Ep16. Nuclear Update, AI Fast & Furious, State of VC | BG2 w/ Bill Gurley & Brad Gerstner

00:00:00.000 | There's a picture you can look up that's kind of disgusting,

00:00:02.540 | so people may not want to.

00:00:03.820 | But there's this thing called a gavage tube, which is what

00:00:08.460 | they use to make foie gras.

00:00:09.940 | It's how they force feed the geese to get them just super

00:00:14.380 | fat.

00:00:14.860 | And that's the image I have in my mind,

00:00:17.380 | like are we overfeeding these startups?

00:00:20.220 | [MUSIC PLAYING]

00:00:23.580 | [KNOCKING]

00:00:32.680 | Hey, Bill.

00:00:33.180 | Great to see you.

00:00:34.380 | Good to see you, Brent.

00:00:35.660 | Man, that was an amazing pod at Diablo Canyon.

00:00:39.220 | The inbound regarding just nuclear

00:00:42.900 | has been off the charts from literally senior policymakers,

00:00:46.700 | senators, and House members on both sides of the aisle.

00:00:49.260 | It really feels like the dam is broke.

00:00:52.580 | We half-joked, Bill, that Microsoft--

00:00:55.440 | when we were down there, we discovered

00:00:58.260 | that there were four unbuilt nuclear reactors that are all

00:01:02.860 | already plotted on the site.

00:01:05.420 | And we half-joked that NVIDIA and Microsoft and Oracle

00:01:08.940 | could come sponsor these reactors

00:01:11.420 | and that they could have a new type

00:01:13.300 | of public-private partnership with the government

00:01:15.660 | and build data centers right next to them.

00:01:18.340 | And it turns out it wasn't so far out there.

00:01:20.280 | I mean, Oracle has announced that they may do some things

00:01:23.340 | with small nuclear reactors.

00:01:25.380 | Amazon is buying this nuclear-powered Talon data

00:01:29.220 | center facility.

00:01:30.180 | And now Microsoft this week announces with CEG

00:01:33.500 | that they're going to bring Three Mile Island out

00:01:36.480 | of retirement.

00:01:37.740 | It's incredible to see the beginnings of what

00:01:40.020 | may be a US nuclear renaissance.

00:01:42.940 | Certainly, the momentum has--

00:01:45.660 | if maybe it was headed up and then it's kind of reached

00:01:48.660 | an apex and kind of fallen over.

00:01:51.020 | The FT article yesterday or today

00:01:54.220 | highlighting that 14 different banks have shown up

00:01:57.860 | at a climate conference with a confirmation of a willingness

00:02:01.900 | to invest is just huge.

00:02:03.460 | And I think there are two things that are big takeaways for me.

00:02:07.700 | One, we were talking about one of the limits on SMR

00:02:11.420 | and on any new innovation in the space

00:02:14.220 | was that utility companies are traditionally very conservative.

00:02:18.100 | And I like to think about it in the framework of crossing

00:02:20.620 | the chasm.

00:02:21.580 | You basically are selling only to laggards.

00:02:24.180 | And that's very difficult, especially

00:02:27.100 | for a capital-intensive startup, to be selling only to laggards.

00:02:30.900 | And what may have transpired literally in the past month

00:02:36.260 | is the hyperscalers--

00:02:38.500 | and this may have started before then

00:02:40.240 | because Amazon did the deal with CEG a little while back.

00:02:43.740 | But if the hyperscalers become part

00:02:47.180 | of the customer set for the nuclear startups,

00:02:51.300 | that may be 10x better than selling just to utilities

00:02:55.900 | alone.

00:02:56.740 | You may have brought innovators to the table

00:02:59.540 | on the purchasing side that may be more open-minded,

00:03:02.300 | that may be more understanding, may be more

00:03:04.620 | willing to share risk, which could be

00:03:06.540 | very positive for the SMR market.

00:03:09.220 | So that'd be my one big takeaway.

00:03:11.300 | And the second one is just that a lot of times,

00:03:15.500 | I think people look at big, big problems

00:03:18.940 | and think they're insurmountable.

00:03:20.700 | And I remember actually in the past two years

00:03:24.260 | being at a off-site conference at a think tank

00:03:28.740 | where we were talking about climate change.

00:03:31.620 | And about 80% of the way in, someone raised their hand

00:03:35.420 | and said, why aren't we talking about nuclear?

00:03:37.300 | And all the scientists in the room said, oh, no,

00:03:40.540 | we're not going to put that back on the table.

00:03:43.060 | That's too far gone.

00:03:44.860 | That's past.

00:03:46.020 | And it turned out that wasn't true.

00:03:48.620 | It turned out there was an opportunity

00:03:51.260 | to get a renaissance in thinking about this.

00:03:53.420 | And it started, I think, with people like Steven Pinker, who

00:03:56.900 | are wildly regarded scientists, saying, no, this

00:03:59.460 | is our best path out.

00:04:01.300 | But then we talked about Patrick Collison.

00:04:04.060 | And others kind of jumped on the bandwagon.

00:04:06.180 | And then there were plenty of pro-nuclear advocates

00:04:10.020 | that were sticking their neck out.

00:04:11.460 | And then Elon gets in the game.

00:04:13.780 | And then this data center thing may

00:04:15.460 | have been just the impetus you needed

00:04:18.020 | to get people over the top.

00:04:20.020 | And we were lucky enough to kind of time our thing

00:04:23.020 | as this transition was happening.

00:04:25.660 | But it is possible to create kind of wholesale change

00:04:30.820 | in how people think about something.

00:04:32.500 | But it takes a lot of work by a lot of people.

00:04:34.820 | And everyone that kind of stuck their neck out early.

00:04:38.060 | Josh Wolf was another one that was sticking his neck out

00:04:40.540 | on this topic.

00:04:41.340 | So I congratulate all of them.

00:04:44.020 | And it feels like the momentum's now behind us.

00:04:46.340 | And I literally feel bad for the citizens of Germany.

00:04:53.420 | One thing that is very apparent is that the easiest thing to do

00:04:58.420 | is bring--

00:04:59.620 | well, start with don't decommission

00:05:02.060 | any of these things.

00:05:03.220 | But second, if any have been decommissioned recently,

00:05:06.820 | try and bring them back.

00:05:08.340 | And I hope there are some sane minds in Germany

00:05:11.940 | that are watching all this.

00:05:13.100 | Because I think the world would benefit

00:05:14.820 | from them reversing that decision

00:05:17.620 | and running back at this.

00:05:19.380 | One of the things I learned as well--

00:05:21.100 | because you and I talked a lot just about how do they

00:05:23.220 | underwrite--

00:05:24.580 | how would the hyperscalers underwrite building out

00:05:27.220 | those nuclear reactors?

00:05:28.820 | And one of the things I learned after our pod

00:05:31.620 | was that these companies that are considering nuclear,

00:05:36.020 | they are spending billions of dollars a year

00:05:38.540 | on carbon offsets.

00:05:39.820 | And you know there's a lot of criticism about these carbon

00:05:42.260 | offset markets.

00:05:43.460 | But I dug up some data.

00:05:45.660 | In 2020, Morgan Stanley estimates

00:05:49.220 | that about $2 billion was spent on the carbon offset market.

00:05:52.900 | And by 2030, they expect that to be $100 billion

00:05:56.140 | by these large hyperscalers that have to buy

00:05:59.780 | these massive carbon offsets.

00:06:01.460 | Now, if instead you're investing in nuclear clean energy,

00:06:05.580 | if the source of the energy that is powering your data centers

00:06:08.740 | is clean, then you actually have to get to buy

00:06:12.020 | fewer of the carbon offsets.

00:06:13.900 | So that may make it easier, again,

00:06:17.020 | for them to pencil out the math.

00:06:18.380 | That may be what we're seeing some of the dam break.

00:06:20.860 | I think a huge part of this is just the public consensus.

00:06:24.340 | Nobody wants to invest in something

00:06:26.060 | that all your customers are against.

00:06:28.740 | And we know, we've shown the data here,

00:06:30.700 | that this is now popular again among consumers

00:06:33.700 | because they understand it's clean, it's carbon-free.

00:06:37.540 | The other data point that broke since we did that

00:06:40.300 | was the Three Mile Island restart.

00:06:42.740 | I think there were rumors of it before we did the podcast

00:06:45.860 | with CEG, with Constellation Energy Group.

00:06:48.540 | And there were quotes in these articles

00:06:51.220 | about a survey of the Pittsburgh residents.

00:06:56.060 | And they were supportive of Philadelphia.

00:06:59.980 | And what that says to me--

00:07:03.220 | and look, this is knowing that the customer is really Amazon

00:07:07.460 | and not the citizenry.

00:07:09.380 | And I was just shocked by that.

00:07:11.060 | And so we all know that one of the reasons this happened

00:07:15.260 | was that there was an irrational public response

00:07:19.540 | to the negative risk of these solutions.

00:07:24.380 | And it is super unfortunate that that takes so long to heal.

00:07:31.020 | But time is the best way to get past something like that.

00:07:35.020 | And it's been a long time.

00:07:36.220 | And I think people have a lot more data.

00:07:38.420 | We're not there yet.

00:07:39.380 | We need to keep the pressure on.

00:07:41.140 | We said we'd like to see Gavin Newsom extend Diablo

00:07:45.100 | by another 15 years.

00:07:46.460 | That's on his desk right now.

00:07:48.620 | That facility has at least another 40 years left in it.

00:07:51.660 | So I think we all need to keep the pressure on.

00:07:54.020 | But the nice thing is there's good bipartisan support.

00:07:57.460 | People view nuclear as not only a matter of climate security,

00:08:01.420 | but now it's a matter of national security

00:08:03.980 | because it's the primitive to all of AI.

00:08:06.860 | And so I'm excited for the momentum.

00:08:08.780 | But speaking of the exploding need for more baseload power

00:08:13.020 | to feed the AI beast, let's talk about our first topic

00:08:17.060 | and where things currently stand in AI.

00:08:19.820 | And I think since we last talked,

00:08:22.660 | people have continued to climb, build this wall of worry,

00:08:25.780 | like whether AI is in a bubble or not.

00:08:28.820 | Of course, last week, OpenAI was rumored

00:08:32.460 | or is now well-known to be raising capital

00:08:34.740 | at $150 billion valuation,

00:08:36.820 | or at least that's the Bloomberg headline.

00:08:39.540 | I can confirm that Altimeter is talking with the company.

00:08:42.060 | So of course, there's some things I can share

00:08:43.660 | and things I can't share.

00:08:45.540 | But you had Kevin Scott, the CTO of Microsoft,

00:08:47.940 | say demand for AI infrastructure

00:08:50.100 | is materially outpacing our ability to supply it,

00:08:53.580 | even as we are building at a pace unseen.

00:08:57.780 | Jensen said at the Goldman Sachs conference

00:09:00.380 | that they will be under-supplied not only this year,

00:09:03.140 | but for a while to come.

00:09:04.700 | So where do you come down on this, Bill?

00:09:06.420 | Let's just start with the demand for training

00:09:09.140 | and inference and data centers and power.

00:09:12.180 | Do you think we're headed for a glut?

00:09:14.940 | Before I answer that question,

00:09:17.820 | which we've talked about many times before,

00:09:20.140 | I will state that since two podcasts ago,

00:09:24.900 | 'cause we focused on Diablo and didn't talk about this.

00:09:27.820 | So you go back, I guess, probably four or five weeks.

00:09:30.860 | If anything, the balance of enthusiasm

00:09:34.860 | and willingness of individuals to commit capital

00:09:41.380 | has gone up, not down.

00:09:44.500 | If you go back maybe a year,

00:09:45.940 | I don't know that Oracle was really in this discussion

00:09:48.780 | from a spin standpoint, right, on scaling things out.

00:09:52.660 | And clearly they are now.

00:09:54.420 | They're at the table.

00:09:55.260 | They want to be considered like one of the hyperscalers.

00:09:58.620 | So in addition to Oracle kind of stepping up

00:10:01.020 | and being a big player,

00:10:02.100 | we get an announcement that BlackRock and Microsoft

00:10:06.100 | are teaming up to raise a fund

00:10:08.740 | that would be 30 to a hundred billion dollars

00:10:11.060 | just to finance data centers.

00:10:12.620 | They also mentioned energy.

00:10:14.380 | But Microsoft was already building their owns

00:10:16.180 | and already supporting CoreWeave.

00:10:17.980 | And so this is just more and more commitment

00:10:21.300 | to rolling out more infrastructure.

00:10:23.300 | I think the Middle East is involved

00:10:25.260 | in this announcement as well.

00:10:26.460 | And there's been a lot of talk of,

00:10:28.180 | and there's a lot of companies like G42

00:10:31.260 | that are pushing for even more spend in the Middle East.

00:10:34.820 | So everything screams even more demand.

00:10:39.820 | You and I have argued or talked about

00:10:41.820 | whether this is really supply or demand,

00:10:43.780 | but presumably these people aren't acting irrationally.

00:10:46.860 | So since we talked last,

00:10:49.260 | I would say the world's gotten more enthusiastic

00:10:53.100 | about AI than it was five or six weeks ago.

00:10:55.780 | Right, right.

00:10:56.620 | And I agree with you.

00:10:58.380 | I think you said something really important,

00:11:00.180 | which is when you're planning out three to five years

00:11:03.700 | and we're talking about tens of billions of dollars,

00:11:06.620 | one has to assume that Sachin, Jensen and Oracle

00:11:11.660 | and then Amazon, et cetera, that what they're seeing

00:11:14.380 | justifies the spend.

00:11:15.740 | And one of the things driving the debate

00:11:18.100 | over that demand bill is all these new models

00:11:20.940 | that have been popping out.

00:11:21.900 | We finally saw Strawberry, the O1 preview model,

00:11:26.340 | whole new vector of scaling around inference time reasoning

00:11:29.700 | out of open AI, which is exciting.

00:11:32.140 | There's a new Anthropic Opus model,

00:11:34.300 | 3.5 that's expected to drop this week.

00:11:36.740 | MetaConnect is in a couple of days, I'll be there.

00:11:39.900 | And Mark Zuckerberg is expected

00:11:42.260 | to announce several new models there,

00:11:44.060 | smaller models and larger models.

00:11:47.060 | And so despite the improvements

00:11:50.060 | in model training efficiency,

00:11:52.260 | the aggregate training and the velocity

00:11:55.020 | seems to continue to grind a lot higher.

00:11:59.060 | We were lucky to have Noam Brown,

00:12:01.660 | who's the inventor and the leader of the O1 preview.

00:12:05.340 | So remember, we talked about Noam before.

00:12:07.860 | He was at Meta and did Pluribus and Liberatus,

00:12:10.780 | won the game of six-handed poker.

00:12:13.380 | He spent a lot of time thinking

00:12:15.060 | about this inference time reasoning.

00:12:17.420 | And we're really early in this journey,

00:12:21.060 | but it's already kind of a big wow, right?

00:12:24.220 | This idea that in addition to pre-training,

00:12:27.140 | we're now going to allow models to reason and think

00:12:31.340 | as part of the response process to the prompt.

00:12:34.620 | I don't know if you looked at any of these tweets

00:12:37.060 | out of Noam Brown or others about the O1 preview,

00:12:40.420 | but just curious if you had any thoughts or reaction.

00:12:43.380 | I know you were blown away by the voice model.

00:12:45.780 | It's been slow in dropping,

00:12:48.100 | but I don't know if you've played with the O1 preview.

00:12:49.900 | Well, yeah, and on the voice thing,

00:12:51.780 | I think they may have been focusing

00:12:54.740 | on their enterprise customers.

00:12:56.100 | I've talked to some enterprise customers

00:12:58.820 | who have that product in-house

00:13:00.540 | and they're very excited about it.

00:13:01.980 | And let me just say quickly on voice,

00:13:04.980 | one thought I had when you think about it being delivered

00:13:08.620 | as an API and not just as a consumer product,

00:13:11.900 | there is this question is,

00:13:13.540 | does the input to the computer become voice

00:13:17.860 | for the first time?

00:13:18.740 | Microsoft was talking about that maybe two decades ago

00:13:23.660 | and Gates was super excited about it.

00:13:26.460 | Yeah, tell me, we were big in tell me,

00:13:29.860 | but if it gets so good,

00:13:32.420 | and this becomes the way you talk to your computer.

00:13:36.900 | And another thing I would say is talk to websites.

00:13:40.260 | Like, could you imagine you walk up to a kayak

00:13:44.780 | or on the website and you just start talking,

00:13:48.740 | that'll be a radical new dimension

00:13:52.660 | for how we use our computers.

00:13:54.740 | So I'm excited to see how that plays out.

00:13:56.900 | On the strawberry release,

00:14:00.580 | the one thing that I think is important

00:14:02.620 | for people to realize is the graphs that were shown,

00:14:05.820 | and we can certainly put some links in the show notes,

00:14:07.940 | but I think everyone's seen it.

00:14:09.500 | The access was logarithmic.

00:14:13.020 | The implication being in order to get linear improvement,

00:14:16.620 | you have to do maybe 10X the amount of processing.

00:14:21.260 | And this is all inference.

00:14:22.700 | So what are the implications of that?

00:14:25.100 | Well, one, you have to figure out what problem cases

00:14:28.980 | are good scenarios for being willing to spend 10X

00:14:33.980 | or 100X as much on inference to get to a better solution.

00:14:41.340 | I don't think it's all of them,

00:14:42.860 | but many people believe it's a lot of them.

00:14:45.780 | And then the second thing that comes out of that is

00:14:48.340 | if there are a lot of them that are willing to pay

00:14:52.740 | 10 to 100X on inference to get linear improvements,

00:14:57.220 | then the percentage of compute,

00:15:00.220 | and you need to run math models

00:15:04.140 | I don't have in front of me now,

00:15:05.220 | but maybe the dollars of compute

00:15:06.980 | are gonna move more towards inference than training

00:15:10.100 | as we move forward.

00:15:11.820 | - I think there's no doubt.

00:15:13.940 | So there's a video of Jensen out there

00:15:15.860 | that we'll put in here.

00:15:17.500 | - One of the things that Sam introduced recently,

00:15:22.500 | the reasoning capability of these AIs

00:15:25.580 | are gonna be so much smarter,

00:15:28.180 | but it's gonna require so much more computation.

00:15:31.180 | And so whereas each one of the prompts today

00:15:34.620 | into ChatGPT is a one pass,

00:15:37.100 | in the future it's going to be hundreds of passes inside.

00:15:39.860 | It's gonna be reasoning.

00:15:40.780 | It's gonna be doing reinforcement learning.

00:15:42.260 | It's gonna be trying to figure out

00:15:43.860 | how to create a better answer,

00:15:45.140 | reason a better answer for you.

00:15:47.020 | - When you use a model like Strawberry,

00:15:48.860 | you're likely to see 100X more inference, right?

00:15:51.660 | Because rather than single shot prompting,

00:15:53.940 | there's a lot of recycling that's going on

00:15:56.460 | as part of the reasoning process.

00:15:58.500 | Just mathematically, we know that's going to lead

00:16:02.140 | to a massive explosion in inference.

00:16:04.340 | Now, if you look at these GB200s that NVIDIA is selling,

00:16:08.580 | the cost or the improvement in inference,

00:16:10.660 | I mean, NVIDIA says it's a 50X improvement in inference.

00:16:14.740 | Other people say it's a 3X improvement in inference,

00:16:17.420 | but there's clearly a lot of focus on inference.

00:16:19.780 | I think the world is inference constrained.

00:16:21.980 | I think part of the reason for that

00:16:24.260 | is that you have new models emerging like this

00:16:28.020 | that are gonna have machines talking to machines,

00:16:30.020 | lots of inference going on in the background.

00:16:32.580 | Of course, you have companies like Grok and Cerebus

00:16:34.900 | that are bringing really fast inferencing to the table.

00:16:37.700 | But again, I think we're talking about

00:16:39.980 | many orders of magnitude increase

00:16:42.460 | to the amount of inference,

00:16:44.020 | which is going to be needed in the future.

00:16:47.380 | And if you're building a data center,

00:16:49.620 | think about this, Bill.

00:16:51.380 | If I'm making a $10 billion investment

00:16:53.780 | in a 300-megawatt data center,

00:16:55.700 | I wanna be able to use that for both training

00:16:57.940 | and for inference when my training run is not happening.

00:17:01.540 | So when you look at the total cost of operation

00:17:03.900 | in one of these data centers,

00:17:05.780 | you're gonna see a lot more activity.

00:17:07.220 | And my hunch is part of the reason

00:17:09.540 | that voice hasn't dropped, Bill,

00:17:12.020 | is that they've all said it.

00:17:15.380 | I think that OpenAI and Microsoft and others

00:17:19.020 | are inference-constrained at the moment

00:17:21.860 | in terms of the demand on these systems.

00:17:24.100 | I think the systems will also get more intelligent

00:17:26.500 | where they'll route the request to the simplest models

00:17:30.820 | to answer the particular question.

00:17:32.820 | So you don't need O1 for really basic questions.

00:17:37.820 | That you might be able to route, in fact,

00:17:39.940 | to a GPT-3-like model.

00:17:41.980 | But having intelligent layering of these models,

00:17:44.380 | an ensemble of models put together

00:17:47.340 | so that you get the answers in the fastest amount of time,

00:17:50.660 | and it's gonna be different models

00:17:52.140 | for different sorts of questions.

00:17:54.260 | Yeah, and whether or not the engine can interpret

00:17:58.820 | which of those it is will be important.

00:18:00.660 | I would go back and highlight

00:18:02.940 | that even in the announcements from OpenAI on Strawberry,

00:18:06.460 | they admit or they disclose or qualify

00:18:12.740 | that there are many instances

00:18:15.180 | where the extra iteration led to worse results.

00:18:19.180 | So you really do need to be able to figure out

00:18:22.420 | the type of problem and whether or not

00:18:25.140 | you're gonna get improvement from that effort.

00:18:28.020 | And I personally don't think we know enough yet

00:18:32.300 | to know which problems fit in there or not.

00:18:35.900 | But I do think people are super excited

00:18:39.180 | about what's possible on that front.

00:18:42.900 | My biggest thing about it is like, listen,

00:18:45.140 | we've been talking about bigger models,

00:18:47.060 | more parameters, all of this.

00:18:48.420 | That's basically been the exclusive vector of conversation

00:18:51.780 | for scaling intelligence.

00:18:53.100 | This is a totally new vector.

00:18:54.940 | And now you have the compounding benefit,

00:18:58.220 | two different ways to scale intelligence

00:19:00.820 | that I think is super exciting.

00:19:02.180 | And just real quick,

00:19:03.580 | I had a second thing before you move forward.

00:19:06.220 | I could restate when you use the phrase

00:19:10.140 | inference constraint,

00:19:11.580 | that may be a financial problem too.

00:19:14.140 | Like it may be super expensive to run advanced voice

00:19:17.380 | relative to what you're charging for.

00:19:19.780 | And so especially when you talk about

00:19:23.340 | logarithmic increase in spend,

00:19:26.100 | I think these companies develop these breakthroughs

00:19:29.220 | and they're eager to share them with the world.

00:19:30.900 | And so they put them out there maybe in a freemium

00:19:34.060 | or maybe in a kind of early test thing.

00:19:37.100 | But if some of them do have much higher underlying costs,

00:19:40.700 | we do need to figure out,

00:19:43.020 | what are the business models for these things?

00:19:45.380 | How much are people willing to pay?

00:19:46.980 | I've heard people on other podcasts say,

00:19:49.580 | well, for a perfect assistant, I might pay 10 grand a year,

00:19:53.420 | but no one has that product on the market right now.

00:19:56.460 | And so I think there's a lot of experimentation

00:20:00.060 | with business models that's gonna have to happen as well.

00:20:02.820 | I couldn't agree more.

00:20:03.740 | I think there's gonna have to be,

00:20:05.100 | I mean, think about this.

00:20:06.220 | There's gonna be massive price discrimination.

00:20:08.340 | You can't charge in the Philippines,

00:20:09.700 | what you're gonna be able to charge in the United States.

00:20:11.500 | You can't charge to the tail end,

00:20:13.140 | what you're gonna be able to charge to the head end.

00:20:14.980 | But one thing that is true is it looks like OpenAI

00:20:19.980 | has something like 200 million weekly MAUs.

00:20:24.940 | That's a number widely reported.

00:20:26.540 | It's a huge, huge number with little to no advertising.

00:20:30.940 | And it seems to me,

00:20:31.860 | this is really this benefit of going first, Bill.

00:20:35.060 | Billions of free ad impressions.

00:20:37.420 | It continues to grow.

00:20:39.180 | And so I asked the team to just to take a look

00:20:41.580 | at the time to reach a hundred million MAUs,

00:20:44.540 | and you could see this chart just,

00:20:47.500 | it took Chachapiti a fraction of the time

00:20:51.300 | that it took YouTube or Instagram or Facebook

00:20:55.220 | to get to the same place.

00:20:56.220 | It took Facebook about five years.

00:20:58.500 | Now that obviously has continued to go up.

00:21:02.220 | Weekly average users are 200 million.

00:21:04.940 | It's pretty extraordinary.

00:21:06.660 | And even by country,

00:21:07.740 | if you look at the spread of Chachapiti,

00:21:09.540 | it's clearly spreading on a global basis.

00:21:12.220 | And then finally this tweet by Vivek Goyal on my team,

00:21:15.340 | it shows just like Chachapiti beginning to run away with it,

00:21:19.140 | Gemini, Meta AI, Claude really are not even keeping up.

00:21:24.140 | So just if you set aside valuation for a second here, Bill,

00:21:30.620 | why do you think the game on the field

00:21:32.820 | as to consumer has changed so much?

00:21:36.380 | Is Chachapiti now in a flywheel?

00:21:40.380 | Have they broken out of the pack?

00:21:41.980 | Do you think they're gonna be the winner in consumer AI?

00:21:46.220 | A couple of things that we've talked about in the past.

00:21:49.980 | So I think everything you said is true.

00:21:52.860 | I do think voice and memory

00:21:55.180 | are areas where you could really run.

00:21:59.060 | And so people are super excited about advanced voice.

00:22:02.940 | People that have it, love it.

00:22:04.380 | Especially when I'm driving in a car,

00:22:08.860 | I'll have long conversations with Chachapiti.

00:22:12.620 | And if the advanced mode makes that even easier,

00:22:16.020 | I think that's very, very reinforcing.

00:22:18.700 | And anyone that wants to compete

00:22:20.940 | would need to catch up on that front fast.

00:22:24.060 | And then the second thing is memory.

00:22:26.020 | And I would just say,

00:22:29.100 | based on all the tools that are out there,

00:22:31.140 | they appear to be experimenting with it more than others.

00:22:34.540 | And we've talked about this over and over,

00:22:36.700 | but you can go into Chachapiti

00:22:38.500 | and look at what it's remembered on your behalf.

00:22:41.860 | I think that's big, really big,

00:22:44.940 | which would be another vector for them to break through on.

00:22:48.180 | You know, another one I feel like is worth mentioning

00:22:51.820 | is Sam Altman just continues to do extraordinary things.

00:22:56.820 | Like he's just surviving the whole board thing

00:23:02.540 | was something most humans couldn't do.

00:23:05.860 | He seems to have remarkable touch in Washington and access,

00:23:11.020 | which regulation appears to be coming at us fast and furious.

00:23:15.780 | And I've often said that could be used

00:23:18.900 | to help reinforce lock-in.

00:23:20.700 | And him having that access and control is super valuable.

00:23:24.940 | And we continue to just hear about new initiatives

00:23:28.260 | or new programs.

00:23:29.380 | You know, he's traveling around the globe.

00:23:31.060 | He's got everyone's ear.

00:23:32.580 | And he appears to be remarkably ambitious

00:23:35.500 | and successful at what he's trying to convey

00:23:39.500 | and talking to people into doing things

00:23:41.500 | they wouldn't do for any other partner, you know?

00:23:44.060 | - Yeah, it's pretty extraordinary, the pace and velocity.

00:23:47.980 | And frankly, we see that on the team side as well.

00:23:51.100 | Just extraordinary team.

00:23:52.380 | The best people continue to appear to go there.

00:23:55.500 | You know, everybody asked me a year ago,

00:23:57.500 | they said, "Oh, Gemini's coming.

00:23:58.940 | Gemini's gonna go first."

00:24:00.060 | But the fact of the matter is,

00:24:01.180 | right through the chaos of the moment, Bill,

00:24:03.980 | everybody's responding to them.

00:24:05.380 | They launched Strawberry in '01 preview first.

00:24:07.700 | They launched Advanced Voice first.

00:24:09.900 | So you gotta give them some credit for that.

00:24:11.620 | But when it comes to 200 million weekly miles

00:24:15.020 | that's reported out there,

00:24:16.140 | I would also make this argument, you know,

00:24:17.980 | and the team's made a chart on this,

00:24:19.700 | and I'm really curious of your thoughts.

00:24:23.180 | So it appears to me that open AI is seeing

00:24:26.300 | more and more of a network effect

00:24:28.460 | as well as scale advantages, right?

00:24:30.780 | So, you know, you've talked a lot about network effects,

00:24:34.660 | but there's been debate as to whether or not

00:24:36.340 | they exist here.

00:24:37.500 | But here's the argument that I would make.

00:24:39.100 | On the network effects side,

00:24:41.340 | it seems like more users is leading to better data,

00:24:45.940 | the data coming from the interactions with those users.

00:24:49.300 | And that's leading to better models and cheaper models,

00:24:53.260 | because you can do more of the work in post-training,

00:24:56.180 | which then leads to more users, right?

00:24:59.460 | And so, you know, here's the chart that we made on it.

00:25:03.860 | Do you buy the network effects argument, that flywheel?

00:25:08.460 | Because if that's in place,

00:25:11.580 | then it, to me, explains why we're seeing them

00:25:14.340 | break away from the pack when it comes to consumer AI.

00:25:17.780 | - Yeah, I don't know.

00:25:19.420 | I mean, I hate for that to be my answer,

00:25:21.180 | but I don't know the material impact

00:25:25.700 | of the users translating into data.

00:25:29.620 | You've likely seen, I've seen, you know,

00:25:31.860 | occasionally maybe one in 20 prompts that I do into open AI

00:25:36.540 | I'll get two results, and it'll ask me

00:25:38.460 | which one I like better.

00:25:39.620 | So that's the kind of thing you're talking about.

00:25:42.340 | And I just don't know if that makes the model 10% better,

00:25:46.180 | 20, or 50, or 100.

00:25:47.980 | There's certainly data that suggests the other models

00:25:52.100 | are right on their heels if you only look at test scores

00:25:55.860 | and, you know, the type of benchmarks.

00:25:58.140 | So that would suggest this isn't true.

00:26:01.060 | You know, the memory side, you know,

00:26:03.580 | the switching costs go through the roof

00:26:05.700 | if you get that right.

00:26:06.700 | - For sure, for sure.

00:26:07.980 | - Now, the one thing we've talked about

00:26:10.340 | that is also at play is that the memory

00:26:13.980 | you'd really like to have is on email and chat

00:26:17.500 | and all the data sources that already exist in your life.

00:26:21.940 | And how open AI would get inside of those systems

00:26:26.660 | is less clear to me.

00:26:28.540 | - Yep.

00:26:29.500 | - It's not impossible, it's just less clear.

00:26:31.780 | And that's where Microsoft and Google have some advantages

00:26:34.780 | and maybe Apple as well.

00:26:36.540 | And so it'll be fun to watch that fight

00:26:40.180 | and how those sources of data,

00:26:42.820 | because I think that's where you get the real lock-in.

00:26:45.780 | If I have an AI partner where I can simply say,

00:26:50.300 | "Who did I send that email to?"

00:26:52.220 | Like, that's really, really powerful.

00:26:55.020 | And I think the switching costs are insurmountable

00:26:58.420 | if someone gets to that place first.

00:27:00.540 | - The data on the field,

00:27:01.660 | and I'm just looking at the data on the field,

00:27:03.740 | I'm looking at the number of users and meta AI, et cetera.

00:27:07.420 | It looks to me like among the new consumer entrants,

00:27:10.460 | and I like the guys a lot at Perplexity, as you know,

00:27:13.580 | and I know lots of people like that product,

00:27:15.980 | but just from a usage perspective,

00:27:18.060 | among the new entrants who have the capital,

00:27:20.780 | who have the surface area to compete,

00:27:22.740 | it looks to me like Chat GPT has now clearly broken away.

00:27:27.380 | And there's gonna be a game between them.

00:27:29.580 | Meta, I think, is probably in the second best position.

00:27:31.980 | Google probably can't be underestimated.

00:27:34.900 | Obviously, Satya has consumer co-pilot with Mustafa there,

00:27:38.700 | but it's really interesting to see that game.

00:27:40.940 | The other vector here that's interesting is,

00:27:43.220 | you know, there's a lot of rumors out there

00:27:45.460 | about open AI's revenue, Bill, you know,

00:27:49.220 | and we'll post some of these, you know,

00:27:50.900 | four or five billion in revenue-

00:27:52.460 | - That's what I've heard, yeah.

00:27:54.020 | - Growing at over 100% a year.

00:27:55.820 | So I'm just intrigued by that.

00:27:58.580 | If that trajectory were to continue, right,

00:28:01.380 | that'd give you like roughly 10 billion next year.

00:28:04.820 | And the round's rumored to be at $150 billion.

00:28:08.460 | So that's about 15 times forward revenue.

00:28:11.900 | So I asked the team two things.

00:28:13.340 | I asked them to compare that to other companies,

00:28:16.700 | namely Google and Meta,

00:28:18.700 | both in terms of the pace to get to five billion in revenue

00:28:22.660 | and the valuations once they got there.

00:28:25.460 | And so this first chart just shows

00:28:27.820 | that open AI was able to get there roughly in, you know,

00:28:32.260 | in two years from the launch of the chat GPT

00:28:35.260 | in November of '22.

00:28:36.940 | You know, it took Google about two or three more years

00:28:40.020 | than that to get there.

00:28:41.060 | It took Meta almost six or seven years to get there.

00:28:45.100 | And what was interesting,

00:28:46.660 | so they got there a lot faster to five billion.

00:28:49.020 | We can agree on that.

00:28:51.100 | But then I asked,

00:28:52.260 | what were the multiples at that point in time?

00:28:55.460 | Because I remember when I bought the Google IPO,

00:28:58.060 | everybody said it's overpriced.

00:28:59.820 | When Microsoft invested in Meta,

00:29:01.540 | everybody said it's overpriced.

00:29:02.900 | But what's interesting is Google IPO-ed in 2004

00:29:05.900 | at about 10 times forward revenue, right?

00:29:08.660 | Microsoft invested in Meta at 2007

00:29:11.340 | at about 50 times revenue.

00:29:13.700 | And then Meta IPO-ed in 2012 at about 13 times revenue.

00:29:18.060 | And now again, if all these rumors are correct,

00:29:20.020 | you're talking 15 times revenue.

00:29:22.140 | So it's basically in a valuation zip code

00:29:25.980 | that is, you know, again,

00:29:27.500 | if you accept the trajectory

00:29:30.140 | that's similar to those other companies.

00:29:32.620 | And I know you have some real thoughts about margin here

00:29:35.500 | and whether or not the quality of those revenues

00:29:37.580 | are the same.

00:29:38.420 | So I thought I'd just throw that out there and ask you.

00:29:41.460 | Yeah, well, look, I think your analysis is exactly correct.

00:29:46.340 | And the only area of risk is what you just said.

00:29:49.460 | And I wrote a blog post years ago

00:29:51.740 | called "All Revenue Is Not Created Equal,"

00:29:54.140 | which we could put a link in for people wanna look at.

00:29:57.260 | But I think the one question I would have in this case,

00:30:01.060 | which is a data point I don't have, is gross margin.

00:30:04.620 | And there, everything we talked about,

00:30:08.300 | you know, the high cost of maybe the GPU usage

00:30:11.860 | to get advanced voice, right?

00:30:13.380 | Like there's a chance that open AI

00:30:17.940 | or anyone else in this field's gross margins

00:30:20.220 | are more in the 10 or 20% range

00:30:22.620 | versus the 57 and 81 that you have here in your charts.

00:30:26.580 | And that would be the one thing that might trip it up.

00:30:29.180 | And how those scale over time is tied exactly

00:30:32.740 | to all the things that we just talked about

00:30:34.740 | in the pricing model and the business model.

00:30:37.380 | So yeah, I think you could come to the conclusion

00:30:40.900 | you just made, but still have exposure in this one area.

00:30:45.060 | Yeah, and I think, you know,

00:30:46.420 | I think it's such an important point to make, right?

00:30:49.540 | Like when you're investing in a company,

00:30:51.540 | you gotta get the top line.

00:30:54.300 | Like who's gonna be the winner?

00:30:55.740 | And then you need to be able to forecast that top line,

00:30:58.140 | but that's not ultimately what drives valuation, right?

00:31:01.180 | What drives valuation as we've often talked about here

00:31:04.580 | is the future cash flows that those revenues can produce.

00:31:07.420 | And there's a real question on the table here

00:31:09.700 | that you've articulated well, which is, you know,

00:31:12.820 | is there going to be a layer, a tax here, right?

00:31:16.460 | That Nvidia and the cost of inference

00:31:19.580 | and the cost of training imposes in perpetuity

00:31:23.060 | on these companies such that it's going

00:31:26.180 | to always be a less profitable business

00:31:29.020 | than a Meta or Google on the consumer side,

00:31:32.580 | or an AWS and, you know, and a Google Cloud

00:31:37.380 | or Microsoft Azure on the enterprise side.

00:31:41.060 | I certainly think you're right

00:31:43.540 | that at the start of the super cycle,

00:31:46.380 | like if you go back to the start of AWS,

00:31:48.820 | remember the debates then, Bill, in 2009, 2010,

00:31:52.180 | can it ever make money?

00:31:53.580 | Can it ever make money, right?

00:31:55.380 | Because they had to get to massive scale

00:31:58.660 | and the cost of delivering that scale had to come down.

00:32:02.780 | So you're betting on two things, I think,

00:32:05.020 | with OpenAI with respect to margins.

00:32:07.460 | The first thing you're betting on is that they can get

00:32:09.540 | to scale because this is clearly a scale business.

00:32:12.820 | The second thing you're betting on is you have to believe

00:32:15.980 | that the cost of inference is going to come down

00:32:19.020 | meaningfully over time, and that the cost of training

00:32:22.300 | will come down meaningfully over time.

00:32:23.940 | Now, we already know the cost of inference has come down

00:32:26.540 | by over 90% over the course of the last 18 months.

00:32:30.780 | And our friend, Sonny, over the weekend,

00:32:32.340 | when we were in our group chat, you know,

00:32:34.180 | said he expects it to come down by another 90%

00:32:37.460 | over the course of the next several years.

00:32:39.620 | So, but those are the types of things

00:32:41.540 | you're going to need to be true

00:32:43.100 | in order to have a margin structure that is consistent

00:32:47.420 | with those legendary businesses like Google and Meta.

00:32:51.380 | - Yeah, and yeah, to be fair,

00:32:53.820 | your analysis had Google and Meta,

00:32:55.500 | but when you make AWS, the comparison that suggests

00:32:59.900 | that Amazon might be a better proxy,

00:33:03.580 | which, you know, trades at three times.

00:33:06.220 | - Certainly for the enterprise side of OpenAI's business,

00:33:08.940 | I think the comp would much more be AWS,

00:33:11.940 | but on the consumer side of their business,

00:33:13.900 | I think the comp is fairer to be somebody

00:33:15.900 | like Google or Meta, but in both instances,

00:33:19.420 | you have to assume that the cost of delivering, right?

00:33:22.260 | Let's be clear, AI and AI inference

00:33:26.500 | is a much, much more compute-intensive activity

00:33:29.900 | than retrieval, which was the business of search, right?

00:33:33.900 | And so, like, we just have to see technology

00:33:36.820 | ultimately drive that cost down,

00:33:38.580 | or there's going to be a higher tax

00:33:39.980 | and it will be a lower margin business.

00:33:41.620 | It doesn't mean that it won't be a great business

00:33:43.500 | or even a good return, but, you know,

00:33:45.860 | to achieve those margins, you got to see the cost of,

00:33:48.540 | you know, delivering the product go down.

00:33:50.780 | - You know, there've been statements along this journey

00:33:53.380 | from both Jensen and Sam Altman

00:33:56.300 | that the input to this thing is compute

00:33:59.820 | and you're going to need tokens.

00:34:01.980 | And those sound like variable cost inputs.

00:34:05.820 | That's, you know, and I guess in the worst case scenario,

00:34:09.140 | it's like an airline where fuel costs are just, you know,

00:34:12.980 | a big part of what drives the incremental profitability.

00:34:16.860 | And so the thing I don't know,

00:34:19.140 | and so I'm not suggesting this is absolutely true,

00:34:21.580 | is, is an AI business inherently a 20% margin business?

00:34:26.580 | You know, AWS is at 30 and Amazon Commerce was at five

00:34:33.700 | until they added, you know, whatever, added advertising,

00:34:36.820 | like, or can it be like Google or Meta?

00:34:39.940 | And I think until one of these things gets public

00:34:43.100 | and we can look at data a little more detail, we don't know.

00:34:47.180 | - Well, it's going to be interesting to watch it unfold,

00:34:49.740 | but, you know, there's certainly a related topic,

00:34:52.220 | one that I know you're amped up about

00:34:54.900 | and is probably even more amped up

00:34:57.340 | given this rumored open AI deal,

00:35:00.420 | is how the venture model is changing

00:35:03.460 | and whether these structural changes are good or bad,

00:35:06.700 | whether they're good or bad for LPs,

00:35:08.260 | whether they're good or bad for GPs and founders.

00:35:11.220 | So why don't you lead us in a discussion on that topic,

00:35:14.060 | you know, on the challenges to venture today?

00:35:19.660 | - Yeah, and two things that I would encourage people

00:35:22.980 | to check out, our friends at All In

00:35:25.940 | talked about this a little bit on their pod last Friday.

00:35:29.180 | And then at their conference,

00:35:30.900 | Thomas Lafon of Co2 had a long presentation

00:35:34.900 | that I think is where everyone worth looking at.

00:35:37.580 | They kind of sets this up,

00:35:38.940 | but despite our massive enthusiasm for AI

00:35:43.940 | and I'd say the entire community's enthusiasm,

00:35:49.100 | you know, we are at a seemingly problematic place

00:35:54.100 | in the venture capital industry

00:35:57.140 | with regard to how much cash is coming out of the system

00:36:00.900 | versus how much is going in.

00:36:02.940 | And everyone seems to be hyper aware

00:36:06.700 | that it's a historically low number of IPOs.

00:36:11.540 | It's like single digits where the average,

00:36:14.140 | even in subpar years has been closer to 70 or 80.

00:36:18.540 | M&A has had quite a hiatus,

00:36:23.220 | partially driven by the restrictions on the Magnificent 7,

00:36:27.780 | although they're finding ways around that.

00:36:29.900 | And so, you know, what's going wrong?

00:36:32.340 | You know, the capital markets seem to be doing just fine

00:36:37.100 | in terms of how well the S&P is performing.

00:36:40.380 | And so why isn't this happening?

00:36:42.140 | And so I would offer a couple of thoughts.

00:36:45.460 | One, I think everyone now believes in power laws,

00:36:49.060 | network effects, scaling laws, that kind of thing.

00:36:51.460 | I think all the investors do.

00:36:53.140 | When I entered the venture industry,

00:36:55.420 | I think it was a competitive advantage to believe in them

00:36:59.620 | when people didn't and you could, you know,

00:37:03.220 | find a way to take advantage of that and make money.

00:37:06.300 | But today I think everyone gets it.

00:37:08.740 | And so the other thing that's happened, I believe,

00:37:12.260 | is many of investors have decided late stage investing

00:37:16.540 | is better than early stage.

00:37:18.180 | And I'm primarily respond, you know,

00:37:20.740 | talking about the venture firms that have gone

00:37:23.700 | from being mostly early stage venture,

00:37:26.140 | traditional venture to having 10 billion or more, you know,

00:37:30.500 | AUM and willing to write checks

00:37:32.820 | in the hundreds of millions of dollars,

00:37:34.660 | which didn't happen a decade ago, right?

00:37:37.580 | And for those type people, you know,

00:37:41.020 | the management fee is on a much bigger, you know,

00:37:43.540 | you get the same percentage on whether you're deploying it

00:37:46.700 | at $5 million a piece or $200 million a piece,

00:37:50.940 | and you get way more dollars deployed.

00:37:53.300 | You don't take board seats, so the work's less.

00:37:55.500 | And the fees are massively bigger.

00:37:58.300 | And I think that for reasons that are just competitive,

00:38:02.260 | our whole world's kind of felt this gravity

00:38:07.020 | pulling them to that place.

00:38:08.900 | And despite the fact that we had this mini correction,

00:38:12.860 | I call it mini 'cause that's what it feels like

00:38:14.700 | now that AI kind of just brought the sunlight out again,

00:38:18.460 | these firms have not had problems raising those dollars.

00:38:21.740 | And so despite the fact that a large number

00:38:25.180 | of the unicorns are still private

00:38:27.460 | from the previous investment cycle,

00:38:30.100 | this kind of behavior continues.

00:38:32.460 | And all of these firms wanna be in the hottest deals.

00:38:35.580 | You know this, you're on this field,

00:38:36.980 | you're a participant in this world.

00:38:39.460 | If there's an interesting company out there,

00:38:41.940 | it's very likely that they're gonna be approached

00:38:45.180 | preemptively and told to take more money.

00:38:48.220 | And so I think you're gonna, until this change,

00:38:51.780 | I think you're gonna have very few companies

00:38:53.860 | that are considered to be doing well

00:38:57.020 | that aren't asked by the industry

00:38:59.860 | to raise $500 million or more.

00:39:02.580 | And that in and of itself is a very unusual,

00:39:07.580 | compared to the traditional venture model from years ago,

00:39:12.380 | it's just super unusual because, I'm almost done,

00:39:15.860 | because this thing is so competitive

00:39:19.940 | and everyone is trying to get in the hottest deals.

00:39:24.060 | The best way to achieve that is to be founder friendly.

00:39:27.260 | And I think we talked about the profile that Thrive had,

00:39:31.660 | I think on the cover of Fortune.

00:39:33.100 | I would encourage people to read that 'cause that's,

00:39:35.580 | it was almost, I would call it almost PR perfection

00:39:39.500 | for Josh and his team in terms of coming,

00:39:43.100 | people vouching for them being founder friendly.

00:39:47.020 | And if you're gonna be founder friendly

00:39:49.340 | and write big checks, guess what?

00:39:50.900 | You're gonna be supportive of founder secondary

00:39:53.580 | and you're gonna be supportive

00:39:56.300 | of broad-based employee secondary.

00:39:58.740 | When you do those things, you are taking away

00:40:03.020 | probably the strongest motivating factor

00:40:07.060 | that pushed people to go public,

00:40:08.780 | that pushed founders and their teams to wanna be public,

00:40:11.860 | which was liquidity.

00:40:13.260 | And so with that off the table, for me,

00:40:15.700 | there's no surprise that IPOs are happening

00:40:18.500 | or not happening or not happening

00:40:20.340 | because there's no incentive for them to go out.

00:40:24.740 | And it's a weird place for me

00:40:28.180 | when I look at the venture industry writ large,

00:40:30.660 | which is what's gonna drive people to go public?

00:40:33.580 | How do large institutions get liquidity?

00:40:36.340 | I don't think large institutions can realistically

00:40:39.740 | get liquidity to the secondary market,

00:40:42.140 | certainly not at a good price.

00:40:44.260 | But my biggest, in addition to all those things,

00:40:48.500 | one thing I would raise provocatively is,

00:40:53.380 | does overfeeding these companies with cash

00:40:56.340 | lead to non-optimal execution?

00:40:58.860 | And you and I were deeply involved in the Uber situation,

00:41:03.860 | but when you start losing a billion dollars a year,

00:41:07.660 | or even I would say $20 million a month,

00:41:11.380 | you're very far away from profitability.

00:41:15.020 | And we talk a lot about focus and constraints

00:41:19.140 | and how that leads to better decision-making.

00:41:21.540 | That's hard to do when you're spending 20 million a month.

00:41:24.420 | And the other thing that's hard to do

00:41:26.740 | is raise $500 million and not spend it.

00:41:29.820 | And so I do propose the question

00:41:33.060 | that maybe one of the things that's a problem

00:41:37.300 | with that previous generation of unicorns

00:41:39.980 | is they were overfed.

00:41:42.740 | There's a picture you can look up

00:41:44.660 | that's kind of disgusting, so people may not want to,

00:41:47.060 | but there's this thing called a gavage tube,

00:41:50.500 | which is what they use to make foie gras.

00:41:53.180 | It's how they force feed the geese

00:41:55.820 | to get 'em just super fat.

00:41:58.140 | And that's the image I have in my mind.

00:42:00.620 | Like, are we overfeeding these startups?

00:42:03.820 | And then they get so far away from profitability,

00:42:07.180 | they're spending on projects

00:42:09.420 | that if they were trying to get to profitability,

00:42:11.540 | they wouldn't spend on that are lower return.

00:42:14.060 | And then maybe they get stuck.

00:42:17.260 | This is my last point.

00:42:19.140 | When I entered the venture business,

00:42:21.060 | one of the things I thought was an advantage I had

00:42:23.500 | coming from Wall Street is I knew what Wall Street wanted.

00:42:26.860 | They were the customer for the venture capital company

00:42:30.700 | that would eventually IPO and trade in their markets.

00:42:34.100 | And there's an interesting dichotomy right now.

00:42:37.340 | If you look at the public markets,

00:42:39.900 | which have become much more sane,

00:42:41.940 | you would know this better than me,

00:42:43.860 | relative to where they were three years ago,

00:42:46.060 | there's a high expectation for profitability.

00:42:51.700 | And so I think there's this incredible mismatch

00:42:55.180 | between what Wall Street wants to see

00:42:58.660 | and the state that a company is forced to be in

00:43:04.340 | as a result of this hyper competitive investment market.

00:43:08.180 | Yeah, no, I mean, listen,

00:43:09.660 | I think that it's a great analysis

00:43:12.660 | and framing really of the issue.

00:43:16.260 | And you bring up a lot of great questions

00:43:18.020 | and I think the right concerns.

00:43:19.100 | So let's try to break them down

00:43:20.260 | because I think they fall into roughly like three buckets.

00:43:24.420 | Let's start first with the question,

00:43:25.860 | just some more dollars, bigger funds,

00:43:27.540 | more competition and higher valuations.

00:43:30.420 | There's no doubt that VC has grown

00:43:33.540 | from 100 billion 10 years ago

00:43:35.260 | to 300 billion over the last 10 years.

00:43:38.420 | If you set aside the COVID period

00:43:40.780 | where we all know because of ZURP,

00:43:43.500 | public markets lost their minds,

00:43:45.020 | venture markets went to high levels,

00:43:47.980 | we're back to kind of this $300 billion level,

00:43:50.900 | which was pre-COVID.

00:43:52.460 | And so while we call all of this venture bill,

00:43:55.860 | one of the big differences I have here,

00:43:58.580 | and I've said many times,

00:44:00.540 | is that the venture market really hasn't grown that much.

00:44:04.220 | Much of the investment that we're counting as venture,

00:44:07.780 | when you look at all these data sources that we pull

00:44:10.540 | is into companies that are higher

00:44:12.460 | than $10 billion in valuation,

00:44:14.700 | with huge revenues that would have been before

00:44:17.540 | captured by public market investors.

00:44:20.020 | And so I think it's important as an industry

00:44:21.940 | that we start thinking about these things as different.

00:44:25.820 | I call them internally and to our LPs,

00:44:28.260 | I call those quasi-public companies,

00:44:30.860 | companies like Databricks, Stripe, OpenAI.

00:44:34.500 | I think it's silly to call them venture at this stage

00:44:36.900 | when they have 5 billion in revenues growing 100% a year.

00:44:40.580 | So yes, I would say this,

00:44:43.300 | the late stage quasi-public market is much more competitive,

00:44:48.300 | just like the public market is more competitive,

00:44:51.700 | because it leads to better price discovery,

00:44:54.580 | but it also means that there's less arbitrage

00:44:57.820 | and returns are more dependent upon long-term compounding

00:45:01.740 | than some misinformation in the market, right?

00:45:04.780 | And as we saw in '20 and '21,

00:45:06.860 | the public market's corrected.

00:45:09.300 | In fact, a lot of the IPOs that happened during that period

00:45:12.660 | are still down over 50%, right?

00:45:16.100 | So the public market's corrected, just did it quicker.

00:45:20.020 | In the quasi-public markets,

00:45:21.940 | we've seen a lot of these companies shut down,

00:45:24.660 | sell, and still down over 50% from the high.

00:45:28.940 | So I don't think there's a lot of difference there.

00:45:31.220 | And when I look at the early stage venture markets, Bill,

00:45:34.540 | I would agree with you,

00:45:35.820 | there's a lot of excitement around AI,

00:45:37.660 | but outside of AI, you look at Series A follow-on rounds

00:45:41.620 | or Series B follow-on rounds,

00:45:43.340 | I mean, they're down dramatically.

00:45:45.100 | If you look at the number of first-time funds

00:45:47.380 | that are getting funded as second-time funds,

00:45:49.660 | those are down dramatically.

00:45:51.460 | So I see a lot of reversion to the mean happening

00:45:54.820 | rather than structural change.

00:45:56.300 | I think the big structural change

00:45:58.620 | that this data leads me to conclude

00:46:00.980 | is that because of the regulatory burdens of going public,

00:46:04.740 | because of the change in Silicon Valley around sentiment,

00:46:08.100 | because of the ability to get liquid

00:46:10.060 | in secondary transactions,

00:46:11.980 | because of the liquidity

00:46:13.380 | of this late-stage quasi-public market

00:46:15.580 | where institutions like Altimeter

00:46:17.700 | or the CO2s of the world or Fidelity's or Thrive's

00:46:21.780 | or whatever that we're here to provide that liquidity,

00:46:25.260 | I agree with you that there's a lot more money there

00:46:28.940 | because those companies are choosing to stay private,

00:46:31.340 | but we should think about these

00:46:32.820 | and compare them to their public company competitors,

00:46:36.100 | not to what's happening in the Series A market.

00:46:39.420 | Two things I would highlight.

00:46:40.340 | One, when you call it quasi-public,

00:46:43.940 | I think you're talking about it

00:46:46.060 | primarily from a input point of view.

00:46:49.020 | In other words, it looks public relative

00:46:52.300 | to how Altimeter would invest

00:46:54.780 | or other late-stage players would invest.

00:46:57.740 | But, and maybe this is where the word quasi comes in,

00:47:02.380 | but if you think about it from an output perspective,

00:47:05.980 | there's no liquidity for anybody.

00:47:08.460 | So you've taken a portion of the market

00:47:11.580 | that used to serve multiple purposes

00:47:15.060 | and now one of those purposes is gone.

00:47:17.580 | If we're saying there's a permanent shift

00:47:20.780 | from one place to the other.

00:47:21.860 | Two, there are these regulatory things that come up

00:47:25.100 | because many people believe that one of the SEC's goal

00:47:29.460 | is to make sure all investors can participate in that.

00:47:32.260 | Now, that's off the table.

00:47:33.780 | I mean, listen, listen.

00:47:35.220 | I've said publicly many, many times,

00:47:37.900 | we should get these companies public faster.

00:47:40.500 | I wanna see Stripe public.

00:47:41.900 | I wanna see Databricks public.

00:47:43.340 | I think it's better for the companies.

00:47:44.940 | I think it's better for the investing public writ large.

00:47:48.540 | I'm just trying to explain the game on the field and-

00:47:51.460 | - I understand.

00:47:52.700 | - Go ahead.

00:47:53.540 | - And that's the third point I was gonna make is

00:47:55.620 | quasi public isn't the same as public

00:47:58.740 | in terms of how the public markets might shape

00:48:03.180 | the motivation and execution of the team.

00:48:06.660 | - 100%.

00:48:07.500 | - We go back to the meta example where they went public,

00:48:10.420 | stock went down, Wall Street says,

00:48:12.380 | "You're not ready for mobile."

00:48:14.060 | You know, and Zuck later said,

00:48:15.260 | "Shit, that actually kicked me into gear."

00:48:18.180 | Those things don't exist when this isn't here.

00:48:21.340 | And I agree with you.

00:48:22.580 | You know, this is playing the game on the field

00:48:24.780 | and I'm not blaming anyone.

00:48:26.460 | I'm just highlighting this is where we've matriculated to.

00:48:30.020 | You know, this is where we stand today.

00:48:31.940 | - I think it's right.

00:48:32.820 | So let's move to the second big point,

00:48:34.380 | which is this liquidity IPOs.

00:48:36.100 | And I think it is true the number of IPOs has been anemic

00:48:40.540 | and the exit amount is now in venture

00:48:42.820 | is now at about $100 billion a year.

00:48:45.540 | While down a ton from the $700 billion peak

00:48:50.300 | in Zerp in 2021, you know,

00:48:53.100 | which was really a one-time COVID high,

00:48:54.980 | we're basically back at the same exit level

00:48:58.020 | for venture that we were pre-COVID.

00:49:00.780 | Lots of people talk about the zombie corns, right?

00:49:03.020 | We have a thousand companies that were unicorns.

00:49:06.260 | A lot of those will never get back there.

00:49:07.860 | I've said 80% of those companies will never get back there.

00:49:11.180 | Those companies need to get, you know,

00:49:13.060 | merged into other companies, need to get sold,

00:49:15.260 | need to get shut down, whatever the case may be,

00:49:17.500 | or do down round IPOs like Instacart did,

00:49:20.100 | which is now off to the races under some great leadership.

00:49:24.420 | But I look at just our pipeline build,

00:49:26.540 | just to give you a counter example here.

00:49:29.660 | I think we may have four IPOs in the pipeline

00:49:32.020 | in the next four to six months.

00:49:33.860 | The rumors out there around companies like Cerebris

00:49:36.260 | and CoreWeave and Databricks,

00:49:37.660 | those are all in our portfolios.

00:49:39.500 | And on top of that, you know, we recently sold Tabular

00:49:42.380 | to Databricks, you know,

00:49:44.020 | that had a price rumored to be $2 billion.

00:49:46.340 | So exits are increasing.

00:49:48.340 | Interest rates are coming down,

00:49:49.940 | which I think will lead to more of that.

00:49:51.740 | The world is healing.

00:49:53.460 | And so again, I'm not so sure this is,

00:49:55.900 | I do think there are some things that are structural.

00:49:58.180 | The regulatory things are structural.

00:50:00.300 | The more dollars in quasi-public is structural,

00:50:03.340 | but there is still enough incentive.

00:50:05.660 | These companies want and have boards

00:50:07.820 | that will get them public.

00:50:08.860 | I just think they may come public a lot longer

00:50:12.420 | when they're at 10 or $50 billion valuations,

00:50:15.060 | rather than they're at a $2 billion valuation.

00:50:18.500 | But, you know, as far as the companies I'm involved with,

00:50:20.780 | I'm pushing for them to come public sooner,

00:50:23.260 | or at least when they're ready.

00:50:24.980 | And I think a lot of that pressure is off.

00:50:28.140 | And I think the number of board members

00:50:31.820 | that are actually willing to push for that,

00:50:34.020 | and Chamath talks about this a lot,

00:50:35.540 | I think is actually few.

00:50:36.980 | You might be one of a few,

00:50:38.780 | but I think most of them don't

00:50:41.500 | because they've been trained to applaud,

00:50:44.860 | and that's what they do.

00:50:46.820 | Well, I mean-

00:50:47.660 | And it's partially just driven by competition.

00:50:49.980 | Once again, I'm not,

00:50:51.420 | I'm trying to give you my best view of-

00:50:53.500 | Yeah, no, I think it's an important,

00:50:55.460 | it's a super important conversation.

00:50:57.940 | And by the way, you talk,

00:50:59.540 | if 80% of the zombie corns are never gonna get out

00:51:02.340 | or are gonna get out at reduced prices,

00:51:04.180 | I'm telling you flat out,

00:51:06.860 | those are being held on the large endowments as LPs

00:51:11.860 | at unrealistically high prices across the board.

00:51:16.940 | I would say yes and no.

00:51:18.220 | So let me just give you a couple of different examples.

00:51:20.860 | I mentioned we have a lot of good things in the portfolio,

00:51:23.060 | but we had a company that wasn't performing

00:51:25.300 | at the levels that it previously had been.

00:51:28.420 | It had been priced at many billions of dollars

00:51:31.340 | at the peak of ZERP in 2021, a company called Lacework.

00:51:34.180 | And we ultimately,

00:51:35.420 | we pushed to sell that company to Fortinet

00:51:37.500 | while we still had hundreds of millions of dollars

00:51:39.660 | of cash on the balance sheet.

00:51:40.740 | It's a great acquisition for that company,

00:51:43.740 | but it's out of the system.

00:51:45.020 | It's marked.

00:51:45.860 | We've distributed the cash to our shareholders.

00:51:47.740 | So we're distributing, okay?

00:51:49.660 | So I'm just saying that this is happening.

00:51:52.620 | And we have other companies in those portfolios

00:51:54.900 | that were marked really high in '21, right?

00:51:57.100 | One is company you and I are both invested in, Clickhouse,

00:52:01.020 | which is growing through those valuations, right?

00:52:03.420 | Or a company like Sigma Computing that we're in,

00:52:05.580 | which is growing through those high valuations.

00:52:07.580 | So you just have to break it down

00:52:09.780 | and look at these one by one.

00:52:11.540 | I think that there are definitely things in there

00:52:14.260 | that are held at too high a valuation,

00:52:16.340 | and LPs should scrutinize that.

00:52:18.100 | But there are other things that are healing.

00:52:19.460 | Let's just, I really want to talk about this question

00:52:21.700 | that I think is the most important one, Bill.

00:52:24.180 | Great.

00:52:25.020 | Which is, does excess capital lead to companies

00:52:28.780 | being overfed, which leads to poorer outcomes

00:52:32.820 | for innovation?

00:52:33.660 | Because I think the potential for that,

00:52:35.780 | like, at the end of the day,

00:52:37.540 | that would be the worst thing, right?

00:52:39.500 | And I think you and I have a lot of shared belief

00:52:43.220 | that too much capital does ruin corporate culture.

00:52:47.380 | It does lead to higher burn rates.

00:52:49.260 | It does lead to lower financial returns.

00:52:51.580 | And it does lead to less innovation, right?

00:52:54.620 | I pounded the table on this topic,

00:52:56.340 | you know, the time to get fit.

00:52:58.180 | And I applaud Mark Zuckerberg.

00:53:01.300 | I mean, him stepping out in February of 2023

00:53:05.580 | and writing a letter called the year of efficiency.

00:53:08.380 | And he said, we started doing this,

00:53:11.060 | just thinking that it was about getting back to the office.

00:53:13.780 | But what we discovered was that smaller is better, right?

00:53:18.780 | He said, flatter is faster and leaner is better.

00:53:22.540 | And what he meant is the cycle time on innovation,

00:53:25.540 | de-layering the organization,

00:53:27.500 | getting rid of layers of VPs, right?

00:53:30.940 | Like really getting the organization tight and fit

00:53:33.780 | was better for the future growth and future profitability

00:53:38.780 | and future innovation of the business.

00:53:40.540 | So I tweeted over the weekend,

00:53:43.140 | it's great news that I saw that Zoom

00:53:46.140 | and Salesforce and Workdays,

00:53:47.860 | they're starting to get sober about stock-based compensation.

00:53:50.700 | That is another component of too much capital,

00:53:55.220 | too many people leads to excessive SBC,

00:53:58.740 | which you and I have talked a lot about.

00:54:00.780 | So I think just because you raise a lot of money

00:54:03.860 | doesn't necessarily mean you're unfit, right?

00:54:06.940 | Remember, OpenAI is not really a VC company at this stage.

00:54:11.580 | Google went public on 2 billion of trailing revenues.

00:54:15.140 | These guys are rumored to have 5 billion already.

00:54:18.140 | So 3X that amount.

00:54:20.500 | And we've talked about building AI

00:54:22.860 | is just a lot more expensive

00:54:24.700 | than building the things that came before it.

00:54:26.940 | So I just think we need to have an apples

00:54:28.900 | for apples comparison.

00:54:29.980 | But if you want a champion to stand with you

00:54:32.180 | on this issue of companies raising too much,

00:54:36.420 | spending too much, they gotta be really careful.

00:54:39.700 | And so I gave some advice

00:54:41.580 | to one of our fellow founders the other day.

00:54:43.540 | He's got money coming in over the bow

00:54:46.340 | at a multi-billion dollar valuation.

00:54:48.420 | And he's like, "We don't need it.

00:54:49.980 | "We already have hundreds of millions on the balance sheet.

00:54:52.180 | "Should we raise it?"

00:54:53.580 | And I said, "Here are all the downsides of raising it.

00:54:56.700 | "You can't go raise another 500 million

00:54:58.980 | "and not have pressure from all your employees

00:55:01.300 | "for employees secondary,

00:55:02.980 | "for spending more money on more projects, et cetera.

00:55:05.620 | "And the NPV on those other activities will be lower.

00:55:08.980 | "And the incentive your employees have to stay with you

00:55:12.180 | "once they sell 10 or $20 million worth of stock

00:55:15.260 | "is gonna be less."

00:55:16.220 | And so finding that balance,

00:55:18.100 | I think is a critical function

00:55:20.300 | of leadership of these companies.

00:55:22.660 | - Yeah, and I would just say to that, Brad,

00:55:24.660 | I do think this is a huge dichotomy.

00:55:27.300 | Like for the, I do believe that the hyper competition

00:55:32.300 | in late stage market leads to incredibly large number

00:55:37.460 | of preemptive rounds where hundreds of million dollars

00:55:40.220 | are being force-fed to a company.

00:55:42.720 | And if you're spending 20 million a month,

00:55:45.660 | you're burning 240 million a year.

00:55:49.360 | If these companies in AI are 50% gross margin or whatever,

00:55:53.420 | that you gotta get revenue to twice that.

00:55:57.060 | You're right at that 500 million run rate

00:56:00.340 | before you could think about being profitable.

00:56:03.020 | And once you've gone to that place,

00:56:05.820 | which is ironically the same number, I think,

00:56:09.020 | where Philippe Lafont said,

00:56:10.300 | "You gotta be to go public these days."

00:56:12.560 | And the thing I would say to you, if that becomes true,

00:56:17.140 | if that kind of path dependency

00:56:20.340 | is cast upon every venture capital company that comes along,

00:56:26.500 | you're gonna end up with an excessive amount of zombie corns

00:56:30.380 | because previous to this evolution

00:56:35.380 | in the late stage markets, plenty of companies,

00:56:40.020 | either were bought at 300 million

00:56:41.880 | or went public at 250 or 500 million

00:56:45.540 | and created positive returns

00:56:48.180 | for their early venture capital investors.

00:56:51.060 | And if it's really 500 million in revenue

00:56:56.460 | or bust, I think there's gonna be a lot more bust

00:57:00.620 | than we've traditionally had.

00:57:01.600 | Yeah, no, net-net, I agree with you.

00:57:05.100 | VC is harder, it's more competitive,

00:57:06.820 | but I think the structural changes that we're seeing

00:57:10.020 | are more reversion to the mean than VC is forever bad.

00:57:14.380 | I think it's, you know, listen,

00:57:15.540 | this has always been a hard investment category.

00:57:18.100 | And I think if you're not backing a VC

00:57:20.500 | who has a right and a process

00:57:22.640 | that gets them into the top decile, right,

00:57:25.220 | then returns are not gonna be great.

00:57:26.940 | It's a power law business,

00:57:28.060 | always been a power law business.

00:57:29.700 | I think these are super important issues,

00:57:31.580 | but why don't we, just in the spirit of time-

00:57:33.420 | Wait, wait, wait, wait.

00:57:34.540 | Oh, here you go, here you go.

00:57:35.500 | I have to have one response.

00:57:37.100 | I would encourage our listener base to look up,

00:57:41.580 | there's a piece of research

00:57:43.420 | that's called the observer effect.

00:57:46.060 | And it started in physics, but it's used more broadly.

00:57:50.740 | And the observer effect is the idea

00:57:53.340 | that observing a phenomena or situation changes it.

00:57:57.260 | And that's really the point I'm making here

00:58:00.140 | is that I think prior to now,

00:58:02.500 | the investor and how they behave

00:58:05.580 | didn't actually impact the situation on the field

00:58:10.340 | in terms of like changing the game and how it's played.

00:58:15.340 | And to me, the way the competition that's evolved

00:58:20.580 | in the venture industry is actually perturbing

00:58:23.780 | and affecting the situation.

00:58:25.700 | So anyway, I'll leave it at that.

00:58:27.060 | No, I think it's another way of saying that

00:58:29.340 | is negative reflexivity, right?

00:58:31.980 | That when dollars come in,

00:58:33.540 | it actually leads to poor behavior.

00:58:35.500 | We'll take a deeper dive on that sometime

00:58:37.860 | because I also wanna get your thoughts

00:58:39.740 | on how that might change.

00:58:41.340 | But why don't we finish just in the spirit of time

00:58:43.660 | with a quick tech check like we always do?

00:58:46.940 | Awesome.

00:58:47.780 | So the big, big event,

00:58:49.220 | and this macro world is more you than me,

00:58:51.620 | but the big event obviously was the Fed decision

00:58:55.060 | to lower by 50 basis points.

00:58:57.740 | What's the impact from your point of view?

00:58:59.620 | Yeah, I mean, you and I talked a bunch about this

00:59:03.340 | along with our friends at All In.

00:59:05.740 | I mean, Goolsby had signaled this, right?

00:59:08.020 | That really we were in historically restrictive territory.

00:59:13.020 | And what that means is basically

00:59:14.580 | we just have a little bit of the emergency break

00:59:16.660 | in the economy.

00:59:17.820 | And so once they were convinced

00:59:20.340 | that inflation was gonna have a two-handle,

00:59:22.300 | which it now has,

00:59:23.740 | and that they were starting to see some slowdown

00:59:27.220 | in the jobs market,

00:59:28.420 | what we did is we wanna take a jumpstart

00:59:32.380 | to reducing the restrictiveness of the economy.

00:59:34.500 | I thought that was a smart thing to do.

00:59:36.500 | I think it's just getting back on side.

00:59:38.500 | I don't think they were seeing anything other

00:59:40.820 | than what we're seeing in the economy.

00:59:43.340 | But it's incredibly significant to the markets

00:59:46.020 | that we're now on our way down.

00:59:47.460 | Remember, we had two plus years

00:59:50.020 | of historically steep rise in interest rates

00:59:53.420 | coming out of COVID.

00:59:55.780 | And so this gives predictability

00:59:57.420 | as companies enter their budget cycles this year, right?

00:59:59.780 | Every company right now is thinking,

01:00:01.620 | what can I invest into AI infrastructure next year?

01:00:04.540 | And so knowing that interest rates are not going up

01:00:06.940 | is a super, super important input

01:00:10.220 | into those pieces of analysis.

01:00:13.820 | But the debate remains.

01:00:14.820 | I mean, I think it's interesting.

01:00:15.940 | You had Jamie Dimon and Gunlach come out and say,

01:00:19.020 | "Hey, this battle with inflation is not over,

01:00:21.260 | deglobalization, all these other things."

01:00:23.340 | So they're suggesting

01:00:24.380 | that you could actually see inflation kick back up.

01:00:26.900 | That would be a negative surprise.

01:00:28.700 | And others are saying

01:00:29.620 | that the Fed's already behind the curve

01:00:31.300 | and needs to go faster

01:00:33.580 | with respect to the slowing economy.

01:00:35.340 | So-

01:00:36.180 | - Where do you come out on that?

01:00:37.260 | - That's the debate that's in the market.

01:00:39.060 | Well, I'll just tell you where we are today.

01:00:42.300 | We talked about taking a bunch of units of risk off,

01:00:44.780 | which we did in June and July.

01:00:47.900 | And as we were heading into these rate cuts,

01:00:49.940 | we're back at average levels of exposure today.

01:00:53.060 | And that's because we saw a lot

01:00:54.620 | of really positive statements coming out

01:00:56.540 | of the earnings this summer.

01:00:58.180 | And the economic data continued to be constructed.

01:01:01.460 | We expected a 50 basis point rate cut,

01:01:03.620 | which I had been sharing with you

01:01:05.140 | and folks on this pod for some time.

01:01:07.860 | And so we think that's a good setup heading into the fall.

01:01:10.340 | Now we have an election we got to work our way through

01:01:13.060 | and we got a lot of other question marks,

01:01:14.980 | but you got to take those data points as you get them.

01:01:18.380 | - Your team is listening

01:01:20.060 | to so many different earnings calls and whatnot.

01:01:22.620 | What's your take on the consumer

01:01:26.260 | and that side of the demand equation,

01:01:30.500 | like irrespective of inflation,

01:01:34.780 | is there a mini recession

01:01:36.420 | or have we literally landed the soft landing

01:01:39.420 | or do we not know yet?

01:01:40.620 | - Listen, I think behind the scenes

01:01:42.020 | we've been going through many recessions

01:01:44.580 | in a bunch of different industries.

01:01:45.940 | Like for example, housing went through a mini recession.

01:01:48.740 | I think things like home renovation

01:01:52.060 | went through a mini recession.

01:01:54.020 | The entire supply chain about that happened.

01:01:56.940 | The S&P we talked about,

01:01:58.700 | the S&P has continued to outperform,

01:02:00.700 | but if you take out the 10 best performers from the S&P

01:02:03.540 | up until a couple of weeks ago,

01:02:04.740 | it was actually down on the year.

01:02:06.780 | So this has been a period of haves and have nots.

01:02:10.300 | I think the economy writ large is pretty stable,

01:02:13.740 | but let's just look at multiples here for a second,

01:02:16.100 | because I think for tech investors

01:02:18.020 | it's really what it comes down to.

01:02:19.500 | So this first chart just shows you

01:02:21.940 | that multiples for these big tech companies

01:02:25.180 | have come up quite a bit,

01:02:26.340 | certainly off of the January 23 lows.

01:02:29.100 | We were trading about 21 times, right?

01:02:31.660 | This is forward PE for these companies.

01:02:34.540 | - For the mag seven,

01:02:35.460 | you're showing forward PE for the mag seven.

01:02:37.540 | - Forward PE for the mag seven.

01:02:38.980 | So if you look at the 10-year average of this,

01:02:41.820 | it's about 25 times.

01:02:43.900 | In January 23, when we're all talking about

01:02:46.180 | Mike Wilson's hard landing,

01:02:47.820 | the economy's gonna crash, like all this stuff,

01:02:50.540 | people have post-traumatic stress from 2022,

01:02:53.060 | we got as low as 21 times.

01:02:54.860 | So now we've run up to 31 times on the forward PE.

01:02:59.540 | So on that dimension,

01:03:01.020 | you would say that looks pretty darn expensive.

01:03:03.500 | But if you go to the next chart, Bill,

01:03:05.100 | which I think is an important one,

01:03:06.660 | which is this is the PE ratio divided by growth, right?

01:03:10.860 | So this is the expected growth rate of these companies.

01:03:14.260 | You can see one of the reasons people are excited

01:03:16.900 | is because they expect a lot of growth.

01:03:18.500 | So on that dimension, it's below the 10-year average.

01:03:22.540 | And so what's my conclusion based on that?

01:03:25.860 | You know, that it looks cheap if you expect,

01:03:29.420 | if you believe in those growth rates,

01:03:31.100 | but if those growth rates don't show up for Microsoft,

01:03:34.340 | for Amazon, for Google, et cetera, next year,

01:03:37.060 | then you can expect that these companies are gonna,

01:03:39.540 | their stocks are gonna go sideways to down, right?

01:03:42.620 | Because the valuations are much more full.

01:03:46.300 | And so I think that that's really the debate, you know, now.

01:03:49.900 | And I think it's a stock picker's market from here.

01:03:52.300 | We have average levels of exposure.

01:03:54.220 | I think, for example,

01:03:55.780 | we think the entire NVIDIA

01:03:59.220 | and AI infrastructure supply chain

01:04:01.140 | is gonna continue to be undersupplied.

01:04:04.020 | So NVIDIA has come off from 140 to 115.

01:04:06.820 | There's a lot of debate in the world.

01:04:08.180 | A lot of people think, you know,

01:04:09.900 | that they're not gonna hit the numbers.

01:04:12.380 | You know, we're kind of at, you know,

01:04:14.500 | 6 million GPUs for next year.

01:04:16.500 | The bearish people are at like 4 1/2 million.

01:04:19.340 | Like the numbers will ultimately tell.

01:04:21.460 | If they do 6 million next year, the stock's going higher.

01:04:24.020 | If they do 4 1/2 million, the stock's going lower.

01:04:26.940 | That's the way this business works, right?

01:04:29.460 | And so we're just out there trying to collect all our data.

01:04:32.580 | In fact, Clark's over in Taipei right now,

01:04:35.020 | talking, you know, meeting with the supply chain,

01:04:36.940 | understanding what's really going on.

01:04:38.500 | And I think, you know, in January '23,

01:04:41.300 | you had a huge margin of safety.

01:04:42.940 | All you had to believe is that the world wasn't ending

01:04:45.180 | and that we're in the start of a new super cycle

01:04:47.260 | and you push chips onto the table, right?

01:04:49.420 | If you understood that, you had pocket kings or pocket aces.

01:04:52.860 | As we sit here today,

01:04:54.820 | the world is much more bowled up, right?

01:04:58.300 | So, you know, even if you have a differentiated point of view,

01:05:01.420 | it's more like sitting on pocket nines,

01:05:04.140 | you know, not like pocket kings.

01:05:05.660 | I think you got to take a more measured view of the market

01:05:08.660 | and think about this distribution of probabilities.

01:05:11.780 | There's certainly we could see the economy slow.

01:05:14.300 | Certainly we could see Blackwell

01:05:16.340 | doesn't get his production levels up.

01:05:17.940 | That would be, you know, a challenge

01:05:19.540 | for the entire ecosystem.

01:05:21.140 | So it's an exciting time.

01:05:23.500 | I don't think there are any no-brainers in the market,

01:05:25.820 | but I can also see how, you know,

01:05:27.700 | when I look at the tailwinds behind tech right now,

01:05:30.260 | both in the private and the public markets,

01:05:32.660 | I couldn't be more excited about the next five years.

01:05:35.220 | It'll be volatile as they always are,

01:05:37.540 | but there's no doubt there are going to be

01:05:38.940 | some big winners produced, you know, in this cycle.

01:05:42.620 | Let's wrap it there.

01:05:43.860 | It's good to see you.

01:05:44.900 | Let's do it.

01:05:45.740 | It's good to see you too.

01:05:46.580 | Look forward to seeing you next time.

01:05:47.420 | No, I'm going to be down in Austin soon.

01:05:48.700 | Let's get a poker game going.

01:05:50.620 | All right, man.

01:05:51.700 | Take care.

01:05:52.540 | (upbeat music)

01:05:55.140 | (upbeat music)

01:05:57.740 | As a reminder to everybody,

01:06:03.580 | just our opinions, not investment advice.

Ep16. Nuclear Update, AI Fast & Furious, State of VC | BG2 w/ Bill Gurley & Brad Gerstner

Chapters