Ep8. AI Models, Data Scaling, Enterprise & Personal AI | BG2 with Bill Gurley & Brad Gerstner

00:00:00.000 | Elon is building a much, much bigger cluster to train a much, much bigger

00:00:04.620 | model as is open AI, as is Zuckerberg.

00:00:07.940 | I mean,

00:00:08.240 | well, Sam just said the bigger models aren't the best.

00:00:11.420 | Well, I mean, he may be doing the same game that everybody else is doing,

00:00:14.940 | Bill, and trying to throw everybody off the scent.

00:00:17.180 | Great to have you guys.

00:00:32.000 | What a week it's been nuts.

00:00:33.840 | There's so much to talk about and we have our good buddy, Sonny.

00:00:37.400 | How are you doing?

00:00:37.920 | Deep Madra in the house.

00:00:39.880 | Sonny, somebody, Bill and I go too often when we're talking through all things.

00:00:44.400 | AI currently at Grok working on the inference cloud.

00:00:48.240 | So you're deep in thinking about all these things, all these models, AI, and

00:00:51.800 | we're going to talk a lot about models today with the release of Llama 3.

00:00:55.400 | So it's good to have you, Sonny.

00:00:57.040 | Good to be here.

00:00:57.640 | Thanks guys.

00:00:58.380 | Bill, why are you in town?

00:00:59.760 | Board meetings, a couple of board meetings.

00:01:01.700 | Good to have you.

00:01:02.700 | Yeah.

00:01:02.960 | I always like doing this in person.

00:01:04.200 | Yeah, I know you do.

00:01:04.800 | Okay.

00:01:05.200 | So let's, so let's roll.

00:01:06.560 | They're my favorite episodes when you do them live.

00:01:08.240 | Okay.

00:01:08.640 | There we go.

00:01:09.320 | There we go.

00:01:10.080 | So models, models, models, models.

00:01:13.360 | If AI is the next big thing, then this felt like another really important week.

00:01:18.360 | I mean, we got models being dropped by Meta with Llama 3.

00:01:22.400 | That was the one that was really, you know, the category five earthquake.

00:01:26.720 | Microsoft, Snowflake, everybody seems to be out with a new model, but let's start

00:01:31.720 | with Zuck, huge Llama 3 unveiling, three distinct models, an 8 billion, a 70

00:01:37.640 | billion, and a 405 billion parameter model, which is still training and still

00:01:41.800 | learning they're telling us, which is pretty fascinating, but what seems to

00:01:46.280 | have, you know, shocked the market is that Meta could pack so much intelligence

00:01:51.320 | into such a small model.

00:01:53.360 | And so both models quickly shot up the rankings this week.

00:01:58.080 | We have some, you know, a screenshot here of that.

00:02:00.880 | Of course the 405 is still training and there, there've been some hints out of

00:02:05.920 | the, a recent podcast with Zuck and Dwarkesh about it may in fact kind of

00:02:10.480 | come in at the top of the polls.

00:02:11.760 | We'll see, it's probably going to train for another couple of months, but I'd

00:02:15.000 | love to hear from both of you guys.

00:02:16.600 | What were your big takeaways from the launch of Llama 3 and maybe start with

00:02:21.040 | you, Sonny, walk us through kind of just the what and the how of Model 3 and why

00:02:25.320 | it really kind of shook things up.

00:02:27.680 | Yeah, I would say, you know, the biggest impact of Llama 3 is its

00:02:33.240 | capabilities and at the size and what, you know, Zuck shared in that interview

00:02:38.920 | was that they basically took the model and kept training it past the chinchilla

00:02:43.360 | point.

00:02:43.720 | And so really by doing that, which is generally considered like sort of the

00:02:48.600 | point of diminishing returns, they were able to pack much more information and

00:02:52.920 | much more capability into this model with the same data set.

00:02:56.400 | So, so, so just for, for, for everybody listening.

00:02:59.240 | So the chinchilla point, if I understand it correctly, right, that it's the

00:03:03.440 | by-product of this paper out of Google, which basically talked about the

00:03:06.760 | relationship between the optimal amount of data to use for an, for a certain

00:03:11.080 | amount of compute.

00:03:11.800 | Yep.

00:03:12.240 | But in the case of Meta, when they were training Llama 3, they were basically

00:03:18.400 | continued with these forward passes of the data.

00:03:21.520 | So they were curating the data, refining the data, pushing it back into the model.

00:03:25.560 | And I think several people who are working on pre-training at Meta said they

00:03:30.480 | were even surprised that it was still learning when they took it offline on

00:03:34.840 | that data.

00:03:35.440 | Yeah, and they only took it offline to reallocate the resources to, you know,

00:03:39.080 | 405 and other efforts.

00:03:40.960 | And I think he said Llama 4.

00:03:42.520 | And Llama 4.

00:03:43.120 | Right.

00:03:43.440 | So the rate of innovation, certainly not slowing down there.

00:03:46.200 | So a 15 trillion parameter model.

00:03:48.680 | 15 trillion tokens used to train it.

00:03:50.840 | Oh yeah, yeah, 15 trillion tokens used to train, you know, the model.

00:03:55.960 | I know at Grok, you guys are deploying Llama 3.

00:04:00.960 | I think you deployed it the same day that it came out.

00:04:03.800 | So how important is this?

00:04:06.440 | How important a development is it in the world of models?

00:04:09.480 | Well, really, you know, Zak came out and threw down for the entire world of folks

00:04:14.680 | that are building models.

00:04:15.680 | And it's really disruptive because when you look at the rankings, you have a model

00:04:20.640 | that's much smaller, so much easier to run on all different types of hardware and

00:04:25.560 | much faster.

00:04:26.440 | And so those two things are like catnip for developers.

00:04:30.880 | And for us, we saw within the first 48 hours, it become the most popular model that

00:04:36.400 | we run on Grok.

00:04:37.280 | And so, really.

00:04:39.240 | Replacing what?

00:04:40.240 | Replacing Mixtral 8x7 for us.

00:04:42.960 | Interesting.

00:04:43.360 | Which was, you know, generally considered the best open source model at that point.

00:04:46.760 | And what the capabilities have happened beyond us sort of running it, the developers

00:04:52.440 | that use it, the use cases we've seen it in are incredible.

00:04:56.400 | And people are doing a direct replacement with OpenAI across the board.

00:05:00.920 | They come to us.

00:05:02.040 | So they come to, you know, all the different providers and they replace out OpenAI and

00:05:06.880 | they don't really see any performance impact or any reasoning impact or, which is

00:05:11.400 | incredible.

00:05:11.600 | And why replace?

00:05:12.480 | What is being optimized in the switch?

00:05:15.400 | Price performance, right?

00:05:17.520 | You get probably on from a GPT-4, you're more than 10 times cheaper, right?

00:05:24.880 | And you're, yeah, 10 times cheaper.

00:05:27.240 | And well, let me just tell you, GPT-4 is $10 per million tokens input and $30 per

00:05:35.480 | million token output.

00:05:36.480 | And LLAMA370B is $0.60 for a million tokens input and $0.70 for a million tokens output.

00:05:45.400 | I mean, Bill, this seems to be playing right into your thesis around kind of just

00:05:51.720 | these models generally.

00:05:53.120 | Commoditization.

00:05:54.680 | Yeah, you've been skeptical about the amount of dollars it's taking to train some of

00:06:00.560 | these venture-backed models and the business models that would come out the other side.

00:06:04.920 | Now we have a business in meta, right, that just announced they're going to spend $40

00:06:09.720 | billion this year on CapEx that just trained a model that is 10x less expensive than, you

00:06:16.400 | know, the most performant model on the market.

00:06:18.400 | I mean, what does this mean for everybody else?

00:06:21.400 | Well, there's a couple of things that I put into the mix as I analyze this and answer

00:06:29.560 | your question.

00:06:30.200 | You know, first, he made meta AI free and he didn't proclaim that this was temporary or

00:06:39.280 | that he might pull it up later.

00:06:40.720 | And so, you know, that combined with, I think, Perplexity claiming they're going to have

00:06:46.160 | an ad portion and OpenAI hinting at that, at least for the time being, I think the $20

00:06:52.280 | concept is gone.

00:06:54.560 | Yep.

00:06:55.080 | And that was a big part of OpenAI's revenue, apparently, or we believe rumored to be.

00:06:59.560 | Yeah, over 50% of their revenue is, I think, from consumer.

00:07:02.160 | And so that's gone.

00:07:03.760 | As long as meta's free, I don't think anyone pays the $20.

00:07:07.560 | And I will say that is as things sit today.

00:07:11.240 | If some crazy feature comes along, you know, we've talked about personal memory, maybe

00:07:16.920 | that comes back.

00:07:18.240 | But for now, it feels dead.

00:07:20.240 | Right.

00:07:20.720 | And then I thought the podcast that you mentioned was just incredibly, like,

00:07:27.920 | disclosive, transparent, thoughtful.

00:07:32.320 | You're talking about Zuckerberg on Dwarkish?

00:07:35.800 | Yes.

00:07:36.120 | Yes.

00:07:36.440 | I thought it was incredible.

00:07:38.080 | And it's funny, because it came out at the exact same time that Sam and Brad were on

00:07:43.360 | Harry Stebbings in 20 BC, and Dario was on Ezra Klein.

00:07:48.240 | And I would encourage people to listen to all three of them, but Dario and Sam talk in

00:07:54.120 | these high-level platitudes about how this stuff's going to cure cancer, and we're all

00:07:58.200 | going to not have to work anymore.

00:08:00.120 | And Zuck was down in the weeds, in the meat, being super transparent.

00:08:06.320 | And I was just like, "Holy sh*t, maybe this guy's in charge now."

00:08:09.960 | Right, right.

00:08:10.640 | You know?

00:08:11.080 | I mean, I saw a lot of people on Twitter saying this was checkmate on all of these

00:08:16.760 | closed models that have gotten started and venture-backed over the course of last year.

00:08:20.920 | And certainly, you know, I'm not cheering for that.

00:08:23.640 | I'm not sure that I would go so far as to declare it.

00:08:26.640 | But I think if you're in the business of producing a closed model, right?

00:08:30.560 | We've talked a lot on this pod.

00:08:32.240 | There are one of two ways that you can build a business.

00:08:35.000 | You either have to sell it to a consumer for 20 bucks a month, or advertising, you

00:08:39.280 | have to get a billion consumers to use your product, or you have to sell it to an

00:08:43.280 | enterprise.

00:08:43.800 | Somebody has to pay you.

00:08:45.120 | And now you have a disruptor coming along and saying, "It's open, it's cheap or

00:08:50.080 | free, and I'm not going to charge for it."

00:08:52.800 | That's hard to compete with.

00:08:54.360 | And I use them all because I'm curious and I love playing.

00:08:57.880 | I'll do the same query on four of them.

00:09:01.000 | But right now, it's hard to believe that any of them, including the Google one, are

00:09:06.840 | going to have escape velocity because of differentiation.

00:09:09.800 | I'm not seeing, maybe you guys are, I'm not seeing an element of differentiation on

00:09:14.680 | the consumer-facing tool that's radically different.

00:09:19.080 | It's going to cause 80% market share.

00:09:21.200 | The differentiation is happening on the infrastructure side.

00:09:23.680 | And there's another thing that, you know, Zuck said in that, which was, I think, a

00:09:27.360 | big throwdown, which is, we're going to spend $100 billion on this.

00:09:31.160 | But if the community, because he's made it open, makes it 10% better, right, in some

00:09:37.240 | parameter, that's $10 billion savings for us.

00:09:40.680 | Right.

00:09:41.480 | If you just look at what they did this week, they pushed AI Search across their entire

00:09:47.280 | family of applications.

00:09:48.680 | They have 3 billion people using those every day.

00:09:51.400 | He said on his earnings call this week, they already have tens of millions of people

00:09:55.200 | running AI Searches on their applications.

00:09:57.560 | And he was even honest in the podcast where he's like, I don't know if this is where

00:10:01.440 | people want to do these searches.

00:10:02.840 | And it just earned so much credibility with me when someone kind of, you know, comes

00:10:08.160 | clean that way.

00:10:08.800 | I think one of the other interesting developments, again, just getting back, if I

00:10:12.400 | had one big takeaway, it was this large versus small.

00:10:18.040 | Okay.

00:10:18.800 | And so I think we have to...

00:10:21.360 | Before you go there, I got a question for you.

00:10:23.000 | Yeah.

00:10:23.600 | Could the CapEx thing be a throw down?

00:10:26.160 | Like, could it be a signal to the rest of the community?

00:10:31.640 | This is where we're going to be.

00:10:33.640 | Like, just to, could it be a move to tell everyone else, if you want to stay in this

00:10:41.240 | game, you got to play at this level.

00:10:42.280 | It's also two birds with one stone, right?

00:10:44.600 | I think there is that, but there's the second one, which is, you know, also talked

00:10:48.480 | about in another set of tweets this week, by putting more effort and resources towards

00:10:54.360 | the training, they reduce the inference costs.

00:10:57.440 | Yeah.

00:10:57.760 | Right, for sure.

00:10:58.680 | And for them, you can imagine everything that you're talking about, making it free

00:11:01.640 | everywhere, putting it inside all the products, millions of people using it.

00:11:05.320 | That's a huge impact.

00:11:06.400 | And I would remind people, you know, something we talked about several episodes

00:11:10.080 | ago, but there was a, there was a podcast I listened to, um, where Amazon was talking

00:11:16.720 | about, um, their Alexa product.

00:11:20.120 | And they said, you know, inference was way more of the cost than, than the

00:11:27.280 | training, like night and day.

00:11:28.800 | And so there's a real world application that's been alive.

00:11:32.160 | And, and if that's, do you believe that's true for almost all development projects?

00:11:36.120 | Oh, yeah.

00:11:36.560 | Yeah.

00:11:37.760 | And so it's, it's weird to, you know, to your question, is capital a signal?

00:11:42.120 | Of course it is.

00:11:43.080 | Yeah.

00:11:43.360 | I mean, in 2021, we talked about capital as, as, as being the kingmaker, like

00:11:48.200 | who would, who, who would win the game?

00:11:50.000 | I think there are four important ingredients, uh, to compete in this market.

00:11:54.600 | Number one, you have to have capital and the leaders are spending 40 billion a year.

00:11:59.320 | There aren't many sovereigns on the planet that can afford to spend $40 billion a year.

00:12:03.720 | Second, what you need, we just, the big, uh, you know, innovation this week is that

00:12:08.640 | data is scaling in the way that compute is scaling.

00:12:12.080 | So you need a lot of data.

00:12:13.640 | They have a massive amount of data.

00:12:15.360 | Third, you need compute.

00:12:16.880 | That's not just about capital.

00:12:18.480 | You have to have people who know how to build and stand up infrastructure.

00:12:22.240 | You have to have relationships with the entire supply chain, right?

00:12:25.840 | And fourth, you have to have distribution, right?

00:12:28.480 | And so they're touching 3 billion consumers.

00:12:30.480 | They have a business model.

00:12:31.680 | I think he said 50% of the content on Instagram was AI generated in the quarter.

00:12:38.360 | Right?

00:12:38.760 | So not AI generated, it was AI suggested, right?

00:12:42.880 | So it was no longer about your friends looking at something.

00:12:45.720 | It was something that, you know, they did.

00:12:47.480 | So that gives them a huge advantage, but I do think there's, you know, when I get

00:12:51.840 | back to this bifurcation and why I think this is important, we have these smaller

00:12:56.560 | models that are going to be specialized and have specialized use cases.

00:13:00.800 | You know, Microsoft is out with Fi and I want you to talk a little bit about that,

00:13:04.760 | Sonny, but we don't see any slowing down on the push to bigger models as well.

00:13:10.680 | So we really see both of these things happening simultaneously.

00:13:14.480 | And an analog I was discussing with our team was if you think about different use

00:13:20.200 | cases, you might build a small rocket to get a satellite into space and you might

00:13:25.360 | build a big rocket to try to get to Mars.

00:13:27.400 | Okay.

00:13:28.360 | Now those are both rockets.

00:13:30.960 | But they have radically different use cases and radically different cost

00:13:34.400 | structures.

00:13:34.920 | And I think that the cost structures that are going to be associated with frontier

00:13:39.560 | level models, there are going to be very few companies on the planet that are going

00:13:43.440 | to be able to build those models.

00:13:45.320 | Because I think the latest discussions, whether it's, you know, Stargate out of

00:13:50.040 | Microsoft, a hundred thousand GPU cluster, Elon's talking about a hundred thousand

00:13:54.560 | GPU cluster, Mark is talking about that.

00:13:57.120 | I just don't know many companies that are going to be able to compete with that.

00:14:00.760 | Yeah, I'll take maybe just a slightly tangential view to that, which is if you

00:14:05.120 | think about, you know, Meta's history in open source, open compute project, PyTorch,

00:14:11.600 | React.js, these are just infrastructure components for them, right?

00:14:15.920 | And they put the investment in so that they can drive improvements in the supply

00:14:19.920 | chain.

00:14:20.440 | They can drive the ecosystem to make it better.

00:14:23.080 | And I think they've really taken that approach with this technology and said,

00:14:26.840 | "Hey, this is a infrastructure level component that, you know, we want the

00:14:30.440 | ecosystem to make better."

00:14:31.800 | And everyone else is in the business of models, whether you're a hyperscaler or

00:14:37.000 | whether, you know, you're one of these model companies.

00:14:38.960 | And I think that's a distinctly different approach for them that puts them at an

00:14:43.560 | advantage.

00:14:44.240 | But by the way, I think this is worth drilling in on.

00:14:47.040 | So, unless one of you correct me, they are not in the cloud hosting business and

00:14:52.640 | remain not in the cloud.

00:14:53.640 | You're talking about Meta.

00:14:54.600 | Meta.

00:14:54.920 | Correct.

00:14:55.520 | And so the people that they're up against have businesses they're running based on

00:15:03.320 | these things.

00:15:04.000 | They're developing this thing, spending some number of billions and putting it out

00:15:10.880 | as open source.

00:15:11.920 | I think it's a little different than the open compute part where I don't think they

00:15:18.040 | felt the differentiation of their architecture had any impact on the strategic

00:15:24.080 | execution of their company, almost the opposite, like it's a commodity, so let's

00:15:28.560 | exploit it like a commodity.

00:15:30.160 | Here, this feels more kind of like a badass throwdown where there's a very

00:15:38.320 | intentional element of burning, you know, the strategic ground out there for

00:15:45.840 | everyone.

00:15:46.480 | You know, similar, I think, to what maybe Google did with Android when they came

00:15:52.560 | out, like just protect all around me by making it very hard to have differentiated

00:15:59.280 | products built on AI that you might come after me with.

00:16:02.920 | I mean, one of the-

00:16:03.680 | Is that fair or is that not?

00:16:04.680 | Yeah, I think it's fair, but I think I just want to make a couple of points that

00:16:07.760 | he, that Mark Zuckerberg talked about on the Dworkish podcast.

00:16:13.280 | One was, he said that they do in fact have revenue sharing relationships with the

00:16:18.560 | hyperscalers such that when they use their models, they ought to get compensated

00:16:23.440 | something for that.

00:16:24.120 | Now, he didn't go, I think he said it wasn't a very big number, but relative to

00:16:27.720 | $165 billion in revenue, nothing is a big-

00:16:30.320 | Any color on that?

00:16:31.200 | Yeah, no, nothing is a big number.

00:16:32.480 | You know, the hyperscalers have definitely been squeezing all the model

00:16:37.680 | makers, right?

00:16:38.760 | And they have a really interesting position because, you know, especially the

00:16:42.840 | ones that are creating their own because they have to create a marketplace and they

00:16:46.400 | have to ensure that they're operating sort of in a free market capacity.

00:16:50.360 | But it's difficult, right, when you have your own models, because there's

00:16:53.320 | obviously a lot of interest to drive that.

00:16:55.200 | I can definitely confirm that the data clouds are paying a revenue share to the

00:17:00.840 | open models.

00:17:01.520 | I don't know what the revenue share is, but there will be some compensation.

00:17:05.000 | And listen, that compensation can change over time.

00:17:07.960 | So that's one bit of it.

00:17:11.120 | The second thing is super important for all of us to listen to this again.

00:17:17.200 | Zuck said, "We believe in open source, but there may come a time where we have a

00:17:23.640 | discovery in our largest model, perhaps, that is fundamental and economic to our

00:17:28.960 | business, where we will elect to no longer open source said model."

00:17:33.120 | So you can see a world where they will always open source the 7B or, you know,

00:17:37.720 | he said he wants to build a 1B or a 500 million parameter model or the 70B, but

00:17:43.720 | you can also see a world where their most sophisticated model is not open source

00:17:47.440 | because he says, "Listen, I want to build the best personal AI in the world.

00:17:51.240 | It's central to what our business is about.

00:17:53.480 | We want to have the advantage associated with that."

00:17:56.400 | So I think the strategy for me, it feels like the reason the earth shook this week

00:18:03.040 | is that this felt like the most significant development and disruptive

00:18:10.000 | element in the model marketplace.

00:18:12.200 | I think it's going to be very difficult for new entrants to be venture-backed

00:18:16.360 | because to, you know, open AI will continue to get funding because they

00:18:21.360 | have this incredible team.

00:18:22.880 | They have a hundred million people using the product and paying them for the

00:18:25.880 | product, but I think for all the other closed models, they're going to have

00:18:29.280 | trouble getting follow-on financing.

00:18:31.360 | And I think any new models that come along, you would have to have something

00:18:34.920 | so different, such as orthogonal angle of attack in order to get funding.

00:18:40.240 | So I think to your point, by throwing down on the CapEx that you're going to

00:18:45.040 | spend, you are clearing the market of potential competitors, right?

00:18:48.960 | It's a very quickly depreciating asset.

00:18:51.160 | Boy, I mean, that's just so, like, unbelievable is the steepness of the

00:19:00.320 | price curve on a slightly older model.

00:19:04.080 | Like, and if people are maximizing ROI on an inference basis, they're going to

00:19:11.000 | use, they're going to take advantage of that like crazy.

00:19:14.480 | I mean, it's going to-

00:19:15.240 | We took LLAMA 2 out of Grok Cloud.

00:19:17.080 | It's not even available.

00:19:18.640 | We just took it out and replaced it with LLAMA 3 and all the

00:19:21.240 | developers went to LLAMA 3.

00:19:22.480 | But it's already, LLAMA 3 is already one 20th, one, whatever.

00:19:28.720 | Yeah.

00:19:29.640 | So there's no reason.

00:19:30.760 | Well, I mean, it seems to me where the value is again, coming back to maybe

00:19:34.040 | we'll switch to, you know, this is the right transition to talk about

00:19:38.880 | enterprise AI because the value is not in the model, right?

00:19:45.680 | Just like the value is not in storage, right?

00:19:48.680 | You could say storage is a part of the AWS cloud, but there's not a lot of

00:19:52.680 | value in that thing unto itself.

00:19:54.520 | The value is in the enterprise relationship.

00:19:56.680 | The value is in, right, the number of services that you're

00:20:01.480 | offering to your customers.

00:20:03.000 | So Microsoft and Google are out tonight.

00:20:05.000 | Both clouds accelerated their growth on the back of AI.

00:20:09.720 | 64% of Fortune 500 customers are now Azure OpenAI customers, which I

00:20:17.520 | thought was pretty extraordinary.

00:20:18.840 | Big numbers.

00:20:19.320 | That's a-

00:20:19.640 | GitHub Copilot growing 35%, quarter over quarter.

00:20:23.600 | And the number of use cases seem absolutely wild.

00:20:26.680 | And what's even crazier is Satya said on the call, the revenue growth

00:20:30.400 | would be even higher, but they're GPU constrained.

00:20:33.160 | Yes, you heard me say it, Bill.

00:20:34.680 | They're GPU constrained.

00:20:36.280 | We'll come back to that.

00:20:37.280 | So I look at this and, you know, GCP is accelerating.

00:20:42.200 | Azure's accelerating.

00:20:44.880 | My assumption is you, we heard it out of ServiceNow, their

00:20:48.240 | demand, you know, is accelerating.

00:20:50.640 | So clearly enterprises are finding value in use, you know, in this.

00:20:56.120 | So Sonny, talk to us a little bit about what you're seeing.

00:20:59.720 | I know you have a hundred thousand developers in the long tail now using,

00:21:03.720 | or I think a lot of big enterprises as well, using the Grok cloud.

00:21:08.600 | But what are these enterprise use cases?

00:21:10.640 | And are you surprised when you see these hyperscalers

00:21:13.480 | racking up these numbers?

00:21:14.680 | I'm not surprised.

00:21:16.160 | And let me level up the question for a quick second into like,

00:21:18.760 | where is that spend coming from?

00:21:20.240 | And right now, and this is, you know, even verified by this report

00:21:24.400 | that Andreessen Horowitz put out a couple of weeks ago around enterprise AI.

00:21:28.080 | And what they really showed is like the distribution of use

00:21:30.840 | is coming from IT to the business units to support.

00:21:34.680 | And it's not in these innovation arms, because when usually you see

00:21:38.240 | these technologies, when they're there, you understand the budgets are limited.

00:21:41.280 | So that's awesome.

00:21:42.840 | They also just as a relative point, they showed that folks are tripling

00:21:46.400 | their AI spend this year, right?

00:21:48.480 | And so that that kind of lines up to what we're seeing there.

00:21:50.880 | Right.

00:21:51.080 | And we'll show these slides.

00:21:52.120 | Yeah, we'll show these charts there.

00:21:53.560 | And, you know, I think the most interesting thing,

00:21:56.480 | and I'll get into the use cases that 82 percent of the respondents said

00:22:00.560 | they are or either already on open source or will move to open source.

00:22:04.280 | So that's the interesting fact that's happening there.

00:22:07.000 | The use cases really.

00:22:09.840 | And, you know, let's maybe we got a little bit of alpha

00:22:13.200 | from Michael Dell a couple of weeks ago when, you know, he really talked to us

00:22:17.520 | about this use case for enterprise rag, right, where there's all this data.

00:22:23.360 | And I want to be able to reason over that data with a model.

00:22:26.960 | Right.

00:22:27.600 | And so, you know, his interests, obviously, what he's selling alongside,

00:22:32.480 | you know, his partners.

00:22:33.840 | But I think in the cloud, you're seeing that heavily happen right now.

00:22:36.640 | Customer support is number two.

00:22:38.240 | I know you guys just financed a company in this space.

00:22:40.280 | So congratulations on that deal, which is really interesting.

00:22:43.280 | And then I think content moderation and content generation.

00:22:46.560 | I think we don't really talk about it enough.

00:22:48.960 | But if you think about a business, this is happening all over the place

00:22:52.160 | all the time. Right.

00:22:53.280 | And we see a ton of use cases still there, where whether it's a daily report

00:22:57.560 | or whether it's something you send out to your customers

00:23:00.240 | and all of that coming out of those enterprise systems and being sent out.

00:23:03.360 | I mean, Bill, do you remember?

00:23:05.040 | I mean, 18 months ago, pre-chat GPT,

00:23:07.920 | I imagine less than five percent of enterprises in this,

00:23:11.160 | you know, were building AI production use cases.

00:23:16.040 | Today, I don't know an enterprise that's not at least running a test use case.

00:23:21.240 | Well, you said it was in a percentage there.

00:23:23.160 | It wasn't 100. No, no.

00:23:25.000 | But that was 64 percent using Azure.

00:23:28.240 | But I think it's probably close to 100 percent.

00:23:32.160 | I can't imagine a company in the S&P 500 that's not at least testing AI.

00:23:36.920 | Right. You would really have to be asleep.

00:23:39.160 | Do you remember any other technologies that went from zero to ubiquity this fast?

00:23:43.560 | I mean, maybe the Internet itself.

00:23:45.680 | People said, oh, my God, I got to get on the Internet.

00:23:47.640 | Mobile, mobile. I think mobile.

00:23:50.000 | Although this one, but I don't think that's a secret.

00:23:53.800 | We've talked about how the incumbents moved very quickly here.

00:23:57.520 | And I think you can give OpenAI a lot of credit because they were out

00:24:01.480 | selling the mission and out talking to the customer base

00:24:05.640 | and doing everything they could to promote.

00:24:08.360 | It's also easy to use.

00:24:10.400 | Like when you talked about some of these other technologies,

00:24:12.280 | like going to cloud was like a real effort. Right.

00:24:14.800 | Right. Using it to migrate your entire database.

00:24:17.760 | Exactly. You had to do a lot of real work.

00:24:20.160 | This is an API call. Right. Right.

00:24:23.120 | And again, credit to OpenAI.

00:24:24.920 | They're the ones that led everyone down that path.

00:24:27.280 | And everyone else now is OpenAI compatible or has a similar looking API.

00:24:31.320 | It's very easy to use.

00:24:32.840 | And part of the reason, you know, one of the things,

00:24:35.480 | you know, you mentioned, Michael Dell, he tweeted the other day

00:24:37.800 | this Barclays survey that I thought was really fascinating.

00:24:42.320 | So this is among the enterprise CIOs moving back to hybrid and on prem.

00:24:47.840 | The number was that 83 percent of respondents

00:24:51.560 | said that they were going to repatriate at least some of their workloads

00:24:56.000 | right back to on prem.

00:24:57.640 | And that was up from 49 percent or 43 percent in 2020. Right.

00:25:02.960 | And so I think it's an interesting case that you're moving back.

00:25:07.920 | My sense is it's because they don't trust certain data in the cloud. Right.

00:25:13.160 | So they want they don't want to run maybe code generation,

00:25:16.320 | you know, tools in the cloud.

00:25:18.040 | And the other one is just data gravity.

00:25:20.040 | Maybe they have on prem databases and they don't want the cost

00:25:23.440 | and the headache associated with moving that to the cloud.

00:25:25.640 | Do you see this in other parts of of your world, Sonny?

00:25:29.640 | Yeah, you know, definitely a lot.

00:25:31.400 | A third one, which is I think there's still a lack of trust.

00:25:35.080 | And this gets expanded every time.

00:25:37.000 | You know, we had that interview with the OpenAI CTO where they asked her,

00:25:40.840 | hey, have you trained this on, you know, data?

00:25:42.800 | And she didn't answer the question quite well.

00:25:45.160 | And so I think and I've heard this,

00:25:47.240 | you know, in conversation with hyperscalers where customers will not trust

00:25:51.080 | and hyperscale will legally sign that they will not train.

00:25:53.520 | They still will not trust. Right.

00:25:55.560 | They just they believe that all these stories around the data

00:25:58.640 | make these models better, that everyone is just wants a way

00:26:01.080 | to get access to that data to make the models better.

00:26:03.360 | So I think the combination of those three factors is 100 percent what we see.

00:26:07.080 | And so what what happens, you know, with us, which is just,

00:26:10.600 | you know, basically maybe a pattern, people come and try something in the cloud,

00:26:14.760 | make sure that it works and then immediately want to get on the phone

00:26:17.720 | with you and say, hey, can I can I get this on prem?

00:26:20.000 | Interesting.

00:26:21.120 | Or at least sequestered.

00:26:22.440 | Or sequestered, yeah.

00:26:23.280 | Like I use on prem as like some virtual.

00:26:25.360 | I think I thought there were two things in these podcasts

00:26:28.920 | that we keep referencing that relate to the enterprise decision making.

00:26:33.720 | One, you know, Zuck said something that kind of makes sense to me.

00:26:39.480 | He just said, like, you know, cramming data in the context

00:26:43.480 | window feels a little hacky or he I don't know what his exact words were.

00:26:47.280 | And so I think there's still this this future in front of us

00:26:52.040 | where data gets deeper integrated in the model and the trust issues there.

00:26:57.240 | And we don't quite know how that's all going to come together.

00:27:00.200 | It's still TBD. Yes.

00:27:02.480 | Yeah. And you guys probably haven't tried it because it's just not,

00:27:05.680 | you know, feasible if you're not a developer, but using like a million,

00:27:08.720 | you know, contact like a context window of like a million tokens.

00:27:12.600 | It's like really hard.

00:27:14.280 | Yes. You can't use it up, you're saying.

00:27:17.040 | Well, you can use it up.

00:27:18.440 | But the amount of gathering and work you have to do to get a million to,

00:27:21.440 | you know, think about it's like several books, you know.

00:27:24.040 | And so, you know, people talk about it like it's this wonderful thing,

00:27:27.240 | but it's not it's not, you know, overly usable.

00:27:29.920 | And then the other one, I thought the most interesting thing out of the the

00:27:33.880 | the Sam podcast was he talked about whether or not developers

00:27:40.160 | were kind of going wholesale on top of open

00:27:43.840 | AI or whether they were just using it in a lightweight way

00:27:47.200 | and then doing a bunch of stuff externally.

00:27:50.600 | And he implied that most people are doing the latter.

00:27:54.440 | But then he said, if you do that, we're going to steamroll you

00:27:58.040 | and you need to bet on us being successful.

00:28:00.560 | And which would mean dumping your data and and trusting open AI more fully.

00:28:07.160 | I don't know. What was your interpretation of what he was trying?

00:28:10.360 | It was exactly that, plus the following.

00:28:12.560 | If you if you take his, you know, take him for what he was saying,

00:28:15.280 | which is the models are going to get better.

00:28:17.080 | Well, what room does it leave for anything else?

00:28:19.640 | Because if you shouldn't be taking a model and wrapping it with your own,

00:28:23.720 | you know, your own code or your own technology or framework,

00:28:26.440 | and then you're going to assume the model gets better.

00:28:28.480 | Well, why do you need what?

00:28:29.600 | Why do you need whatever I'm building if the model can just do everything?

00:28:32.400 | I actually I actually thought Sam and Brad were really articulate

00:28:36.680 | on this point, whether you believe them or not.

00:28:39.800 | And I think it was consistent with the tweet that Aaron Levy sent out yesterday,

00:28:43.600 | which is people are not thinking ambitiously enough

00:28:48.120 | as to where these things are going.

00:28:50.200 | And, you know, today we're really in the land of answers, right?

00:28:53.920 | We're running some rag over some, you know,

00:28:56.480 | HR data that we have in our company and building a little chat bot

00:29:00.000 | so it can answer questions more efficiently than my HR group can

00:29:03.520 | can answer questions.

00:29:04.720 | But they're saying it really needs to think about agentic thinking.

00:29:07.200 | Like, what is that multistep reasoning that can be done in the business?

00:29:10.760 | And, you know, I know how big your HR group.

00:29:14.160 | Well, here it's pretty easy to do.

00:29:16.560 | Here it's pretty easy to do.

00:29:18.080 | But, you know, so my sense is that

00:29:22.320 | I'm kind of in this Aaron Levy camp that when you look out two or three years.

00:29:27.000 | I mean, listen, every week we're blown away

00:29:30.840 | by, you know, how these models are progressing.

00:29:33.720 | It's hard for me to think in three years at the rate of progress

00:29:37.440 | and the amount of investment that's going into this,

00:29:39.600 | that we're not going to be a lot further down the path in terms of this

00:29:42.560 | in terms of this reasoning.

00:29:44.680 | And when we get there, I think people are going to want that

00:29:47.360 | to be more proprietary, because I think the advantages

00:29:50.000 | that are going to inure to the enterprise are even more.

00:29:53.600 | Let me throw one other thing in here.

00:29:56.560 | You know, I was sitting with my team this week and we're trying to figure out

00:29:59.200 | who are the winners and losers, not of the providers of the arms,

00:30:03.400 | but the buyers of the arms. OK.

00:30:05.760 | So if every Fortune 500 company is buying AI,

00:30:09.280 | one of the things that Bill often reminds me is fine.

00:30:14.000 | It will give a little improvement to an airline that starts using AI.

00:30:18.000 | But airlines are a competitive industry

00:30:20.720 | and they're just going to compete away all the profits.

00:30:22.880 | And so that's a defensive move, right?

00:30:24.560 | You don't actually improve the business model

00:30:27.080 | because all the earnings get competed away.

00:30:29.200 | So what you want to find is a market leader, somebody who has 70

00:30:32.960 | or 80 percent of a given market who gets to hang on, right, to all of this.

00:30:37.800 | Or compound their lead. Right. Or compound their lead.

00:30:40.520 | And so, you know, there's a company coming public

00:30:43.680 | in a few weeks called Lineage, which is in the cold storage business.

00:30:47.920 | So they basically are an integral part of the food supply chain.

00:30:51.240 | You know, any refrigerated storage of, you know, a food.

00:30:55.760 | And I think they have a huge percentage of the market.

00:30:58.440 | And I think they have 50 data analysts and scientists now in San Francisco,

00:31:03.360 | because if they can turn the screw a quarter of an inch on spoilage,

00:31:07.120 | a quarter of an inch on energy consumption to keep this food

00:31:11.160 | all bottom line, and so it's all to their bottom line.

00:31:13.720 | And by the way, it doesn't get competed away. Yeah. Right.

00:31:16.280 | And so they're looking at leveraging, you know, I happen to know

00:31:19.800 | because they were a Snowflake customer and they were using some Snowflake

00:31:23.640 | AI, you know, to improve these use cases.

00:31:27.600 | And so I think they're going to be a whole host of businesses, Bill,

00:31:30.400 | industrial businesses that capture some of these profits

00:31:33.920 | and get to hold on to them.

00:31:35.320 | I'd be interested in Sonny's reaction to to your question

00:31:38.880 | and maybe to my answer, which is I

00:31:42.720 | when when I meet a company

00:31:46.720 | and see them using AI in a way that feels like ultra compelling

00:31:51.120 | from us, improvement of their own strategic business position.

00:31:54.720 | It's almost always a more traditional AI model

00:31:58.800 | that's running a very particular optimization problem.

00:32:03.240 | It's not an LLM application.

00:32:06.320 | And this stuff's all happening simultaneously.

00:32:09.200 | You know, I think I think that I think that's true.

00:32:12.000 | I don't think it particularly matters because what generative AI has done,

00:32:15.600 | what the chat moment has done is it's caused every enterprise

00:32:20.160 | to get off their ass to get all their data organized

00:32:23.360 | because that's a condition required to benefit from any of this stuff.

00:32:26.600 | But then I think what they do figure out along the way is some basic,

00:32:31.320 | you know, machine learning around time series or forecasting

00:32:36.200 | or things that have been around for quite a while, Bill,

00:32:38.480 | is where they get the most bang for the buck, maybe not from the generative AI,

00:32:42.120 | but they might get there because they got into the pool

00:32:45.320 | because they were motivated by generative AI.

00:32:47.840 | I certainly think it's an accelerant based on everything we're seeing.

00:32:51.440 | Yeah. You know, I'll disagree with you.

00:32:53.960 | I think, you know, what what this technology really enables is,

00:32:58.280 | you know, we get spoiled in Silicon Valley

00:33:00.840 | because we can get the best engineers to build like the most difficult things.

00:33:04.280 | But I think for the average business to do most problems,

00:33:08.080 | whether it was, you know, pre generative AI was very, very difficult.

00:33:12.240 | Now you can basically take a generative AI model

00:33:15.200 | and have it do one of the most advanced things in the world.

00:33:19.120 | And, you know, we've shared an example in our chat, right,

00:33:21.720 | where you can take a picture of a plate of food

00:33:24.600 | and tell it to return to you what's in that food

00:33:27.600 | and how many calories might it be and what's the portion size.

00:33:30.120 | Right. That's done sort of, again, with one prompt.

00:33:33.520 | And so now you've given that ability to every business, every small business.

00:33:36.960 | Right. It's like this business you're talking about.

00:33:38.920 | They can do a lot of improvements without having to have 50 people in San Francisco.

00:33:42.720 | So I think that's where the improvements are really going to come.

00:33:45.200 | Although I could push back on you and use your own

00:33:48.760 | statement about open AI.

00:33:51.920 | And if they achieve everything like what they may be commoditized just because.

00:33:57.600 | Well, someone still has to take it and apply it to that business. Right.

00:34:00.920 | And it may just be the one, you know, the one tech person in that business.

00:34:03.960 | It's a I was on a walk last weekend

00:34:07.080 | with a great economist over at Stanford, and we were talking about

00:34:11.440 | whether or not about the amount of productivity improvement

00:34:14.800 | that would be unleashed into the economy because of AI.

00:34:17.320 | And what was interesting is, you know, productivity

00:34:20.960 | has actually been under assault in this country

00:34:24.120 | because we've limited immigration, which was a huge source of productivity

00:34:28.520 | because de-globalization is actually hurting productivity

00:34:32.360 | because we're not moving the productions of goods and services

00:34:36.280 | to the lowest cost places, you know, anything that's causing friction.

00:34:39.560 | So it's like all the goodness to come out of AI.

00:34:42.520 | We need it just to replace the headwinds

00:34:45.440 | that we have on productivity in other places.

00:34:48.120 | But I digress. Let's move on to the.

00:34:51.080 | Well, can I add one thing to that, which is I think just building

00:34:54.400 | on the point that you said, like Aaron Levy was talking about,

00:34:56.720 | we're not thinking about it big enough.

00:34:58.560 | And, you know, where I where I, you know, someone share this on Twitter

00:35:02.280 | and I can't find the original author.

00:35:03.800 | But if we can, some point we'll share it, which is in the industrial revolution.

00:35:08.000 | You saw, you know, car making go from something bespoke

00:35:11.320 | one car per day to a factory making a thousand.

00:35:13.840 | Same for clothing, same for farms.

00:35:16.040 | And, you know, we've looked at technology as this huge accelerant,

00:35:19.120 | but we really haven't had the industrial revolution for technology.

00:35:22.560 | It's still pretty bespoke.

00:35:24.360 | You know, one developer writing code and and now you have this idea where,

00:35:28.240 | you know, go back to a place where you spend a lot of time travel search.

00:35:31.080 | Right. You could have one agent do a thousand or an agent

00:35:35.400 | and a thousand instances of it, do a thousand searches for you

00:35:38.680 | and find what you're looking for.

00:35:40.320 | We haven't seen that in technology yet.

00:35:42.600 | And I think that's that's the era we're really about to go into,

00:35:45.800 | which ties back to, you know, the point that you said

00:35:48.640 | that you were having on your walk around efficiency for for for society.

00:35:52.720 | Yeah, I mean, I think I think about it in the context

00:35:55.480 | of what we've called business intelligence, right?

00:35:58.080 | We've been investors in companies like Tableau,

00:36:00.440 | you know, obviously Snowflake, et cetera, over the years.

00:36:04.480 | And, you know, it's not really business intelligence, right?

00:36:08.440 | Issuing me a report that tells me how many black T-shirts

00:36:11.720 | I sold yesterday, right, is nice, but it's not all that informative.

00:36:15.680 | What you would like is an agent to scour all of your data,

00:36:19.520 | compare it to all the data of other companies and say,

00:36:22.480 | here's something that is anomalous or we can predict something

00:36:26.880 | or suggest something based upon patterns we're seeing in other businesses.

00:36:30.880 | That's all. You know, we've been talking about that for a decade, right?

00:36:34.760 | I actually think we're getting a lot closer to that moment

00:36:37.400 | where now we're going to be able to have these resources,

00:36:40.280 | because what are these things do really well, Bill?

00:36:42.360 | They devour data.

00:36:44.320 | They spot patterns and they predict. Yes. Right.

00:36:47.200 | Take what you said times a thousand shouldn't be a single age.

00:36:50.200 | It could be a thousand of them doing it on your data.

00:36:52.280 | We beat up what it's going to do in the enterprise.

00:36:55.200 | But, you know, one of the areas that I'm even more excited about

00:36:58.440 | as these models get smaller is what it's going to mean

00:37:01.520 | for personal search and personal AI.

00:37:04.480 | So when we think about that, you know, Google reported tonight

00:37:09.000 | they had billions of what they call their SGE searches.

00:37:12.640 | So these are, you know, their AI searches.

00:37:16.520 | They talked about dramatically driving down the cost of inference

00:37:19.840 | of those searches that you can probably tell us a little about.

00:37:23.280 | You know, Meta has rolled out search across all of their apps.

00:37:26.760 | There's a search bar on Facebook, on IG, on WhatsApp.

00:37:30.760 | And you can search any topic.

00:37:32.840 | You can go there and say, hey, show me the recipe for fried chicken

00:37:36.680 | or show me how to, you know, play a guitar or show me where I should stay

00:37:41.440 | at a hotel, you know, when I'm visiting Milan.

00:37:44.400 | And Zuck did say in his announcement, kind of as a shot across the bow

00:37:49.240 | at Claude and at ChatGPT

00:37:53.880 | that they had the most cap, the most capable, free

00:37:58.360 | personal assistant, right, you know, that you could get out there.

00:38:02.200 | You know, we had Apple announce Open ELM, which were these models

00:38:06.480 | from 270 million parameters to three billion parameters.

00:38:09.920 | You know, it seems like the next step that everybody's looking at

00:38:13.520 | is really the smaller models that can get us to,

00:38:17.160 | you know, a personal assistant on device, whether it's on phone,

00:38:21.240 | whether it's on glasses, et cetera.

00:38:23.520 | So when you looked at the announcements this week, you can go to either of you.

00:38:28.240 | It felt to me like the disruption caused by LLAMA 3

00:38:33.320 | was almost more impactful to what we're going to see

00:38:37.160 | along the lines of consumer AI and search than it was in the enterprise.

00:38:41.440 | Any thoughts about that?

00:38:43.600 | Yeah, I think, you know, it ties back to a point we touched on earlier, right?

00:38:47.040 | The as we make smaller models more capable

00:38:50.720 | and we make even smaller and smaller models that can maybe reference

00:38:54.520 | those larger models, we're on to a place where it becomes more affordable.

00:38:59.520 | Right. What we don't really think about, you know,

00:39:02.480 | if you think about the larger models is even so crazy.

00:39:06.160 | It's like a year ago. Right.

00:39:07.920 | You know, all the way back a year ago, back a year ago.

00:39:10.360 | You're using a, you know, thirty thousand dollar plus unit of compute

00:39:15.600 | to run this thing with, you know, hundreds of gigabytes of memory.

00:39:18.640 | Now, whether you look at the Apple stuff or fi

00:39:22.200 | that came out of Microsoft, you can run that on your phone.

00:39:25.280 | People are already running it on their phone.

00:39:26.840 | I saw a demo of some folks running it in Apple Vision Pro. Right.

00:39:30.200 | No specialized hardware. Right.

00:39:32.440 | And the key is, you know, if we're going to run it on the phone,

00:39:35.080 | we got to compress all of that intelligence into a smaller

00:39:39.080 | and smaller model that's less power consumptive. Right.

00:39:42.600 | If you put one of these larger models on, it burns up the battery,

00:39:46.080 | burns up the phone, too much heat, you know, generated by that.

00:39:49.720 | Bill, you referenced a quote, you know, Zuck from Dworkash, where he said,

00:39:53.920 | I don't think in the future we're going to be primarily shoving

00:39:56.680 | all these things in the context window to ask more complicated questions.

00:40:00.160 | There will be a different stores of memory or different custom models

00:40:03.720 | that are more personalized to people.

00:40:06.520 | One of the things that I was most intrigued by in that interview

00:40:09.840 | was his focus on the personalization to people.

00:40:13.120 | He went so far as to say, understanding the content around emotions

00:40:17.480 | is a different modality unto itself, which got me thinking, you know,

00:40:23.040 | not only are they producing smaller models, but they probably have

00:40:26.800 | the largest store of human emotions.

00:40:29.280 | What reactions to one another emoticons to one another?

00:40:34.240 | You know, biggest, certainly social graph on the planet,

00:40:38.040 | which seems to put them in a really good position

00:40:41.280 | when it comes to this personal assistant that we all talk about.

00:40:45.400 | I know you your view is we're not going to get anywhere close

00:40:48.560 | until we get memory and we haven't solved memory.

00:40:51.440 | Well, I mean, he hinted at it, but everyone hints at it.

00:40:54.320 | It comes up a lot.

00:40:55.520 | And there's a and I mean, I push it to Sonny, but it's unclear

00:41:01.360 | whether you can accomplish what people hope

00:41:03.960 | to be achieved in a personal assistant with rag, with fine tuning,

00:41:09.120 | or if you really need a model to be actually,

00:41:12.960 | you know, trained on my data.

00:41:15.720 | And that latter part, no one knows how to do a fat cost effectively.

00:41:19.800 | Yeah. So I don't know.

00:41:21.800 | I don't know what pieces have to fall in place for us to get to that place.

00:41:25.520 | Yeah. And there's no secret.

00:41:27.760 | Like everyone seems to be aware that that's the end goal.

00:41:31.240 | But I don't I think there are a few breadcrumbs that were.

00:41:34.600 | I don't know if you say, you know, I'll suggest some of them.

00:41:37.600 | Maybe you can say a few breadcrumbs that were dropped out there,

00:41:40.520 | both by Apple and by by Zuckerberg in this regard.

00:41:44.680 | I mean, I'll just kick it off by saying what he said in that podcast is like,

00:41:48.520 | listen, in the first instance, what we do is we build software around the model

00:41:53.400 | that kind of hacks this stuff together and we see kind of what works.

00:41:57.560 | And so, yes, in the first instance, it may in fact be

00:42:01.520 | you have a really small model, you do some rag on it.

00:42:04.480 | Maybe in certain instances it communicates with a more sophisticated model.

00:42:08.880 | But in that, you know, in that rag can be a lot of personal information.

00:42:13.480 | I think Apple has said the same thing.

00:42:15.880 | But then what he importantly said is if that works,

00:42:19.240 | then on the next go around, we figure out how to build that into the model itself.

00:42:23.560 | Yeah, I think building on that, if you look at the breadcrumbs

00:42:28.160 | from, you know, all the major folks, and I think there was like a

00:42:30.960 | a Wired article that came out where Sam said, you know,

00:42:33.920 | the next model necessarily won't be bigger.

00:42:36.080 | I think he did say that.

00:42:37.200 | I thought that was interesting.

00:42:38.240 | Yeah. And I and the reason is,

00:42:41.240 | and you know, you had a thing, Brad, last year at the Barn

00:42:45.280 | where you had Brad Lightcaps being right.

00:42:48.400 | And the general message that keeps coming out of the open

00:42:52.800 | AI contingent is that customization and memory.

00:42:56.960 | And so my and I don't have anything beyond this, but I would say my guess is

00:43:01.200 | that's what they focused on with GPT-5.

00:43:04.960 | That's an important point.

00:43:06.360 | Like, I think in GPT-5, it's not going to be the final state,

00:43:10.640 | but I think you're going to see the beginnings of memory

00:43:12.480 | and the beginnings of actions. Right.

00:43:14.280 | And this is, you know, months away.

00:43:15.720 | And you and I have a bet on this. I know.

00:43:18.040 | Well, well, yeah, but that could be a major another, you know, tremor.

00:43:22.200 | But one is one interpretation of the statement

00:43:25.640 | that the models aren't going to get bigger.

00:43:27.720 | One, it could be a mea culpa to the thing.

00:43:30.360 | Like, OK, like, I don't want to play this game anymore.

00:43:33.480 | But two, it could mean that the the LLM

00:43:37.680 | training has kind of just run its course and you got to go do this next thing.

00:43:43.240 | But the next thing is not a necessarily an exponential leap.

00:43:50.000 | It may it may be like an early alpha or beta,

00:43:53.160 | and it may be a little more stumbly as you.

00:43:55.800 | I don't think I see little evidence that the scaling has run its course.

00:44:00.640 | I mean, like the smartest people on the planet

00:44:02.840 | who are putting their own money, real money up against this.

00:44:05.880 | Elon is building a much, much bigger cluster to train a much, much bigger model

00:44:11.640 | as is open AI, as is Zuckerberg.

00:44:14.280 | I mean, what Sam just said, the bigger models aren't.

00:44:17.680 | Well, I mean, he may be doing the same game that everybody else is doing.

00:44:21.320 | I'm trying to throw everybody off the scent of building a bigger model.

00:44:25.200 | Why is he trying to build his own chips, nuclear power plants

00:44:28.360 | and everything else if he's not going to build big models?

00:44:30.960 | I mean, you only don't take him at his word.

00:44:32.960 | Well, I'm just saying that I think the world, as I said earlier,

00:44:37.040 | is bifurcating into two like a world of specialized models.

00:44:41.280 | We are going to have very large frontier models.

00:44:43.640 | There will be a point at which you hit the you hit diminishing returns.

00:44:47.240 | Jan LeCun has said we're going to need a different architecture to get to AGI.

00:44:50.840 | He speculated that it's probably two or three generations

00:44:54.040 | more of scaling before we get to that point

00:44:57.480 | where it no longer makes economic sense to continue to scale it.

00:45:01.280 | But we're going to I mean, that's a lot of if it continues apace.

00:45:05.040 | That's a lot of developments over the course of the next two, three

00:45:09.280 | generations before we hit the upper limits of that.

00:45:11.960 | And by the way, I think we're already seeing some creative things

00:45:15.280 | like the data scaling that we saw, you know, past the chinchilla point.

00:45:20.080 | Those are really creative innovations to get around

00:45:22.960 | or to augment kind of the compute problem.

00:45:25.640 | So to me, I come back to this and I, you know,

00:45:30.840 | it makes me really excited again about where we are

00:45:35.000 | in this state of consumer search,

00:45:37.880 | you know, and personal assistance.

00:45:41.120 | Google's probably innovating better than they ever have

00:45:43.520 | because they're pushed out of their monopoly position by everybody else.

00:45:47.080 | Now, it sounds like, you know, they're seeing some great results

00:45:50.840 | come out of that.

00:45:52.440 | You know, I thought it was really interesting, you know, when you see,

00:45:55.880 | you know, Dolly and and and David Woodland, who's the product lead on on

00:46:00.640 | on Metaglass's talk about what they announced this week.

00:46:03.640 | Now it has meta AI with vision.

00:46:05.720 | It's, you know, now available to everybody.

00:46:08.080 | You know, not only can you use these things to call and to message

00:46:12.000 | using WhatsApp, but as all these integrations and these overlays.

00:46:16.800 | So, I mean, we haven't seen this kind of shake up in the world of search

00:46:20.840 | and in the world of consumer products in a while.

00:46:24.000 | And now, you know, there was this all this noise this week about Humane.

00:46:29.080 | Yep. Right. The startup up in San Francisco.

00:46:31.680 | And, you know, it got panned in a consumer review,

00:46:34.440 | you know, and one of the biggest challenges with that product,

00:46:39.200 | because I use the product as well, right, is the models weren't small enough.

00:46:42.880 | It doesn't have it can't run the inference on device.

00:46:46.560 | So it has to go out to the cloud to do it.

00:46:48.400 | And the second you have to go out to the cloud, it ruins the experience

00:46:51.560 | because now you have latency.

00:46:52.840 | We're a year away, probably max from that thing, being able to have a billion

00:46:57.560 | parameter or 500 million parameter model that basically has all the capability

00:47:01.640 | you need it to have. Totally agree.

00:47:03.880 | And we're also compressing

00:47:07.000 | the amount of time that it takes to go out to the cloud

00:47:11.080 | because we'll get those models to start running faster.

00:47:13.320 | So we're going to see a convergence there on two fronts, the local

00:47:16.840 | and then the ability for that model to reach out in the cloud

00:47:19.440 | and get a faster response out of the cloud.

00:47:21.160 | That's what I think, you know, is being underestimated.

00:47:26.240 | You know, just swinging back around to model size, right.

00:47:30.000 | The smaller models run faster just naturally.

00:47:32.680 | And so that gets you to faster responses.

00:47:36.280 | And we know the Internet's been on a huge push for lower latency

00:47:40.720 | across whether it's loading web pages or search results or whatever it is.

00:47:44.160 | And so I think we're starting to see a push in that direction.

00:47:46.600 | We all got kind of comfortable with the pace of chat GPT.

00:47:51.280 | But if you kind of go away for a second and try one of these smaller models

00:47:54.800 | somewhere else and go back to chat GPT, you'll really have that.

00:47:58.040 | Like we all had that moment for a bit between dial up and high speed Internet

00:48:01.680 | where we maybe had dial up at home, still in high speed at work.

00:48:04.280 | That's the feeling that you get when you switch between those two things.

00:48:07.440 | One of the debates I know, Sonny, you've been having, you know, with our team

00:48:11.000 | and I'm firmly, you know, in your camp on is, you know, this idea

00:48:15.680 | dating back 20 years when it comes to consumer products, even,

00:48:20.640 | you know, speed improvements that are barely perceptible

00:48:25.080 | at Google have pretty important implications for their revenue.

00:48:29.480 | And so I think what we're seeing with these smaller models

00:48:34.160 | and all of these other developments and you guys are helping,

00:48:36.480 | you know, certainly to lead the way at Grok, you're just seeing massive

00:48:39.800 | improvements in token per second.

00:48:41.840 | And I think, you know, when you start having agents talk to agents,

00:48:45.440 | you take humans out of the loop, right?

00:48:47.920 | Now, computers can talk really fast to one another,

00:48:51.200 | but we have to have low cost, fast inference, you know, that's able to support that.

00:48:55.360 | We do. And we think of the use cases that we all like.

00:48:58.000 | And I think we all love perplexity.

00:49:00.120 | But you think about, you know, what happens behind the scenes

00:49:02.920 | when you type like a small request, it shoots off something

00:49:06.200 | into a couple of different places, including a search results, pictures

00:49:09.440 | and all those kind of things that all that has to be processed by the LLM

00:49:12.440 | like really quickly.

00:49:13.600 | Did you play with the Meta AI picture generator?

00:49:17.080 | I did. Where you just add another little word or add to it?

00:49:20.200 | Yes. What do you think of it?

00:49:21.320 | That speed is insane.

00:49:23.200 | Yeah. And compared to like a year ago when you were doing it,

00:49:26.560 | when you'd wait, you know, 15 seconds to get one.

00:49:28.640 | Right, for the next image.

00:49:30.240 | This gets back to Sonny's point as well.

00:49:32.040 | When you're doing that, just think if the cost of inference was really high,

00:49:35.960 | there's no way he could roll that out to three billion people, right?

00:49:39.840 | Because all of a sudden people would start playing with it.

00:49:42.120 | And his OpEx on, you know, on the business would blow up in his face.

00:49:45.600 | Part of the reason he's, I think, pushing toward these smaller models,

00:49:49.600 | opening these models, you know, and he said in that podcast interview,

00:49:53.440 | they helped me lower the cost of inference.

00:49:56.000 | You know, we eat the cost of training.

00:49:58.400 | So, you know, we can we can lower the cost of inference.

00:50:01.360 | Well, maybe to, you know, just to wrap, we can.

00:50:04.880 | I want to hit on a few topics, Sonny, that we've covered

00:50:08.160 | over the course of the last few weeks.

00:50:10.480 | Of course, Bill and I did have been doing a couple of deep dives

00:50:13.840 | on full self-driving at Tesla, as well as,

00:50:18.000 | you know, their ride share project that's now moved front and center

00:50:21.800 | because of the breakthroughs they've had on FSD.

00:50:24.080 | And on the Tesla earnings call this week, they answered some of our questions.

00:50:28.960 | So a couple of the questions Bill and I had is,

00:50:31.440 | is this going to occur within the Tesla app?

00:50:33.480 | Well, you can see here, you know, this beautiful depiction,

00:50:37.360 | you know, of a ride share within the Tesla app.

00:50:40.800 | There are a lot of Tesla app holders.

00:50:42.920 | We had a question was whether or not it was going to be owned and operated

00:50:45.560 | or whether it was going to leverage

00:50:46.720 | the millions and millions of cars that are out there in the fleet.

00:50:49.080 | And Elon, I thought, elegantly put this, you know,

00:50:51.760 | we're going to be both Uber and Airbnb.

00:50:54.000 | You know, we're going to, you know, own some of the fleet.

00:50:57.400 | We're also going to let those people who buy cars from us,

00:51:01.480 | you know, put their cars into the fleet.

00:51:04.240 | My own hunch is that it will also be distributed both one P and three P,

00:51:08.880 | although he didn't go so far as saying that.

00:51:11.680 | And what I mean by that is not only distributed in the Tesla app.

00:51:14.600 | My hunch is that as this scales, it'll make sense to do a partnership with Uber.

00:51:18.600 | And frankly, I wouldn't be surprised to see some of the people

00:51:21.760 | who operate on the Uber platform become fleet operators of Tesla's for Tesla.

00:51:27.200 | And so I think there's a really interesting opportunity

00:51:30.640 | for an integration there.

00:51:32.680 | But I thought that was pretty consistent.

00:51:34.480 | We weren't too far off in terms of our estimation there.

00:51:38.800 | And there's I mean, this is kind of they obviously have already made it clear

00:51:44.000 | they're going to be talking about this for a long time.

00:51:45.880 | But this is the kind of first draft, if you will.

00:51:49.360 | I think there's a lot to see as this stuff rolls out.

00:51:52.160 | You know, Waymo's had to apply for these licenses

00:51:56.200 | to get these cars on the street. Yes.

00:51:58.240 | We don't have Tesla's, one, they haven't even applied for those things,

00:52:03.160 | but we don't have them driverless on the road yet.

00:52:07.360 | Right. Which would be a step that would need to take place

00:52:11.560 | before this was rolled out.

00:52:12.640 | But it intersects with that really big purchase of,

00:52:16.600 | you know, H100s that they talked about as well.

00:52:19.640 | You know, we now have a lot of companies that are reported

00:52:24.960 | and there's not one of them yet that has not raised their capex guidance

00:52:30.080 | to buy more to buy more GPS.

00:52:33.040 | I mean, Elon himself is going to let his in the past four weeks,

00:52:38.200 | the incremental purchases they've signaled are in and not just Tesla.

00:52:44.080 | But there's data on the internets today that X.AI has raised six billion.

00:52:50.400 | Presumably most of that's going into infrastructure as well.

00:52:54.920 | And, you know, you and I have a couple of bets going.

00:52:58.040 | But, you know, when it comes to whether or not

00:53:01.240 | GPUs are undersupplied or oversupplied, you know,

00:53:03.840 | what I've stipulated is every supply shortage

00:53:06.880 | does ultimately result in a glut.

00:53:08.920 | But people have been calling for this glut now for,

00:53:11.960 | you know, 12 months anyway.

00:53:14.680 | And they're calling for it again this year.

00:53:16.080 | We're not going to see it again this year.

00:53:17.480 | There's, you know, and-

00:53:18.560 | So you bought the dip?

00:53:19.480 | We own plenty.

00:53:21.480 | And, you know, and you just see it, you know, in fact,

00:53:25.800 | you know, Meta was down 15 or I think it ended up down 10 or 11 percent.

00:53:31.640 | And one of the major reasons it was down is Zuckerberg said,

00:53:34.840 | I'm going to put the, you know, the accelerator to the floor.

00:53:37.560 | He increased the midpoint of his CapEx guide by three or four billion dollars.

00:53:41.400 | You know, which I said, you know, I had a lot of people inbound to me and say,

00:53:45.400 | hey, you know, they're no longer being efficient or they're no longer being fit.

00:53:49.040 | Which, you know, to which I responded by saying in two years,

00:53:53.080 | that company has gone from 22 billion in net income to 55 billion in net income.

00:53:58.120 | They've reduced their headcount from 85,000 people to 69,000 people.

00:54:02.160 | What they are demonstrating is what you can do when you're efficient.

00:54:06.840 | You can redeploy all of that incremental profitability into investing,

00:54:11.040 | not in some 10 year project that we don't know what the payback is,

00:54:14.400 | but directly into GPUs and AI, where you can see the payback

00:54:17.640 | in a pretty short period of time, leveraging it in their core business.

00:54:20.920 | And so, you know, it was, you know, while we're on it,

00:54:26.520 | you know, well, he went and bought gym equipment.

00:54:28.680 | I did see that, too.

00:54:32.760 | I did see that, too.

00:54:34.200 | Bill, you and I talked about IPOs.

00:54:36.320 | You know, Dan Primek, you know, came out with this article

00:54:40.800 | that was pretty controversial, I think, among VCs.

00:54:43.640 | I saw a lot of people responding to it.

00:54:45.560 | It was entitled VCs, you're blowing it.

00:54:48.960 | And there was one line in there that caught my attention,

00:54:52.800 | you know, where he said VCs let startups stay private too long,

00:54:57.360 | often well past their hyper growth phase that justified sky high valuations.

00:55:01.680 | You and I debated this last week.

00:55:04.480 | How much revenue do you have to have to go public?

00:55:07.640 | I think you and I are both in the camp

00:55:10.240 | that if you have 100 million of trailing revenue, you're growing well.

00:55:14.720 | You have great unit economics.

00:55:16.160 | You can certainly go public if you price it right.

00:55:19.040 | I'm you know, I've said it on Twitter.

00:55:21.520 | I say it in boardrooms.

00:55:22.720 | I think being in the public markets is a great place for companies to be.

00:55:27.400 | I think it, you know, it puts them in the big league.

00:55:30.080 | It makes them you know, there's plenty of room to innovate there.

00:55:33.240 | However, what I would say to Dan is, you know, when we sit on the board,

00:55:37.880 | we can advise, but ultimately we're not the decision maker.

00:55:42.640 | Right. It's got to be a collaboration with the founder of the company.

00:55:46.200 | And ultimately, I think the company should go public

00:55:48.800 | when it's the right time for the company to go public.

00:55:51.200 | And for some companies, that is at that earlier phase.

00:55:53.920 | I think a lot more could go public at that earlier phase.

00:55:57.160 | But I also think there are certain situations, you know, take SpaceX,

00:56:02.440 | for example, where I think it's behooved them to stay private longer. Right.

00:56:06.560 | And they've had plenty of access to the private markets to raise capital.

00:56:09.800 | So I didn't know if you had any reactions to the Dan.

00:56:12.360 | Well, I mean, I think

00:56:15.720 | in a lot of ways, I agree with what you're saying.

00:56:18.320 | And I disagree maybe with the way it was positioned.

00:56:21.360 | But keep in mind, Dan's one of the very few

00:56:24.160 | analysts and writers that focuses on LPs

00:56:28.120 | like most of these writers focus on VCs or the founders themselves or whatever.

00:56:34.040 | And he's he's constantly talking to LPs.

00:56:36.760 | And I think there is a a very real situation, especially where we came out

00:56:42.000 | of Zerp, where there's a vast amount of paper marks

00:56:46.320 | that are sitting on these LP books that are aging out,

00:56:49.880 | that are that are exposed to dilution, you know, on an annual basis.

00:56:55.040 | And I suspect they're very nervous and I suspect they're talking to him

00:57:00.200 | and that that's where he's building this thesis.

00:57:02.480 | And I think that's probably right.

00:57:04.720 | I also think that a number of people, you know, that invest in late stage

00:57:09.920 | and people that we know have have built a business model where

00:57:13.440 | they kind of like companies staying private longer.

00:57:16.560 | They are the ones that everyone talks about.

00:57:18.480 | Amazon went public at this price and then the public captured it.

00:57:21.760 | They they kind of view their game as capturing

00:57:25.400 | that growth instead of the public markets.

00:57:27.720 | And the third thing I would just say is our business has got nothing

00:57:32.080 | but more competitive from the minute I entered it to today.

00:57:35.400 | And I think that's going to keep happening.

00:57:37.360 | And that competition forces people to be very founder friendly,

00:57:42.480 | to say what they want to hear, to support secondaries, which we've talked about,

00:57:46.880 | that when you support massive secondaries,

00:57:51.160 | you're taking the number one pressure

00:57:53.200 | out of the system that used to lead founders to to want to go public

00:57:59.040 | because their employees are like, I need liquidity, I need liquidity.

00:58:01.480 | So you do a release valve and you take that away.

00:58:05.520 | And and I do think there will be a lot of situation.

00:58:09.840 | And then actually one last thing to mention, just

00:58:12.240 | because of where we came from, evaluation perspective,

00:58:15.240 | we know a lot of people are sitting there afraid

00:58:18.960 | that they can't meet their last mark. For sure.

00:58:21.120 | And so then you're kind of in Never Neverland.

00:58:23.520 | And how do you get out of this?

00:58:24.720 | And where are the odds you're going to grow back to that?

00:58:26.840 | And there's kind of a lack of.

00:58:28.560 | So anyway, those dimensions, I think he's hitting at it right.

00:58:31.960 | I agree with you.

00:58:33.320 | And this is where I think you got it wrong, is that like no single VCs

00:58:37.000 | going to stand up and make a company go public, right?

00:58:40.120 | That's not going to happen.

00:58:41.200 | I mean, I do see the market evolving.

00:58:43.080 | Listen, you got to get you're in you raise 10 year funds.

00:58:47.240 | You need to get liquidity.

00:58:48.880 | You're in the business to provide returns and liquidity to your partners.

00:58:53.040 | If I look at the private equity business, right, they evolved in such a way

00:58:57.480 | where they didn't have to take the company's public.

00:58:59.480 | They would just sell to another private equity company.

00:59:02.160 | And it may very well be in the VC landscape, Bill.

00:59:04.760 | And I see this more and more.

00:59:07.480 | You know, I know a big company right now raising it over 10 billion.

00:59:11.160 | And I know a lot of early stage VCs who are selling into that round.

00:59:16.320 | Right. And so that I thought there was one.

00:59:19.280 | Didn't Rippling say 600 million was going to people that are early?

00:59:23.720 | So that's not the one I was referencing.

00:59:25.840 | But Rippling may, in fact, be one of those.

00:59:27.880 | And so the market may be responding to some of these imperatives.

00:59:32.200 | You know, by LPs to get liquidity.

00:59:34.240 | But what I would say to Dan is certainly with the with the venture

00:59:39.360 | capitalists sitting around this table, you have two people who think

00:59:42.840 | that the public markets are a great place for companies to innovate,

00:59:45.520 | companies to thrive, companies to raise capital, companies to recruit,

00:59:48.920 | build, build brands, et cetera.

00:59:50.720 | And and we know we're in the business of liquidity.

00:59:53.360 | Seeking as an entrepreneur, though, can I add one thing to that, which is

00:59:56.760 | I think we're going to see like a bifurcation

00:59:59.440 | and builds on our conversation from earlier.

01:00:01.480 | There's going to be companies that are cheaper and cheaper to build and run.

01:00:04.520 | And, you know, that has one impact.

01:00:07.200 | But there's companies that are going to get a lot more capital to build and run.

01:00:10.160 | And so that may force folks into the public markets sort of like the way

01:00:13.880 | it was in the late 90s with a lot of those business

01:00:15.720 | because they needed a lot more capital.

01:00:17.280 | What do you guys think about that?

01:00:18.480 | I there was a there was a interesting article

01:00:22.880 | that Tim O'Reilly wrote about a month ago.

01:00:26.240 | I put the link in there, but he implied that he also attacked the VCs

01:00:30.920 | and said they were going about it all wrong.

01:00:32.600 | But he he implied that the AI was now in its Uber phase where the, you know,

01:00:38.560 | talk about Uber and Lyft and DoorDash all raising billions

01:00:42.520 | and it's spilling out on the floor.

01:00:44.480 | And I personally think part of it is simply

01:00:47.880 | a recognition by the investment community

01:00:52.040 | writ large that network effects exist and increasing return exists.

01:00:57.360 | Yes. And so when they think this is the next big thing

01:01:01.520 | and they see open AI take a lead, their gut response is,

01:01:05.880 | well, if I had invested early in Amazon or Google or whatever,

01:01:09.040 | I'd get paid almost no matter what the price.

01:01:11.520 | And so it's the institutionalization of a belief in network effects

01:01:16.200 | that's leading to the the money pouring in.

01:01:18.960 | And then and then it's a competitive dynamic.

01:01:21.280 | Like once your company raises, you know, 200 million, a billion,

01:01:25.760 | if you're in that market, you raise it, too.

01:01:28.000 | And it does create chaos.

01:01:29.920 | Like, I do think it creates chaos.

01:01:32.240 | Maybe we'll just we'll end with just kind of the volatile week

01:01:37.240 | it's been in markets.

01:01:39.280 | You know, we talked to, you know, you asked me what I thought

01:01:42.680 | was going to happen this week.

01:01:44.920 | You told me you'd give me a scorecard.

01:01:46.360 | But, you know, the reason it's so volatile this week

01:01:51.240 | is not just because we've had some mixed earnings reports

01:01:54.600 | or at least mixed reactions.

01:01:57.000 | But I think the economic backdrop is, you know, is unsettling.

01:02:00.720 | GDP came in a lot weaker than expected this morning.

01:02:04.440 | At the same time, the PC for Q1 came in a little bit higher.

01:02:09.120 | Now we got the monthly PC report coming out tomorrow.

01:02:11.800 | We'll see where that shakes out.

01:02:13.240 | But this idea that we could have a slowing economy

01:02:17.000 | at the same time that we have inflation continuing to go up.

01:02:20.080 | Right. This is this very fearful place called stagflation

01:02:23.600 | that nobody wants to be in.

01:02:25.160 | Now the market is now pushed out the rate cut forecast

01:02:28.320 | to December of this year. Right.

01:02:30.480 | So this higher for longer is now in place.

01:02:32.960 | Remember, when we started the year, we thought we were going to have six

01:02:35.200 | rate cuts, very accommodating Goldilocks environment.

01:02:38.520 | What's surprising to me, to be perfectly honest, is how well

01:02:42.080 | technology stocks have performed outside probably software,

01:02:46.000 | but how well they perform notwithstanding this fact.

01:02:48.840 | And the only reason they've been able to do that

01:02:50.960 | is the reacceleration caused by AI.

01:02:53.600 | So if you look at what happened tonight, Google beat.

01:02:56.600 | They issued a dividend for the first time.

01:02:58.640 | They announced a buyback. Right.

01:03:00.480 | So they're really they're their margins are expanding.

01:03:03.920 | So they're finding efficiencies in that business.

01:03:06.000 | They're listening to the markets.

01:03:08.360 | But, you know, I think impressively,

01:03:11.120 | you know what they're doing in that core business, the things they announced

01:03:14.160 | around SG, you got to give that management team a lot of credit.

01:03:17.160 | We talked about what's the stock.

01:03:20.160 | The stock was up, I think, 10 or 15 percent after hours.

01:03:23.160 | So I think it's at an all time high and and and doing incredibly well.

01:03:27.480 | Metta, you know, missed is down 10 percent, but still up 30 percent on the year.

01:03:32.840 | So I imagine those those two companies are about in the same area.

01:03:36.440 | You know, year to date, we talked about why they why they got hammered

01:03:40.240 | is because, you know, they're going to invest even more aggressively in A.I.,

01:03:43.920 | which is a long term investor.

01:03:45.400 | I'm pretty thrilled about.

01:03:46.560 | And then if you look at Azure or Microsoft's quarter,

01:03:49.240 | it was pretty blockbuster.

01:03:52.000 | And we'll show this chart by jamming.

01:03:53.640 | But Azure, I contributed seven percent of growth this quarter.

01:03:56.920 | So it's now translates into about a four billion run rate business,

01:04:01.840 | you know, that didn't even get broken out until five quarters ago, you know.

01:04:06.520 | So, again, I think if you look at technology generally, it's performing.

01:04:12.440 | It's performing really well, despite this kind of volatile economic backdrop.

01:04:17.400 | And we'll see where where PC rolls in on Friday.

01:04:22.720 | We'll see where the rest of technology comes in.

01:04:25.600 | My hunch is that the largest companies in technology,

01:04:29.240 | back to your network effects and your scale advantages.

01:04:32.240 | I'm not sure that smaller tech technology

01:04:35.280 | companies are seeing the benefits that the largest technology companies.

01:04:39.160 | Certainly, it looks like, you know, so we'll see that as it reports.

01:04:42.600 | But I think the largest data platforms and hyperscalers continue to benefit.

01:04:46.560 | Boys has been fun. Yeah, great.

01:04:49.320 | Let's do it again. Thanks for being on, Sonny.

01:04:51.680 | Until next time.

01:04:53.560 | Bye.

01:04:54.080 | I.

01:04:55.080 | I.

01:04:56.080 | I.

01:04:57.080 | I.

01:04:58.080 | I.

01:04:59.080 | I.

01:05:00.080 | I.

01:05:01.080 | I.

01:05:02.080 | I.

01:05:03.080 | I.

01:05:04.080 | I.

01:05:05.080 | I.

Ep8. AI Models, Data Scaling, Enterprise & Personal AI | BG2 with Bill Gurley & Brad Gerstner

Chapters