many companies you can invest in that can give you power law outcomes.

00:24:18.480 | And so, you are almost inherently widening the aperture to do more

00:24:22.320 | than what could be a power law outcome, the larger and larger you get.

00:24:25.680 | I agree with that.

00:24:26.480 | So, I think that the burden of proof is on the picking of the large funds to say,

00:24:33.120 | we will pick and hit those power law outcomes, but then have a much higher hit rate on the rest.

00:24:39.760 | And so, it won't be this big power law outcome. It will have the power law and then be more

00:24:45.120 | evenly distributed in the long tail, which it can happen. It's just really hard, especially

00:24:50.960 | when your incentives are just to deploy to raise the next fund. It's, are you giving up or saying,

00:24:56.400 | "Hey, I don't care. Picking is too hard. I don't need to do it." In which case,

00:25:00.160 | maybe you get the power law, maybe you don't, but you're just naturally going to capture more

00:25:04.080 | of the long tail, which I think ultimately will dilute a lot of those returns.

00:25:08.800 | And so, we'll see how it plays out. That's the fun part of it.

00:25:11.680 | Or not, or not. I mean, this is kind of Bill's point. So, if you're on a run rate where you're

00:25:16.560 | raising 5 billion bucks every two years, okay? So, give or take, that's $50 billion over a 10

00:25:23.200 | year period of time. A 2% on 50 billion, it's a billion. Okay? So, now it just-

00:25:30.080 | A billion a year. Let's just be clear.

00:25:33.520 | Right, right, right. So, underscore that for everybody, Bill, right? Because

00:25:38.320 | these funds layer on top of one another. So, you're getting paid on multiple funds.

00:25:43.280 | Correct.

00:25:43.760 | And that's the guaranteed portion. Now, these firms are much larger firms,

00:25:48.560 | much more institutionalized. To Jamin's point, they're going to have to spread their bets over

00:25:53.760 | a much, much wider field. So, the potential that you're going to have enough ownership

00:25:59.360 | in a power law business to earn venture-like returns is much lower. But I think you also

00:26:06.080 | made the point that maybe what you're doing is your mortality rate goes down.

00:26:10.400 | And so, you're giving more confidence. But Bill's point is, we don't know.

00:26:15.040 | You remember that.

00:26:15.600 | The experiment's never really been run in venture at this scale. And we would have thought that

00:26:21.200 | coming out of the Vision Fund experiment, right? We had all these comments during Vision Fund

00:26:26.800 | that we wouldn't see, you know, kind of this cadence of these size of funds.

00:26:31.040 | And then if you want to take some targeted big bets, say in an open AI or something like that,

00:26:38.240 | there may be only one or two of those that come around every couple of years.

00:26:42.640 | Every year, maybe every 10 years.

00:26:44.160 | And so, you place a lot of wood behind a couple big bets. But then your traditional venture fund

00:26:51.360 | is smaller such that it can generate those historical venture-like returns.

00:26:56.880 | But I haven't seen any math that I find compelling how you can get to a 4 to 5X fund

00:27:02.960 | on, right, that $5 billion pool of capital.

00:27:05.680 | Well, in fact, yeah. I mean, in 2011, there was this famous report by the Kauffman Foundation

00:27:11.280 | that argued that billion-dollar-plus funds had never had good returns.

00:27:17.280 | And of course, after that, everyone raised billion-dollar funds.

00:27:20.240 | But I have this story I want to share with you that relates to this in a funny way.

00:27:24.640 | I was at one point in time invited into the office of this PE firm. I didn't know PE very well.

00:27:31.200 | But they had backed DoubleClick and Google bought it for $3.1 billion.

00:27:36.560 | And this particular firm made-- I forget, they had a lot of it.

00:27:41.200 | So they made like a billion or a billion two.

00:27:43.200 | And I said to him, I always say, "Congratulations, man. That's incredible."

00:27:47.280 | He goes, "Well, it's in a $4 billion fund."

00:27:50.240 | Wow, wow.

00:27:51.200 | You know?

00:27:52.000 | So you have an incredible outcome.

00:27:55.200 | Right, right.

00:27:56.080 | But the denominator is so big that it's just hard to claw it back.

00:28:03.120 | It's hard to get back to that number.

00:28:04.720 | Which is why you can't get to the 4 or 5X because you have a power law outcome,

00:28:09.680 | and it returns 25% or a third of the fund.

00:28:12.960 | Correct.

00:28:13.440 | Right?

00:28:14.000 | What we've witnessed is kind of this bill coming out of the '22 period,

00:28:19.120 | these two paths effectively, you know, that the firms have followed.

00:28:23.920 | We've seen some firms downsize.

00:28:26.240 | Founders Fund, Altimeter, a few others have downsized the fund size.

00:28:30.000 | But we've seen other funds that just said,

00:28:32.320 | you know, we're going to consolidate and we're going to get way bigger.

00:28:35.360 | And so I do think that, you know, if we look at a little bit of returns math on this.

00:28:41.680 | So we have a few slides.

00:28:44.000 | But, you know, I love, you know, Fred Wilson's rule of thumb.

00:28:47.360 | You know, a third of companies fail, a third basically underperform,

00:28:51.760 | and then a third gets you your 5 to 10X returns.

00:28:54.400 | And recent breaks it down even more, which, you know,

00:28:57.760 | says it all comes down to the 10% that produce, you know, kind of that 10X fund.

00:29:03.200 | And if you look at this stepstone data, you know, it actually shows that, you know,

00:29:08.320 | basically what differentiates those top tier funds.

00:29:12.560 | And by top tier, what I mean is that the top 5% of Decile funds,

00:29:19.680 | they're earning about a 4.5X, you know, cash on cash return in their funds.

00:29:25.120 | You know, you go back and you can look across all of these vintages over the last 20 years.

00:29:29.520 | Now, I think there's a question whether or not any of these mega funds on these

00:29:34.400 | $4 or $5 billion funds are going to be able to, you know, generate those returns.

00:29:38.080 | And maybe the reality is that these new LPs, pension funds,

00:29:42.800 | sovereign wealth funds, and others, look at these as strategically important.

00:29:47.440 | And they, you know, they're less concerned about venture-like returns,

00:29:52.160 | and they want to protect the downside more.

00:29:54.240 | And so maybe you can sell the fact that you're going to have less mortality, right?

00:29:59.680 | And you're going to have less downside risk to the funds.

00:30:02.560 | But I mean, I think that that is really where the rubber meets the road.

00:30:06.560 | You know, when we look back at this, you know, this analysis, whether or not these funds are

00:30:11.200 | going to be able to generate returns that are equivalent to the top Decile of historical.

00:30:17.520 | One of the things we don't know that it would be really hard to gather data and prove,

00:30:22.640 | but this zombie unicorn class, pre-LLM zombie unicorn class, you know,

00:30:29.600 | had it not been flooded with money.

00:30:32.320 | And, you know, there's a notion in science called the observer effect, you know,

00:30:37.840 | where if, you know, VCs are too big and like,

00:30:42.640 | do you actually impact the results of the game on the field?

00:30:46.240 | And, you know, had that not happened, and these companies grew up on a normal way,

00:30:52.720 | you know, in a more organic, you know, growth path, would they have had more liquidity?

00:30:58.240 | Would they have had more investor returns?

00:31:00.560 | Would they have been more acquirable without these high marks on their head?

00:31:05.680 | And so did you take out, you know, you're so, we're so focused on the outlier,

00:31:12.160 | you know, breakout Google meta types.

00:31:14.640 | Did you ruin-

00:31:15.520 | - The two, three, four X returns?

00:31:17.680 | - Yes.

00:31:18.080 | - Yeah.

00:31:18.320 | - Did those go away?

00:31:19.360 | - Yeah.

00:31:20.080 | I think you totally removed it as an outcome path, right?

00:31:24.000 | Which is to say, when you are investing in these companies at high prices,

00:31:27.600 | the only thing as an investor you care about is, are you an outlier outcome?

00:31:30.880 | Are you a 10X plus?

00:31:32.000 | - Right.

00:31:32.560 | - As a founder, you might say, I don't need that.

00:31:36.320 | As employees, you might say, I don't need that.

00:31:39.040 | And if I was sitting on a $600 million valuation, not 6 billion,

00:31:42.640 | it's going to be much different to go and sell the business for 2 billion

00:31:46.960 | when the prep stack isn't one and a half, right?

00:31:48.800 | So I think we've just, we removed a thick middle of an outcomes

00:31:54.800 | that would be relevant to everyone outside of the investors,

00:31:58.240 | where now the only outcome that's relevant to anyone is, are you a 10X or are you a zero?

00:32:03.920 | And you made it a binary outcome-

00:32:05.360 | - I agree.

00:32:05.840 | - For all constituents.

00:32:07.120 | - And you force everyone into that game.

00:32:09.120 | - Right.

00:32:09.680 | - Obviously, all three of us are in an industry.

00:32:14.880 | We believe it's the engine, you know, that provides the fuel for,

00:32:18.960 | you know, incredible founders to go innovate the future.

00:32:22.880 | And it's really just a question of whether or not we're overcapitalizing the business,

00:32:28.400 | whether funds are getting too big, but what recommendations, Bill,

00:32:32.160 | might you have or Jamin, for founders or funders, right?

00:32:38.640 | Jamin, what I hear you saying is founders just be aware, right?

00:32:44.400 | Ask the questions.

00:32:45.440 | - Then know what you're signing up for.

00:32:45.920 | - Think through the people you're putting in your capital structure

00:32:49.520 | and like what their incentives are.

00:32:51.840 | Are they sitting on 20 different boards?

00:32:54.240 | Are they clearly in the deployment mode rather than the investment mode?

00:32:59.280 | And if they're in the deployment mode, just know that.

00:33:01.680 | - And know what you're implicitly signing up for.

00:33:04.880 | What bets are you making, not just on your behalf,

00:33:07.440 | but on your entire employee base's behalf, where do they want the binary bet?

00:33:12.720 | Binary bets are okay for investors.

00:33:14.720 | I don't think early employees or founders necessarily want that.

00:33:18.720 | But I'd say that, you know, know what implicit bets you're making.

00:33:21.920 | Buy the rounds that you're raising.

00:33:23.760 | - Yeah, I mean, to that point, like this is super simple math,

00:33:30.960 | but if you raise 400 million or let's just say 100 million,

00:33:35.600 | even if you're being conservative, what are you going to spend that over?

00:33:40.160 | 36 months, you know, something like that.

00:33:42.880 | You're still signing up for a three, four million a month burn rate.

00:33:48.240 | And there's no way that doesn't represent risk in some way.

00:33:53.040 | And the bigger you take that number,

00:33:57.040 | I used to always think about venture capital as,

00:34:00.080 | you know, you invest in a company and you grow that burn up until a point.

00:34:06.000 | And then the job is to converge back against it and go to profitability.

00:34:10.240 | And with all this money, I mean, we're taking that point up to 10 million a month,

00:34:17.520 | 20 million a month.

00:34:18.560 | And this gets into who can grow out of that.

00:34:22.240 | - Right.

00:34:22.640 | - And it ties into did you destroy that middle?

00:34:26.160 | Because some of them just can't ever get there.

00:34:29.440 | - And you then rely on the beneficence of the capital markets

00:34:32.960 | to keep feeding that machine.

00:34:34.480 | And sometimes it runs out and then you're in a massive course correction.

00:34:39.680 | - But my general response is unfortunately more sanguine,

00:34:46.640 | which is I think these systems are emergent.

00:34:50.560 | I don't think any one person made a willful decision to force this reality.

00:34:56.800 | - Correct.

00:34:57.200 | - I think it emerged out of a combination of what was happening with interest rates

00:35:02.640 | and Zerp and then the rise of AI and maybe the globalization of fundraising

00:35:09.600 | that brought sovereign wealth in.

00:35:11.520 | And it all kind of combined and created this mix that we're living with.

00:35:16.880 | And for the most part, I think we're all actors stuck in the game.

00:35:21.360 | - Right, right.

00:35:22.160 | - You know, I'm not sure you can escape it.

00:35:24.320 | - You know, a couple of observations.

00:35:26.720 | - Is that the positive take on it?

00:35:28.160 | - No, I know he wanted one.

00:35:30.400 | I apologize.

00:35:31.360 | - Well, I mean, I'll give you my spin on that.

00:35:35.920 | Number one is, you know, I do think that pensions and sovereign wealth funds

00:35:41.040 | have, you know, stepped in to replace some of the money

00:35:45.920 | that came out of the endowment ecosystem, right?

00:35:48.640 | Because endowments, you know, I think about the legends of that business

00:35:52.880 | who were allocating those dollars, you know,

00:35:55.840 | the Phil Rotners of the world or the Swensons of the world, right?

00:35:59.920 | Like they would phone up everybody on Sand Hill Road

00:36:02.800 | and they would say, "You're getting over your skis.

00:36:04.480 | You need to back it off a little bit," right?

00:36:06.480 | That industry is long gone.

00:36:08.160 | But I would also say, I just, you know, I saw news this morning,

00:36:12.400 | you know, at FII, which is doing, you know,

00:36:16.240 | like I think an incredible convening of investors and leaders

00:36:20.240 | from around the world.

00:36:21.280 | I think they're doing incredible stuff, you know,

00:36:24.000 | in the GCC and Saudi in particular.

00:36:26.560 | But they said, "We're going to deploy more of PIFS dollars

00:36:30.320 | on investments inside of Saudi Arabia than outside of Saudi Arabia," right?

00:36:34.800 | So I think there was like, and I've heard this as well from, you know, the UAE.

00:36:39.760 | They're going to consolidate the number of GPs.

00:36:42.240 | And so I think there is increasing focus coming out of sovereigns as well.

00:36:46.000 | I hear the same.

00:36:46.720 | I was just down in Texas talking to the Texas pensions,

00:36:49.760 | the same thing happening there.

00:36:51.200 | But when they consolidate, that probably means

00:36:54.400 | more dollars into fewer, right, platforms.

00:36:57.920 | So maybe it's starting to crack, but I don't know.

00:37:00.320 | And so I don't know, you know, we've seen from that slide that I showed,

00:37:04.160 | the total amount of funding has definitely come down

00:37:07.600 | by about 30 or 40% over the course of the last couple of years.

00:37:11.040 | So that's one half of it.

00:37:12.160 | The other half is there is no doubt that the outcomes

00:37:15.600 | that we're now talking about are much bigger

00:37:17.840 | than the outcomes we were talking about 10 or 20 years ago.

00:37:20.880 | So it would make sense to me, if you looked at a trend line for the industry,

00:37:25.040 | the industry and the number of dollars deployed

00:37:27.600 | and the size of funds should be much bigger today

00:37:30.640 | because companies are scaling much faster today

00:37:33.120 | and the outcomes are much faster.

00:37:34.960 | OpenAI got to, you know, $4.5 billion of revenue

00:37:39.200 | in a fraction of the time it took Google and Meta.

00:37:42.160 | And, you know, and so like that to me supports more dollars, larger funds.

00:37:48.160 | But I do think the law of economic gravity prevails.

00:37:51.200 | There's a certain fund size I think it can deploy

00:37:54.640 | if you're aiming for traditional venture like returns,

00:37:58.160 | then I don't think there are 100 growth companies, right,

00:38:02.080 | that you can go put in a fund of $5 or $10 billion,

00:38:05.440 | equally weight them and get a 4 to 5x over any reasonable period of time.

00:38:09.920 | And we can, I mean, let's move into that.

00:38:12.160 | I mean, I think the enthusiasm around AI is part of what led to that reality.

00:38:18.560 | And if it didn't feel spectacular, it wouldn't have happened, right?

00:38:24.000 | Right.

00:38:24.400 | So it's tautological that those two things contribute to one another.

00:38:28.720 | Well, it's a, you know, so one of the topics we've been talking about all year is,

00:38:32.400 | you know, is AI and GPU scaling.

00:38:36.000 | Will these AI models continue to scale?

00:38:37.840 | Will GPUs continue to scale?

00:38:40.640 | There was a lot of question whether or not NVIDIA would grow in 2025.

00:38:44.800 | But just this week, we saw a few announcements.

00:38:46.880 | First, we saw out of XAI, the Colossus cluster in Memphis,

00:38:51.120 | which I discussed at length with Jensen.

00:38:54.000 | They announced this week that they're scaling.

00:38:56.160 | Elon said, we're now going to go from 100,000 GPUs to 200,000 GPUs in that facility.

00:39:04.560 | Did you see the video that Supermicro put out on this?

00:39:07.920 | No, you told me about it though.

00:39:09.280 | Right.

00:39:09.520 | So it's-

00:39:11.120 | Describe it.

00:39:11.760 | Or you put a link in.

00:39:14.000 | It's incredible to watch what they built.

00:39:16.480 | Yeah.

00:39:17.200 | And if I think about for the last decade, data center engineering and construction

00:39:23.440 | has been in the background, right?

00:39:26.320 | Like, you know, Meta had breakthroughs.

00:39:28.560 | Google had breakthroughs.

00:39:29.840 | Like we'd talk about it, but it wasn't something that people were posting videos on.

00:39:34.880 | I think there's some Equinox shareholders that might disagree with you.

00:39:38.960 | But I would say we generally took it for granted.

00:39:42.560 | But now it's clearly becoming a much more important source of competitive advantage.

00:39:48.160 | And I would just say, you know, it's super impressive what they pulled off in Memphis.

00:39:54.160 | It all gets to this question of, you know, you were asking at the beginning of the year,

00:39:58.800 | will we need or will we get to clusters of 200,000?

00:40:02.720 | Now those questions are clearly off the table.

00:40:05.520 | And so we have all these mega caps reporting this week.

00:40:09.840 | It's expected that CapEx will continue to, you know, continue on at this $250 billion run rate.

00:40:16.480 | Google came in tonight at $13 billion, just above the $12.7 billion that was expected.

00:40:22.720 | So they revised that higher a bit.

00:40:25.920 | Bill, what do you just make out of, you know, as we sit here toward the end of 2024,

00:40:31.440 | do you expect that this is going to, you know, continue at this pace?

00:40:38.560 | Is this a sort of, you know, I've heard people describe it as a Pascal's bet sort of situation?

00:40:45.120 | You know, how are you thinking about, you know, where this all leads?

00:40:49.440 | I'd love to drill down on that, but let's come to it in a bit.

00:40:53.680 | Look, it's something we've never seen before.

00:40:57.040 | You know, if I think back to, and you and I were discussing this,

00:41:01.840 | but if I think back to the breakout companies of the previous generation of VC backed companies,

00:41:09.600 | none of them had massive CapEx.

00:41:12.640 | The biggest one was Amazon, actually, with their distribution facilities.

00:41:18.960 | And it caused a lot of skepticism.

00:41:22.320 | And they raised a lot of debt, and they did things other people hadn't done before.

00:41:28.000 | And, you know, I've discussed it.

00:41:29.440 | I don't think we've ever seen kind of a CapEx race before.

00:41:33.760 | And it's pretty mind boggling that these are being built at this scale.

00:41:41.360 | It's interesting, you know, to try and count how many people are going to want to do it.

00:41:47.200 | Like, you know, the hyperscares obviously are reselling this.

00:41:50.560 | So, they're like an arms merchant to everyone else.

00:41:54.400 | You know, in Tesla's case, and extended AI's case, and Meta's case,

00:42:00.240 | they're consuming this, you know, for their own good.

00:42:04.400 | And, you know, I'd throw this back at you.

00:42:08.400 | Do you think there'll eventually be 50 of those that are Tesla-like,

00:42:13.840 | that will want to build these out for themselves?

00:42:17.040 | No, I think there are going to be a handful of companies, you know, like take Mag7.

00:42:20.960 | Because the scale is too high.

00:42:22.880 | Yeah, think about it this way.

00:42:25.360 | You know, if you look at Google, in the quarter,

00:42:28.080 | GCP has accelerated its revenues by up to plus 35% year on year, okay?

00:42:34.720 | So, just to put that in perspective,

00:42:36.480 | they added about two and a half billion of new ARR in the quarter.

00:42:39.920 | That's like adding a Databricks in the quarter, right?

00:42:44.320 | So, it's only companies of that scale that can afford to spend $50 billion a year.

00:42:50.480 | And, you know, I do think there are increasing advantages to scale,

00:42:54.240 | both when it comes to data and compute.

00:42:56.720 | And so, you know, I found this Pascal's bet to be a good framing of this,

00:43:02.000 | which is, if I were running any one of those companies,

00:43:05.120 | it would now be sufficiently clear to me.

00:43:07.360 | Explain, explain this, Pascal.

00:43:09.120 | You know, so Pascal's this French philosopher.

00:43:11.120 | He says, I don't know whether God exists or not,

00:43:13.440 | but if he does exist and I don't believe,

00:43:16.240 | then it's going to be a really bad outcome for me.

00:43:18.640 | And if he does exist, I do believe I have eternal life.

00:43:21.440 | So, if you believe that the upside of AI, right, is infinite, is eternal,

00:43:26.960 | right, is some really big outcome and all your competitors are doing it,

00:43:32.000 | what else are you going to invest your money in?

00:43:33.840 | We said here several weeks ago, you can buy back your shares,

00:43:38.480 | you can issue a dividend, or you can buy GPUs, right?

00:43:41.920 | Do you think any one of these founders or any of these companies-

00:43:44.800 | You can do licensed deals with companies you want to acquire.

00:43:47.680 | You know, so I think, but I think most of these people are in this game for this moment,

00:43:55.200 | but I think it raises a really interesting alternative question,

00:43:59.680 | which is, it's pretty clear to me now that there are very few new entrants

00:44:05.680 | that are going to be able to play in that game, right?

00:44:09.120 | You may be able to build an application that rides on those rails,

00:44:12.720 | but the idea that you're going to be able to build that level of compute

00:44:17.040 | or build a frontier model that's going to require that level of compute

00:44:21.360 | on an annual basis and compete with those companies,

00:44:24.160 | I think it's very, very difficult.

00:44:25.840 | Yeah, and another thing that I think the law of large numbers

00:44:29.440 | just starts to catch up as well.

00:44:31.280 | I think there's some CapEx spend from even a Mag7.

00:44:35.520 | I'd mentioned that maybe the CFO's voice goes up a little bit,

00:44:40.240 | but the investors will start asking as well.

00:44:43.680 | And there was another data point this week that may seem trivial,

00:44:47.280 | but there was something that just came out yesterday

00:44:50.960 | that OpenAI is talking to TSMC about building a chip

00:44:54.880 | and has put their foundry ambitions on hold.

00:44:57.520 | Right.

00:44:59.280 | I always thought the foundry ambitions were a little too ambitious,

00:45:02.560 | but in some ways, that's an odd recognition that,

00:45:05.680 | oh, well, that may have been too much money.

00:45:08.320 | And with the burn rate that's rumored that they have,

00:45:11.120 | like, oh, okay, maybe, you know.

00:45:13.360 | And so, there is some...

00:45:15.520 | One thing I would say about the Pascal thing

00:45:18.560 | is if that's truly what's on everyone's mind,

00:45:22.800 | that's exactly how you would go over the top.

00:45:25.680 | Right, for sure.

00:45:26.720 | The exact formula.

00:45:27.760 | First, I mean, listen, there is some non-zero probability.

00:45:31.440 | Just as Elon says, there's a non-zero probability

00:45:34.480 | that AI is going to end up being bad for humanity.

00:45:37.680 | There is a non-zero probability here

00:45:39.920 | that at least for some of the players,

00:45:42.400 | they don't see a return on that investment.

00:45:44.480 | Or that you get a supply-demand imbalance in the middle.

00:45:47.600 | Or you get a diluted return, right, that you all overspend.

00:45:51.600 | And had you just waited and spent slower,

00:45:55.280 | the cost of the technology would come down.

00:45:57.360 | It would cost you less over time.

00:45:58.400 | I mean, look, there are people that will argue that the,

00:46:02.720 | you know, we're going to get in here

00:46:05.360 | and the first pass is we're going to place

00:46:07.600 | a bunch of programmers.

00:46:08.960 | But then we're going to replace a bunch of paralegals.

00:46:11.440 | And then we're going to replace, you know,

00:46:13.600 | a bunch of, you know, imaging analysts.

00:46:17.920 | And then, you know, and the longer that list gets,

00:46:21.760 | and if you are successfully fine-tuning models

00:46:26.800 | specifically for those particular tasks,

00:46:29.840 | and that gets longer and longer and longer,

00:46:34.400 | that's big, I mean.

00:46:36.240 | So it leads me perfectly into this comment

00:46:38.560 | out of Masa Sun today, you know,

00:46:41.840 | that we couldn't not talk about.

00:46:43.920 | But at any rate, Masa at FII today said,

00:46:46.880 | "9 trillion of cumulative CapEx on 200 million GPUs

00:46:53.600 | is very reasonable."

00:46:55.280 | He said, "In fact, I think it may be too small."

00:46:58.240 | And he went on to say, you know,

00:47:00.400 | that if you took the most critical estimate,

00:47:04.960 | so of the value of AI.

00:47:08.000 | So he said the lowest estimate that he's seen

00:47:10.960 | on the value of AI is that it could do the, you know,

00:47:14.880 | what he calls artificial super intelligence,

00:47:17.200 | which is what he thinks we'll have

00:47:18.880 | if we spend $9 trillion,

00:47:20.160 | is that it would replace 5% of the global workforce.

00:47:25.600 | And it just so happens, if you do the math on that,

00:47:27.760 | 5% of the global workforce costs about $9 trillion a year.

00:47:31.200 | So he said, "Imagine you were a single entity,

00:47:33.920 | single enterprise, the global workforce was your workforce,

00:47:37.600 | and you could spend 9 trillion cumulatively

00:47:40.320 | over the next seven years,

00:47:42.000 | that you would then recoup in a single year

00:47:44.880 | through the efficiencies gained in that workforce,

00:47:46.800 | it would obviously be an NPV positive investment."

00:47:50.240 | Bill, is Masa's 9 trillion the top?

00:47:54.240 | Is it where all this, you know?

00:47:57.120 | I think the key math for that number

00:48:00.480 | is that it was bigger than any number anyone else has said.

00:48:03.760 | There you go, there you go.

00:48:05.120 | I do think though there's, I mean,

00:48:06.720 | many have made this argument, right?

00:48:08.160 | You look at IT budgets as a whole,

00:48:09.680 | about half of that generally goes to headcount and people,

00:48:12.560 | a smaller percentage of that goes to software,

00:48:14.720 | and even smaller percentage of that

00:48:16.240 | goes to new spend in the year.

00:48:18.400 | Most of the software spend is just renewal.

00:48:21.040 | You spent it last year, you're gonna spend it this year.

00:48:22.880 | A lot of people have written about this,

00:48:25.200 | but it's the great services to software rotation.

00:48:27.600 | Services markets are usually 10x bigger than software markets.

00:48:30.560 | If we look at how can spend attack IT budgets as a whole,

00:48:35.360 | it's not just let's replace this software spend

00:48:37.920 | with this software spend,

00:48:38.960 | it's let's replace these call center agents.

00:48:41.280 | Let's replace these insurance people

00:48:43.680 | filling out forms in the back office.

00:48:45.360 | It's let's replace a lot of this headcount spend.

00:48:47.520 | I mean, most of these firms replace,

00:48:49.280 | you know, employ hundreds and hundreds of thousands of people,

00:48:53.760 | and we're seeing those numbers.

00:48:55.120 | That's where, you know, I started,

00:48:56.560 | you know, the conversation about my walk with Michael.

00:49:00.960 | And, you know, again, this is looking over

00:49:04.320 | a 30 to 40 year period of time,

00:49:06.320 | never has he seen, or have I seen,

00:49:11.040 | the ability for companies to take out 10,000 people

00:49:15.120 | and actually see customer satisfaction scores increase, right?

00:49:19.840 | There's always a cost to letting people go.

00:49:23.120 | But in this case, you know, you hear Zuckerberg talk

00:49:26.240 | about take out 20,000 people.

00:49:28.560 | Flatter is faster, leaner is better.

00:49:30.640 | We're getting more done with less, right?

00:49:32.960 | When you look at span of control in companies, right?

00:49:36.320 | There are some companies, you know, that still have,

00:49:38.720 | and that's the number of employees per manager, right?

00:49:42.160 | So you can imagine that will widen, you know,

00:49:45.200 | because AI will allow you to, you know, have more people,

00:49:48.800 | you know, who you can effectively manage.

00:49:51.600 | - It'll be interesting, I do think that some

00:49:54.880 | of the fine tuning of the math is still to come.

00:49:58.960 | And I'll use the coding example.

00:50:01.280 | You know, when we started, you know,

00:50:03.280 | oh, we're gonna replace engineers.

00:50:05.600 | But if you talk to someone, a CIO type,

00:50:10.480 | and say, what kind of lift do you get

00:50:12.240 | from a co-pilot product?

00:50:14.080 | They're kind of generally in the 15 to 30% range,

00:50:17.280 | which isn't, you know, if you're replacing somebody,

00:50:20.160 | it's 100% or infinite percent.

00:50:23.360 | And so whether or not, and maybe customer service

00:50:28.560 | is different than coding in this case,

00:50:30.960 | where you can, you know, replace 100% or 90% of the people.

00:50:36.080 | But it'd be interesting to see as the analysis come out

00:50:39.440 | of these things, what those different numbers look like

00:50:42.800 | for each of these different verticals,

00:50:44.560 | so that you can get a better sense of that.

00:50:47.360 | But if you really can, you know,

00:50:51.360 | if Sierra, Brett Taylor's company is right,

00:50:54.160 | and you can replace 100% of your customer service agents

00:50:58.800 | with this technology and get a more satisfied customer,

00:51:02.320 | and if there are multiple verticals like that,

00:51:05.760 | then you're right, this could go on for a long, long time.

00:51:09.280 | - I'd say, you know, there's no doubt in my mind

00:51:13.760 | that there's some sloppy spending going on, right?

00:51:16.320 | Like, you can't add this magnitude of spending

00:51:20.160 | without there being a little bit of waste, okay?

00:51:23.200 | But it's also clear to me that the reward on the other side,

00:51:26.880 | there is nobody taking their foot off the accelerator.

00:51:30.240 | - Yeah. - Right?

00:51:31.040 | When I talk to CoreWeave, when I talk to Azure,

00:51:33.520 | when I talk to OpenAI, when I talk to Amazon, Google, et cetera,

00:51:37.120 | not one of them is talking about anything other than

00:51:41.360 | how do we find the facilities that can support a gigawatt

00:51:45.440 | or two gigawatts or three gigawatts of power

00:51:48.240 | to power up the clusters that they wanna invest in.

00:51:51.680 | And remember, these are three, four, five-year-long investments.

00:51:55.520 | - Another thing that came out this morning

00:51:57.760 | that I was kind of waiting on as a proof point,

00:52:00.160 | 'cause it didn't make sense to me that it hadn't happened yet,

00:52:02.640 | TSMC said they're gonna do across the board 20% price increase.

00:52:08.240 | And that's, I was like, this should be happening.

00:52:10.720 | - Right. - And so seeing that

00:52:11.760 | is also indicative of this, right?

00:52:14.000 | - Indeed, indeed.

00:52:15.760 | And maybe we could close out here

00:52:19.120 | by talking about the markets a little bit.

00:52:21.840 | We always talk about the markets at the close.

00:52:23.760 | But one of the questions I keep getting asked, Bill,

00:52:26.720 | is whether or not this election is priced in.

00:52:29.360 | I mean, in a lot of our text threads we have with our friends,

00:52:34.240 | everybody's trying to figure this out,

00:52:36.240 | all eyes on the election, which is here in seven or eight days.

00:52:40.640 | And the question is like,

00:52:41.600 | a lot of people looking at the prediction markets

00:52:44.880 | saying increased probability that Trump wins the election.

00:52:48.320 | So they're wondering, is there still time for me

00:52:50.720 | to adjust my portfolio, right, ahead of this?

00:52:53.680 | And so you heard Druckenmiller talk about this this week.

00:52:57.760 | Paul Tudor Jones talked about it.

00:52:59.600 | I think Ray Dalio all said the market is fully banking

00:53:04.000 | on a Trump victory at this point in time.

00:53:06.320 | I probably wouldn't put the market is at 100%,

00:53:10.240 | but, you know, there's 70 to 80% baked into the cake.

00:53:13.280 | You can see that a lot of stocks have moved a lot.

00:53:15.920 | And I peeled this back a little bit.

00:53:17.520 | - This is unbelievable.

00:53:18.400 | I'd love to see what's in this democratic spread trade bucket.

00:53:21.600 | It's down 20%, 30 days.

00:53:24.320 | - Well, I'll tell you, you know, I looked at it,

00:53:26.400 | you know, some companies in the solar complex,

00:53:28.880 | like Enphase, et cetera, down 25%.

00:53:30.960 | Companies- - In 30 days.

00:53:33.040 | - In 30 days.

00:53:34.400 | You know, companies like in the healthcare trade,

00:53:37.520 | Humana and others down significantly

00:53:41.360 | because obviously they provide Medicare Advantage

00:53:44.080 | and other products that will get hurt,

00:53:46.720 | that are expected to get hurt under a Trump presidency.

00:53:50.160 | And so if you look at the policy consequences

00:53:53.680 | of these two candidates, they're very stark, right?

00:53:57.200 | And I would argue that most- - He's also talked about

00:53:59.520 | an MFN clause for drug prices inside versus outside of US.

00:54:03.680 | - So at this point, would I sit here

00:54:05.840 | and I try to fine tune my portfolio for the election?

00:54:08.320 | Probably not.

00:54:08.880 | You know, I think that markets have been quite efficient

00:54:12.720 | on this over the last 30 days.

00:54:15.360 | And my own sense is that the bigger surprise, right?

00:54:20.240 | If you thought that it wasn't 60 or 70% likelihood

00:54:23.920 | that Trump was going to win,

00:54:25.120 | let's say you thought it was actually even odds

00:54:27.760 | and you believe the polls that are out there, right?

00:54:30.880 | That whole trade there will unwind if Harris wins.

00:54:34.240 | So there would be a massive move in the other direction,

00:54:37.760 | right, if Harris were to win.

00:54:39.680 | I put both of them in the too hard bucket,

00:54:41.760 | but I thought I'd put them out there

00:54:43.200 | because we get certainly a lot of chatter about it.

00:54:46.880 | - All right, we're in the earning season, Brad.

00:54:50.240 | So, and we had a couple,

00:54:53.120 | like right before we just started today,

00:54:55.040 | I guess Google and AMD,

00:54:58.960 | what's your quick interpretation of what you're seeing

00:55:01.520 | and what do you expect?

00:55:02.400 | - Well, I think we came into this week,

00:55:04.560 | we and lots of others,

00:55:07.200 | like if you look over the course of the past several weeks,

00:55:09.360 | the Mag 5 has underperformed software and other categories.

00:55:14.320 | And I think people are pretty concerned

00:55:16.080 | about the CapEx guides.

00:55:17.600 | There's been a lot of talk about that.

00:55:19.280 | I came into this week and said to my team,

00:55:22.240 | I think it's largely priced in at this point in time.

00:55:24.880 | I said, I thought the whole Mag 5

00:55:26.560 | would probably trade up three to 5% this week,

00:55:29.520 | including Google.

00:55:30.480 | Google came out tonight,

00:55:31.680 | they grew their search revenues by 12%.

00:55:35.440 | They grew YouTube revenues by 12%.

00:55:38.960 | The stock's up, I think, four or 5% after hours.

00:55:42.560 | That all smells right to me.

00:55:44.160 | It had traded off a lot going into this period.

00:55:48.000 | I think the more interesting thing, Bill,

00:55:50.320 | is if you look at kind of--

00:55:51.600 | - Hey, can I ask you just a quick follow-up.

00:55:53.600 | When you said they were worried about the CapEx guide,

00:55:56.080 | which way were they worried?

00:55:57.360 | - They were worried that it was gonna be even higher.

00:55:59.920 | - Yeah. - Okay, so--

00:56:01.040 | - But from your analysis earlier,

00:56:03.040 | it's a good thing if they wanna spend more.

00:56:04.720 | - Well, no, I think that there is,

00:56:08.800 | I mean, remember, these CapEx expenses

00:56:12.480 | have gone fairly parabolic in the last couple of years.

00:56:15.280 | - Yeah. - But operating income,

00:56:19.040 | so margin expansion and top-line growth

00:56:21.920 | have outpaced the increase in CapEx.

00:56:24.800 | So people have tolerated it.

00:56:26.560 | What won't work well, right?

00:56:28.640 | Let's say we had a recession next year,

00:56:30.960 | and then earnings came in a lot lower than people expected,

00:56:35.120 | or top-line growth came in lower,

00:56:37.120 | but you were unhinged on CapEx.

00:56:40.720 | I don't think that would be a particularly good equation

00:56:43.040 | for the stock.

00:56:43.760 | So some people were quite worried about this.

00:56:47.040 | But I think, let's talk about Google in particular, right?

00:56:52.800 | I think there's a lot of concern and chatter in the world.

00:56:55.360 | I've been out there talking about it.

00:56:57.200 | Chat GPT has consumed an extraordinary percentage

00:57:01.360 | of the time and energy that I used to spend in search.

00:57:04.720 | So it's kind of hard to believe

00:57:07.520 | that it's not really having an impact on search revenues,

00:57:10.080 | but it's not.

00:57:11.200 | And so if you look at this chart

00:57:12.560 | that shows the search revenue growth

00:57:14.880 | over the course of the last eight quarters,

00:57:16.400 | I mean, they've accelerated search revenue growth

00:57:19.920 | from end of '22 and early '23,

00:57:24.000 | from 2%, 5%, all the way up to 14%,

00:57:27.360 | still at 12% today.

00:57:29.520 | So part of the question is,

00:57:31.760 | how long can search revenue growth stay at this level?

00:57:34.640 | And one of the key questions there

00:57:36.480 | was whether or not the AI-enhanced search results

00:57:41.040 | were going to monetize at the same level

00:57:43.920 | as non-AI-enhanced search results.

00:57:46.000 | So they call these SGE.

00:57:49.040 | And it turns out that they said on the call tonight

00:57:53.120 | that they're monetizing at the same level.

00:57:55.120 | So that's a really good data point for Google, if true,

00:57:59.760 | which is they can layer in the answer at the top,

00:58:03.440 | push the blue links down the page

00:58:06.000 | without losing a lot of revenue.

00:58:07.440 | Although, I mean, this has been argued

00:58:11.280 | both as a positive and negative,

00:58:12.800 | but the searches that were, I guess,

00:58:15.920 | stolen most quickly by a chat engine

00:58:18.320 | are more of these kind of encyclopedic-type searches

00:58:21.680 | where you're looking for information,

00:58:23.360 | whereas the big money searches for Google

00:58:27.040 | are travel and new car and a home

00:58:30.320 | and those kind of things, big purchases.

00:58:32.640 | And so those may not, in fact, be the ones

00:58:35.440 | where they're inserting much of this.

00:58:37.120 | So you may be talking about the long tail

00:58:41.760 | of their searches anyway

00:58:43.600 | that aren't that relevant to the big dollars.

00:58:46.240 | You know what I'm saying?

00:58:46.800 | I don't know, for sure.

00:58:48.320 | I mean, a lot of people think of these

00:58:49.680 | as the knowledge graph searches.

00:58:51.120 | You know, I've been using, you know,

00:58:53.200 | all last night I was using ChatGPT

00:58:55.120 | to help Lincoln study for his AP History exam.

00:58:59.200 | There's not a lot of monetization of that

00:59:01.040 | in Google anyway.

00:59:01.600 | So that page could be the same.

00:59:03.280 | Right, so that monetization is about the same.

00:59:06.560 | Now, we've also, as you know,

00:59:09.200 | we saw this week that Anthropic released computer use,

00:59:12.480 | to take control of my browser,

00:59:14.880 | to, you know, what we've been discussing,

00:59:17.680 | book that hotel or book that airline ticket.

00:59:20.320 | It's kludgy, you know, a lot of people, you know,

00:59:23.840 | but it's a step in the direction.

00:59:25.520 | A lot of people thought that Google

00:59:26.720 | was going to release Jarvis this week,

00:59:28.480 | which is their version of this.

00:59:30.160 | Someone just funded a new browser company,

00:59:32.160 | Kleiner did, I think.

00:59:33.120 | OpenAI will certainly do this.

00:59:35.200 | For the agentic browser.

00:59:36.160 | So, you know, the fact of the matter is,

00:59:38.560 | Google is already discounting this.

00:59:40.080 | It's trading at the lowest multiple,

00:59:41.680 | you know, in the mag five.

00:59:43.520 | So I think people are saying at some point,

00:59:45.760 | it's going to have some effect.

00:59:47.040 | We don't know exactly when,

00:59:48.720 | but I would just say for this quarter,

00:59:50.160 | you know, super solid quarter out of the company.

00:59:53.120 | And then you asked about AMD.

00:59:54.400 | AMD's down about 10% after hours,

00:59:57.280 | despite, you know, continuing to have blistering growth.

01:00:00.480 | And I think what's going on there is,

01:00:02.640 | there was some talk at the beginning of the year,

01:00:04.800 | that AMD could capture maybe 10% of the data center market.

01:00:10.560 | And the fact of the matter is,

01:00:11.920 | when they gave some forward guidance tonight,

01:00:14.560 | that suggests they're going to stay at about 3%

01:00:17.120 | of the GPU kind of data center market,

01:00:21.200 | relative to NVIDIA at about 90%.

01:00:23.680 | The company's done extraordinarily well.

01:00:26.640 | Lisa is an amazing, you know, CEO.

01:00:29.280 | And I think there are a lot of good things in the works,

01:00:31.920 | but, you know, they're not closing the gap

01:00:34.560 | in terms of share of market.

01:00:36.160 | The tide is rising for both parties,

01:00:38.320 | but the opportunity clearly for them.

01:00:40.240 | - And it's off big now.

01:00:41.680 | - Right, is to try to pierce a larger share of that market,

01:00:47.120 | but we didn't see any move forward.

01:00:49.440 | - You know, back on the Google thing real quick,

01:00:51.200 | I just have a really hard time, like in my brain,

01:00:55.840 | separating kind of my own user experience shift

01:01:00.000 | and how fewer Google searches I do,

01:01:02.400 | to accepting the idea that,

01:01:05.600 | oh, well, they're going to add it in too.

01:01:07.600 | Like, I don't go back there and say,

01:01:10.800 | oh, this is better, like with it added in

01:01:13.040 | and the ads and stuff.

01:01:14.240 | I prefer the--

01:01:15.840 | - Perplexity or--

01:01:16.800 | - Perplexity or chat GPT, I really do.

01:01:18.960 | So I don't feel--

01:01:20.880 | - Right, and so that to me is the leading edge, right?

01:01:23.840 | Like at the end of the day,

01:01:25.200 | could Google give you that same clean experience

01:01:27.840 | that chat GPT gives you?

01:01:29.040 | Of course they could, right?

01:01:30.720 | But they cannibalize, that's what innovator's dilemma means.

01:01:34.000 | They cannibalize their core business.

01:01:36.640 | And so they either have to, you know,

01:01:39.360 | have a product that looks different,

01:01:40.960 | that still has the blue links,

01:01:42.320 | that still has the sponsored ads, right?

01:01:44.720 | And you risk losing user affinity, right?

01:01:48.560 | Or, you know, so to me it's--

01:01:50.480 | - Well, they have this other challenge

01:01:51.840 | that I think is hard for people to understand,

01:01:54.560 | but I've explained it a couple of times,

01:01:56.080 | which is their revenue per visit is going to fall

01:02:01.120 | if they complete the transaction.

01:02:03.600 | Because someone is spending marketing dollars

01:02:06.800 | to take that customer to their website

01:02:08.880 | and run the transaction with the hopes

01:02:11.600 | that they'll come directly back to that website.

01:02:14.000 | So they'll spend 50 to 100% of first transaction.

01:02:18.000 | There's no one that's gonna,

01:02:19.120 | if Google or chat GPT or whoever convinces Kayak or whoever

01:02:26.000 | to be a white label rails in the background,

01:02:30.480 | the amount of money they're gonna give you for that,

01:02:32.560 | it's gonna be like 5% of the transaction.

01:02:35.760 | - Bill, you brought something interesting.

01:02:37.600 | So like this new browser, right?

01:02:38.800 | It's company called Browserbase.

01:02:40.320 | And I think this is a really interesting concept

01:02:42.960 | to think about kind of transitionary states

01:02:45.120 | versus end states, right?

01:02:46.400 | The transitionary state today

01:02:48.560 | is we're gonna have agents

01:02:49.600 | that essentially can scrape websites

01:02:52.000 | and view it as if I'm doing it.

01:02:54.240 | They're gonna recognize the button and click the button.

01:02:56.160 | They're gonna go here.

01:02:56.880 | What if we reinvented the browser

01:02:59.600 | and built it in a way

01:03:00.720 | where it's not a button for a human eye,

01:03:02.640 | but everything is some composable API

01:03:05.360 | that an agent can recognize

01:03:06.960 | and an agent can go interact with.

01:03:08.320 | And so you're not scraping the website

01:03:10.880 | where you would lose the eyeballs.

01:03:12.720 | You're building a new browser for agents versus humans.

01:03:16.560 | Is that the end state?

01:03:17.440 | There's lots, I don't know the answer,

01:03:18.640 | but it's an interesting question.

01:03:20.360 | - You have to rebuild all the rails too.

01:03:22.400 | I mean, everything has to start to look like Stripe.

01:03:25.760 | There has to be a travel Stripe.

01:03:28.640 | There has to be, like you need a transactional network

01:03:34.400 | and that's not what we have today.

01:03:36.640 | Like we have this browser-based world

01:03:42.240 | that sits between the consumer and all these businesses.

01:03:45.760 | But it could happen.

01:03:47.840 | And boy, we've talked about voice.

01:03:49.360 | Like if you just start talking to your computer,

01:03:51.760 | it's a very different world.

01:03:54.080 | That's not a browser or it's not the browser we know.

01:03:58.400 | We'll be a different entity.

01:04:00.400 | - Yeah.

01:04:00.800 | - Well, I think the way you go over the top

01:04:03.520 | and why I'm so excited

01:04:06.800 | about the equivalent of computer use across the board

01:04:10.480 | is in the same reason perhaps that Elon says

01:04:13.360 | you use a humanoid versus a non-humanoid robotic form

01:04:17.360 | because a humanoid works in a world

01:04:20.000 | that was designed for humans, right?

01:04:22.080 | And the thing about computer use,

01:04:23.600 | you don't have to build all these new rails and everything,

01:04:26.400 | which takes a really long time to negotiate all of that stuff.

01:04:29.760 | Instead, you just give it ubiquitous use

01:04:32.400 | of the world that already exists straight out of the gate.

01:04:37.280 | - I think this is something I really wanted to make a point on.

01:04:42.160 | I think it's gonna be super interesting

01:04:43.600 | to watch which categories

01:04:45.600 | and which verticals fall into place first and why.

01:04:49.760 | There was an article out this week

01:04:52.000 | that was looking at, I think, open AI as a scribe.

01:04:56.160 | And they highlighted that it failed a lot.

01:05:00.000 | And this is also an issue with self-driving.

01:05:05.680 | If the cost of a 5% error or a 10% error is super high,

01:05:12.080 | it's gonna take AI a long time to get there.

01:05:15.200 | I would argue programming's that way.

01:05:16.880 | You can't have 5% or 10%--

01:05:18.800 | - Self-driving cars.

01:05:19.760 | - Wrong.

01:05:20.000 | Self-driving cars.

01:05:20.480 | - The error rate is death.

01:05:21.440 | - It may turn out that customer service, that's just fine

01:05:25.360 | because the best in class today already has that error rate

01:05:29.600 | just because of human failure.

01:05:32.080 | And so it will take AI a while to get

01:05:36.560 | to where it has the kind of nines you need

01:05:39.840 | to solve certain problems.

01:05:41.760 | And that's why I'm so intrigued by which verticals

01:05:44.480 | are gonna line up first and get knocked over.

01:05:46.640 | - Well, another good conversation, Jamin.

01:05:50.000 | Great blog.

01:05:50.480 | - Thanks for joining us.

01:05:51.040 | - Thanks for having me.

01:05:52.120 | - Great blog and I imagine we're gonna get a lot of feedback

01:05:57.560 | on this one from LPs and founders.

01:05:59.320 | So we'll make sure to put it in the notes.

01:06:02.040 | Great to see you both.

01:06:02.760 | Thanks for being here.

01:06:03.240 | - Good to see you.

01:06:03.880 | Take care.

01:06:04.280 | (upbeat music)

01:06:14.600 | - As a reminder to everybody,

01:06:16.280 | just our opinions, not investment advice.

Ep19. State of Venture, AI Scaling, Elections | BG2 w/ Bill Gurley, Brad Gerstner, & Jamin Ball

Chapters