# The Architects of Value: Mark Edelstone and Colin Stewart on the Economics of Silicon Valley

https://www.youtube.com/watch?v=B0e8er_AD7Y

[00:00] Today, we're going to do something a bit
[00:01] different. We're going to host a master
[00:03] class.
[00:04] We're going to talk about markets,
[00:06] cycles, capital intensity, and the
[00:08] question sitting underneath everything
[00:11] right now.
[00:12] Is AI a revolution, a bubble, or both?
[00:16] I recently hosted an episode with Zach
[00:18] Dychtwald about China. And we explored
[00:21] how large systems evolve. They move
[00:23] slowly until suddenly they don't.
[00:26] AI feels like one of those inflection
[00:28] points. Capital spending is exploding.
[00:31] Semiconductor companies are now among
[00:33] the most valuable businesses in the
[00:35] world. Governments are treating chips
[00:37] like strategic weapons. And investors
[00:40] are trying to answer a deceptively
[00:42] simple question. Are we early or are we
[00:45] euphoric?
[00:46] History tells us something
[00:48] uncomfortable.
[00:49] Revolutions are real, bubbles are real,
[00:52] and sometimes they happen at the same
[00:55] time.
[00:56] So today, we sit down with two Morgan
[00:58] Stanley market veterans, Mark Edelstone,
[01:01] managing director, and Colin Stewart,
[01:04] managing director and vice chairman.
[01:06] They'll help us understand this moment,
[01:08] how to think about it, where we are in
[01:09] the cycle, and what it means for markets
[01:12] and the world order.
[01:17] [music]
[01:21] All right, Colin and Mark, thanks for
[01:22] being here. Thanks. I appreciate it. I'm
[01:24] looking forward to it and appreciate
[01:26] being on A Bit Personal. Very excited.
[01:29] Thanks for having us.
[01:30] this dialogue is going to be a little
[01:31] bit different than our other episodes.
[01:33] This is going to be more like a master
[01:34] class on the markets, okay? Because I've
[01:36] had the opportunity
[01:38] to have on A Bit Personal
[01:41] people like Jensen Huang and Lisa Su,
[01:43] Matt Murphy. And you know, after you
[01:45] listen to these guys talk about AI, you
[01:47] get very excited about where we are. And
[01:49] you realize,
[01:51] okay, this is something enormous.
[01:53] But you guys have been through enormous
[01:54] before, right? So you've been through
[01:55] telecom, you've been through dot com,
[01:58] you've been through many, many
[02:00] semiconductor market cycles.
[02:03] And so what I would love is for you guys
[02:05] to kind of help us put into perspective
[02:07] the moment that we're in today.
[02:10] And you know, are we at the beginning of
[02:12] you know, something extraordinary? Are
[02:14] we in the middle of something
[02:15] overheated? And maybe before you answer
[02:19] that question, maybe start by giving us
[02:22] a little bit of background on that some
[02:23] of the cycles you've been through, the
[02:25] patterns that you've been through that
[02:26] help has have helped shape your
[02:28] perspective on this particular moment.
[02:30] And maybe start with you, Mark. Sounds
[02:32] good. Well, first off, your podcasts so
[02:34] far have been fantastic.
[02:35] Thank you.
[02:35] Uh so hearing Jensen, Lisa, Matt, and
[02:37] others has been really inspirational.
[02:40] Thank you. Uh so in terms of of my
[02:42] background seeing cycles, I've been
[02:43] doing this now since 1989. That's when I
[02:45] started covering the semiconductor
[02:47] industry.
[02:48] [snorts]
[02:48] And the first 18 years that was as a
[02:50] research analyst. It's really where I
[02:52] began to understand the details of the
[02:54] industry and really begin to build uh
[02:57] perspective. And the last 19 years has
[02:59] been as an investment banker. So I've
[03:01] seen an awful lot. I've lived through
[03:03] all these cycles, and it's been
[03:06] tremendous change that we've seen. I I
[03:09] love the place we're in today Mhm. Uh at
[03:11] this point, cuz if you go back in time,
[03:13] the industry was growing very, very fast
[03:14] historically, like typically five times
[03:16] faster than GDP.
[03:18] When I started the industry, it was only
[03:19] $50 million in size. Amazing.
[03:21] Next year, we're going to be or actually
[03:23] this year, we'll be at a trillion
[03:25] dollars this year. So run rate off of
[03:27] the first half of this year will be at a
[03:29] trillion dollars. So that's just
[03:30] unbelievable growth to see in an
[03:32] industry in one's career.
[03:34] And you know, lots of cycles. In the
[03:37] past, they were really driven by
[03:39] mistakes that people made, putting too
[03:41] much capacity relative to demand. Okay.
[03:43] Too many companies doing the same thing.
[03:45] So all the R&D dollars going after the
[03:47] same sockets. And so that led to price
[03:50] compression and just unabsorbed overhead
[03:53] and capacity and so on. That created a
[03:55] lot of the cycles. As we moved out over
[03:57] time and the complexion of the industry
[03:58] changed, we started to see things
[04:01] uh transform a little bit. Uh the
[04:03] industry used to be very vertical.
[04:05] Everybody had manufacturing. They did
[04:07] their own tools. They designed their own
[04:09] chips. Then of course, we went to a
[04:11] fabulous model, which you were a big
[04:12] part of,
[04:14] uh with the FSA and then moving to the
[04:16] GSA. And so that helped a little bit
[04:19] with just overall capital intensity for
[04:21] the industry.
[04:22] But um
[04:24] it still is an industry that's very much
[04:25] tied to the economy.
[04:27] And as we get bigger and bigger today,
[04:29] we're just slightly smaller than 1% of
[04:32] global GDP.
[04:33] Okay. I think it should be worth a lot
[04:34] more than that. So it will be over time,
[04:36] but we're still fairly cyclical off of
[04:38] what happens in the economy, because a
[04:40] lot of the industry is consumed by
[04:43] consumer, automotive, PCs, handsets, and
[04:47] so on. This cycle's a little bit
[04:48] different. This cycle's being driven by
[04:51] massive
[04:52] spending for AI, and we're early on, and
[04:56] the rate of change of everything is just
[04:58] so massive, which I'm sure we'll talk
[04:59] about as we go through here.
[05:01] But it's just it's just incredible
[05:03] what's happened, and it's increased the
[05:05] value and the type of drivers that we're
[05:09] getting out of the semiconductor
[05:10] companies today.
[05:12] Uh it's already pretty pervasive in
[05:14] terms of the usage of semiconductors.
[05:17] But this unbelievable growth engine
[05:20] right now for AI, which is helping
[05:22] companies like Nvidia and Micron and
[05:24] SanDisk and Marvell and many others, uh
[05:27] it's truly tremendous. And so it is
[05:29] different today. Okay. Um I think it's
[05:31] very sustainable. Like we're early on in
[05:34] all the drivers need to happen to
[05:36] support AI.
[05:38] That's just in the data center. So we've
[05:40] trained up all these large language
[05:42] models. We're now taking advantage of
[05:44] the inference side of it. But then
[05:45] there's a whole another big driver going
[05:47] to happen with physical AI. Okay. And
[05:49] that's still in the future, and there's
[05:51] still a lot more around it. So things
[05:53] have changed, but I think we're in a
[05:55] really, really good spot for this
[05:56] industry.
[05:57] Yeah, I think that the first time that
[05:59] we heard that the industry was going to
[06:01] a trillion by 2030, Mhm. I mean, people
[06:04] were like, yeah, right. I mean, it was a
[06:07] it was a big forecast.
[06:09] Yeah. One and I think the first time I
[06:11] heard it was maybe 3 years ago. Exactly.
[06:13] And I would have said the exact same
[06:15] thing, and yet it's going to happen in
[06:16] 2026, which is pretty amazing.
[06:19] is. And so it's still cyclical. Of
[06:21] course, you know, we lived through
[06:21] COVID, very cyclical. So if people let
[06:24] supply chains get uh you know, out of
[06:27] whack, get inflated, build inventories,
[06:30] that's a problem. That can always
[06:32] happen. Um you know, COVID was our last
[06:35] real cycle, and it took a while to sort
[06:37] of work its way through. That's now
[06:40] behind us, and we'll see where this
[06:42] cycle goes, but I think we're going to
[06:43] find it's going to probably last longer
[06:45] than what people are currently, you
[06:47] know, concerned about.
[06:48] Okay.
[06:50] Okay, Colin, how do you feel about
[06:51] everything right now? So uh just
[06:53] quickly, so like like Mark, I've been
[06:55] doing this a long time. This is my 38th
[06:57] year at Morgan Stanley. And while Mark
[06:59] has been more focused on semiconductors,
[07:01] I've been focused more broadly on either
[07:03] emerging markets or technology. So I've
[07:05] seen a lot of cycles in those 38 years.
[07:08] So I've seen everything from, you know,
[07:11] crises caused by geopolitics, whether
[07:13] it's the first Iraq War, the second Iraq
[07:15] War. I've seen Asian financial crisis in
[07:17] the late '90s, and then sat through the
[07:19] internet bubble,
[07:21] sat through the global financial crisis,
[07:24] maybe minor SaaS crashes along the way,
[07:26] but now and now, you know, to AI. So
[07:29] I've seen these paradigm shifts happen,
[07:30] and I have sort of a rubric that I that
[07:33] I use, which is every time you get these
[07:35] paradigm shifts, there's a 10x increase
[07:37] in the capital investment for these
[07:39] cycles. And so if you go back and you
[07:41] look at the move from whether it's, you
[07:44] know, mainframe to PC to internet to
[07:46] mobile to cloud to AI, there's always
[07:49] been this large step up in investment
[07:51] cycle. And if you take the cloud in
[07:54] total, it's probably about a trillion
[07:56] dollars of CapEx. So that means the
[07:58] order of magnitude, one order of
[08:00] magnitude more is is AI. That's a $10
[08:02] trillion CapEx bill.
[08:03] Amazing. If you take the chat GPT moment
[08:06] from that point to to end of 2027, it's
[08:10] around two and a half trillion of CapEx
[08:12] at the big hyperscalers and some of the
[08:14] other players in the market. So that
[08:16] would tell you we're kind of in the very
[08:18] early innings of of that move. And so
[08:20] like Mark, I think I believe that
[08:22] there's a big wave that's coming, and
[08:24] it's going to have its own cycles, for
[08:26] sure.
[08:27] But this wave is a little bit different
[08:29] than prior prior waves, in particular
[08:31] with prior bubbles. Um when to have a
[08:34] bubble, I think you need to have a bunch
[08:35] of conditions precedent. You know, one,
[08:37] you have to have a an amazing narrative,
[08:39] right? AI is going to change the way we
[08:41] work and the way we live. That's clearly
[08:43] a great narrative for, you know, the
[08:44] makings of a bubble. But the second
[08:46] piece
[08:47] is look, easy liquidity. You need a lot
[08:49] of capital, low interest rates, or just
[08:51] money that's sloshing around. Kind of go
[08:52] back to 2021, and look what happened to
[08:54] software and the tech market at the
[08:56] time. You know, when you had zero
[08:57] dollars because of ZIRP, or zero cost of
[09:02] valuations exploded.
[09:03] You also need high valuations. I mean,
[09:06] Nvidia is probably one of the cheapest
[09:07] growth companies you can buy in the
[09:08] world. It trades at 20-ish times, a
[09:12] little over 20 times earnings, GAAP
[09:14] earnings,
[09:15] um growing you pick your number,
[09:18] anywhere between 40 to 70% in the next
[09:20] couple of years. So you really don't
[09:21] have like valuations that seem
[09:23] stretched. You don't have easy money.
[09:25] You also don't have as much leverage in
[09:26] the system as we've had before. And so,
[09:29] you know, there are a lot of things that
[09:30] say this still could run.
[09:32] But I think the the the the thing that's
[09:34] probably the most defining thing that
[09:35] we'll be trying to figure out over the
[09:36] next year or two is
[09:38] all this money has gone into
[09:40] infrastructure, gone into capital
[09:41] expenditures to drive AI.
[09:44] The return is the big focus. Like where
[09:46] do you start to see the return for that
[09:48] investment? Is it manifested itself in
[09:51] either better revenue growth, better
[09:53] margins, increased productivity? Are you
[09:55] seeing what more wide-scale adoption
[09:57] across the enterprise? That's still kind
[09:59] of on the come or it's evidence is
[10:01] starting to show, but that's the big
[10:02] question. The market grapples with that
[10:04] on a daily basis.
[10:06] So in in 1999 when we had sort of the
[10:10] the internet
[10:13] the
[10:14] the markets were funded by equity. Okay,
[10:16] in 2008 maybe credit. So how are they
[10:18] being funded today? I mean and I ask
[10:21] this because everybody will say, well,
[10:22] the amazing thing is is that it's free
[10:24] cash flow. Yeah. But there was I guess a
[10:26] report last week with
[10:28] how meta defined free cash flow was
[10:30] maybe different than than others. So do
[10:32] you worry about that of how this being
[10:34] funded? I think that's I think it's an
[10:37] Look, I think free cash flow is a large
[10:39] portion of it. We Again, there's I'm
[10:40] sure that everyone has their own various
[10:42] definitions of free cash flow. We look
[10:43] at the balance sheet and the gap numbers
[10:45] for free cash flow.
[10:47] You know, companies like Microsoft,
[10:50] Amazon, Meta, Alphabet, you know, these
[10:53] companies their free cash flow, you
[10:55] know, historically has been somewhere
[10:57] between 60 to 80, 90 billion dollars a
[11:00] year. I mean Nvidia's free cash flow is
[11:02] over a hundred million dollars a year.
[11:04] So there's a lot of capital there.
[11:06] They've also historically not been big
[11:08] users of the debt markets.
[11:10] So when you think about the percent of
[11:12] the equity market cap that these
[11:14] companies represent and then you look at
[11:16] the percent of the credit markets, the
[11:18] debt markets that they represent, it's
[11:20] it's like polar opposites. You know,
[11:22] these four, five, these mag seven, mag
[11:24] eight, mag 10 40, 45% of the equity
[11:26] markets, but they're probably still
[11:28] single digit percent of the credit
[11:30] market. So they have an ability to
[11:32] retain investment grade credit, to find
[11:33] that credit. They're also starting to do
[11:36] more creative things, right? They're
[11:37] looking at ways that they can
[11:39] provide a guarantee to these facilities.
[11:42] So Google, Meta, you know, are using now
[11:45] kind of a look through to their credit
[11:47] quality to help finance some of the data
[11:49] centers by guaranteeing taking some of
[11:51] the off-take or they're using the
[11:52] compute
[11:53] or helping kind of figure out the
[11:55] funding for the equity piece of these
[11:56] these these these data centers. So I
[11:58] think it's it's definitely a question
[12:00] because when you look at that that
[12:01] number I mentioned, that two and a half
[12:03] trillion or 2.7 trillion, free cash flow
[12:05] only gets you
[12:06] a portion. It gets you a large portion
[12:08] of it, but then there's also
[12:11] there's debt, there's going to be sort
[12:13] of, you know, structured debt, and then
[12:14] there's also just equity, too, which is
[12:16] coming from sovereigns, coming from
[12:18] large
[12:19] um you know, players in the kind of in
[12:21] you know, institutional market. So
[12:23] it's not
[12:25] anywhere close to the things we've seen
[12:27] from a crisis. You know, like when you
[12:28] talk about bubbles where there's been a
[12:29] credit-driven or leverage-driven crisis,
[12:32] again, these are the most
[12:33] high quality, you know, invest all high
[12:36] investment grade companies that are
[12:37] leveraging their their their their
[12:39] incredibly strong businesses to invest
[12:41] in a once-in-a-lifetime opportunity.
[12:43] Doesn't feel like they're doing things
[12:44] today that are risk
[12:46] 25 years I've heard CEOs ringing their
[12:49] hands about no one values us correctly
[12:53] and it it started changing I guess, you
[12:55] know, really about six years ago.
[12:57] Yeah, but
[12:58] great.
[12:58] But you know, to the point that Colin
[13:00] just made about say Nvidia as an
[13:01] example, uh Jensen and Nvidia has
[13:05] changed the world more than anything
[13:06] else. It's why they're the most valuable
[13:07] company in the world today. But they
[13:10] trade at a basically a market multiple
[13:13] despite growing dramatically faster. Um
[13:16] in fact, if you took the the 10 most
[13:19] valuable technology companies in the S&P
[13:22] 500,
[13:24] so that's obviously the hyperscalers
[13:25] plus Nvidia
[13:27] plus Broadcom plus Micron. So three of
[13:29] those 10 are semiconductor companies.
[13:32] And all but one of the other nine
[13:35] are doing semiconductors internally.
[13:38] Those companies collectively represent
[13:40] about 37% of the S&P 500.
[13:43] They're only 16% of revenues of the S&P
[13:46] 500, but 32% of profits. So they are
[13:50] growing fast, they're very profitable,
[13:53] and they deserve the valuations, but
[13:56] if you aggregated them, they only trade
[13:58] at a slight premium to the market today,
[13:59] which is amazing.
[14:00] trade at at or slightly below or
[14:03] slightly above the market multiple.
[14:05] Okay.
[14:06] So why
[14:07] why is that? Just because Nvidia is sort
[14:09] of the poster child for
[14:11] things gone
[14:12] crazy or the bubble that everybody
[14:14] fears? I think people It's funny, you
[14:16] know, the last earnings that they put
[14:18] out were phenomenal and yet the stock
[14:20] kind of went down on the back of it. So
[14:22] I think a lot of people were scratching
[14:24] their head like, how can a business grow
[14:25] that fast, be that profitable, be that
[14:28] dominant from a share perspective and
[14:29] yet not
[14:31] be rewarded?
[14:32] I mean, I think a lot of people point to
[14:34] just it's a big company, right? It's the
[14:36] largest market cap company in the world
[14:38] and it takes a lot to move that.
[14:41] Some people say, well, maybe all the
[14:43] good news is in the price, right? Like
[14:44] they'll they'll they'll that will be the
[14:46] line of logic, which is
[14:48] we've just seen every hyperscaler
[14:49] increase their capital expenditure
[14:51] budget for this year. Like is it really
[14:53] going to get better? You know, and
[14:55] they're obviously among the biggest
[14:56] customers of it.
[14:59] Again, all those are kind of, you know,
[15:01] maybe very short term or very sort of
[15:02] like kind of things that the market
[15:05] loves to focus on it, but I think that
[15:06] the long term, you know, I think we we
[15:08] think it's still undervalued and will
[15:10] grow fast. And there's no reason to
[15:12] think that the market, the multiple it
[15:13] gets to capitalize its business, you
[15:16] know, shouldn't expand just because of
[15:18] the quality of what they do. Right.
[15:20] Yeah, I think the big thing that
[15:21] investors are struggling with
[15:25] everything I think in tech and in
[15:27] semiconductors in particular is we see
[15:30] this unbelievable growth. How
[15:32] sustainable is it? And so is that demand
[15:35] real? So if you've got this incredible
[15:39] driver that has gotten companies like
[15:41] Nvidia or Micron or others to your level
[15:44] they're at today, question is how
[15:46] sustainable is it? So I think Colin said
[15:48] earlier, if you take this view that says
[15:50] we'll take what was the cloud era and
[15:53] we'll increase that by an order of
[15:54] magnitude in terms of everything that
[15:56] comes from it. All the demand drivers,
[15:59] all the supply you need to basically go
[16:00] make it happen. So we're still early on
[16:03] in that. That's going to be the debate
[16:05] that takes place here. I think investors
[16:07] will figure it out.
[16:08] And if we come to a conclusion that just
[16:10] simply says that this is going to last
[16:12] for a while, then maybe you don't see a
[16:14] lot of multiple expansion, but obviously
[16:16] you'll get paid for that tremendous
[16:19] growth that's happening. Those those 10
[16:21] companies I mentioned, the 10 most
[16:23] valuable in the S&P, they're expected to
[16:25] grow 33% this year.
[16:27] That's way faster than the economy, way
[16:29] faster than the other 490 companies in
[16:32] the S&P. And so the faster you grow, the
[16:35] more valuable you should be. And so I
[16:37] think that's how it will ultimately play
[16:38] out. But people are just worried about
[16:39] the cyclicality and is there, you know,
[16:42] this whole concern about the bubble. You
[16:44] know, I think AI to me it's
[16:45] extraordinary. I think it will last for
[16:47] quite some time. And, you know, the
[16:49] debate is there's massive opportunities
[16:52] for companies, but there's also going to
[16:54] be a lot of disruption. Right. And so
[16:56] that's a big debate that's being waged
[16:57] now about software, how it just
[16:59] completely changed the way software
[17:01] companies maybe get built and and run in
[17:04] the future.
[17:05] Well, it seems like So let's talk a
[17:07] little bit about the market's
[17:08] interpretation rather than just the
[17:10] companies. So seems that the market is
[17:13] very emotional. I mean, so
[17:15] I guess it was last week, maybe the week
[17:17] before that there was this scenario
[17:18] analysis that went viral about, you
[17:20] know, that that AI is going to wipe out
[17:25] white collar jobs and therefore that's
[17:27] going to impact the economy and you
[17:29] know, put together this whole kind of
[17:30] fictionalized account of what will
[17:32] happen.
[17:33] These headlines that people almost are
[17:35] just waiting for. Right? So one is why
[17:39] is the market so emotional about this?
[17:41] And at the end of the day, what you guys
[17:43] always say is the market's always right.
[17:46] Is it always right?
[17:49] Well,
[17:49] [sighs]
[17:50] so my foundational belief around markets
[17:53] is that the markets generally are very
[17:56] poor at predicting the future.
[17:59] I mean, I think if you can go back and
[18:00] look over time like what everyone
[18:02] everyone says they have a view, they
[18:03] have a they have a view. Could be
[18:05] optimistic, could be pessimistic, but
[18:06] generally speaking, their ability to
[18:08] predict, whether it's an economist,
[18:09] whether it's a strategist, whether it's
[18:10] an investor or portfolio manager,
[18:13] it's really hard to predict the future.
[18:15] And I think in particular what's really
[18:16] hard to predict are the surprises to the
[18:18] upside, the surprises to the downside.
[18:20] And so I think but that's what people
[18:22] focus on when these reports come out. So
[18:25] I I kind of joke with my team like
[18:27] people are investing by Substack
[18:29] articles today. People who are deep in
[18:31] it are educated, but there's not data
[18:33] yet that says that's going to be the
[18:35] outcome. And I think because these
[18:36] paradigm shifts are so there's such a
[18:39] tectonic event in the industry that
[18:41] people are kind of grasp at these
[18:44] outcomes because they sort of start to
[18:46] look at their portfolio and they they
[18:47] worry about the whether it's a black
[18:49] swan or sort of this long tail outcome
[18:52] that could be very, very disruptive. And
[18:54] but again, the data's not there to
[18:56] support it. Again, we'll only know in
[18:57] hindsight who was correct. The reality
[19:00] is there's only usually one, you know,
[19:02] Nassim Taleb or there's only one Michael
[19:04] Burry who predicted the credit crisis.
[19:06] Everybody else got it wrong. And so the
[19:08] vast majority of these things, I think
[19:09] people are a little frustrated. I'm
[19:11] frustrated because if you look at the
[19:13] what what these reports what they say,
[19:15] what they contextualize,
[19:17] it could be a scenario, but there's no
[19:18] probability that's being assigned to it.
[19:21] Yet it kind of flows through the market
[19:22] as if it's kind of gospel. So the the
[19:25] report was not only was it bad for
[19:27] software that day, but it was bad for
[19:29] financial services and payment companies
[19:31] like Visa. It was bad for insurance
[19:34] stocks. It was And again, like you're
[19:35] just sort of extrapolating and it feels
[19:37] more like these reports are sort of
[19:40] catalyzing
[19:43] people who have relatively short-term
[19:44] trading views in mind and are taking
[19:47] advantage of kind of just the
[19:49] disruption, which is a it's a difficult
[19:51] thing to see through, right? It's sort
[19:53] of like the analogy we made the fog of
[19:54] war. Like, you don't know if you're a
[19:56] portfolio manager how it's going to play
[19:58] out. And so, your defense is, well, I'm
[20:00] just going to be either less invested or
[20:02] less exposed is the only way I can
[20:04] control my outcome until things begin to
[20:06] come a little bit clearer. So, again,
[20:08] take enterprise software as like the
[20:10] case in point. Like, that marketplace
[20:12] has been, you know, kind of eviscerated
[20:15] over the last, you know, kind of 6
[20:17] months or so and particularly the last
[20:19] couple of months where whether it's
[20:20] Cloud Code or Citrine or others and
[20:23] people are just wondering what's the
[20:25] what's the end state for this this
[20:26] businesses. I would say the end state is
[20:28] clearly it's going to be less companies,
[20:31] but there will be some amazing winners
[20:32] in it, but the market's not willing yet
[20:34] to sort of underwrite the winners yet or
[20:36] that thesis because they just it's too
[20:38] hard to tell who's going to make it, you
[20:40] know, make it out until you see more
[20:42] data. And the data would be are these
[20:44] companies putting intelligence into
[20:45] their product? Is that intelligence
[20:47] actually, you know, providing real value
[20:49] to their customers?
[20:50] Mhm. Are you seeing that value then
[20:51] being manifested in better revenue
[20:53] opportunities and revenue growth at
[20:55] these businesses? You haven't seen that
[20:57] yet. And so, until you start to see
[20:58] that, then I think you can answer some
[21:00] of those questions.
[21:02] But in this environment we're living in,
[21:03] it's it's it's it's hard to get people
[21:05] to sort of make a stand or put a put a
[21:07] put a put their foot down so this is
[21:09] this is value, this is interesting, this
[21:10] is cheap.
[21:11] It's easy to sensationalize, I think.
[21:14] And so, if you're a pundit or you're
[21:16] doing research and you want to come out
[21:18] with a bold prediction, no one's going
[21:20] to know if you're wrong or right for a
[21:22] little while. I mean, you think about
[21:23] where we are.
[21:23] headlines. You think about where we are
[21:25] today. You know, ChatGPT was launched on
[21:27] November 30th of 2022. It's not that
[21:30] long ago. And think about what's
[21:31] happened since that. It's pretty
[21:33] amazing. And so, you now have a market
[21:35] that's trading at like 21, 22 times
[21:37] earnings. That historically is not
[21:39] particularly low.
[21:41] [snorts]
[21:41] And so, when you've got that and you got
[21:44] everybody who is trying to figure out
[21:45] what the future looks like, it's very
[21:47] easy to come out and have some negative
[21:49] commentary that causes people to to
[21:52] flinch. And that's driving volatility
[21:54] here at the moment. And again, some just
[21:56] amazing stats. If you go back to 2020,
[21:58] go back to the cloud comments that that
[22:01] Colin talked about, total capital
[22:03] spending for for the cloud CapEx was
[22:06] like $120 billion a year in 2020. This
[22:09] year it's going to be 800 billion.
[22:12] It's going to be over a billion or or
[22:13] trillion, I think, in 2027. That's just
[22:16] massive. And so, when it lays to that's
[22:18] obviously all AI-driven or broadly
[22:20] AI-driven. And yet, we don't know what
[22:22] we're going to use AI for for sure when
[22:25] we look out in the future because the
[22:26] rate of change here in terms of what's
[22:28] been happening
[22:30] is just extraordinary. As we've gone
[22:32] from generative to reasoning to agentic,
[22:36] uh it's just so fast. We've never seen
[22:38] anything quite like it in in the world.
[22:40] liken like the the the ability for AI to
[22:44] code, right? Maybe it was 12, 18 months
[22:47] ago it was coding like a sixth grader,
[22:49] right? And then like this now it's
[22:50] coding like a PhD student, right? Like,
[22:52] it's happened so fast. Like, the
[22:54] improvements in these models have gotten
[22:56] so good and it's so fast and so quickly
[22:59] that now you're, you know, people are
[23:00] extrapolating that to other segments,
[23:03] you know, in whether it's your
[23:04] infrastructure, whether it's in your,
[23:07] you know, workflows. Like, you can now
[23:09] start to extrapolate, well, if it's
[23:10] doing that well on this, certainly it
[23:12] can do better across these. And so, you
[23:15] can see how disruptive it can be. So,
[23:17] some legitimacy to every headline,
[23:19] right? I mean, but
[23:20] Yeah. Yeah. until you dig deeper and
[23:22] Yeah. But what's kind of amazing, if you
[23:23] look at the broad market, it's been just
[23:25] kind of going sideways. Yes. So, cuz
[23:28] we've seen a lot of rotation happening
[23:30] within the market. And I think that's
[23:32] probably pretty reasonable to think.
[23:34] Right. Uh if you go back to where
[23:35] liberation day was, where the market
[23:37] came under pretty significant pressure,
[23:40] you know, our view at Morgan Stanley is
[23:41] that was probably the start of a new
[23:44] bull market. And bull markets don't just
[23:46] last for 9 or 12 months. So, um you're
[23:49] going to get some volatility, you're
[23:51] going to get some churn that happens,
[23:53] but the economic environment is just
[23:56] sort of okay. And so, as long as that
[23:58] holds together, you're going to see
[24:00] broadening out into things beyond just
[24:04] this whole AI trade that'll be quite
[24:06] helpful to the broad market.
[24:08] This episode is brought to you by Morgan
[24:10] Stanley.
[24:11] Morgan Stanley celebrated its 90th
[24:13] anniversary last year and its tech
[24:15] investment banking team has prided
[24:17] itself in working with the most
[24:19] innovative entrepreneurs and leadership
[24:21] companies for decades.
[24:23] During that time, they've led IPOs for
[24:26] hundreds of tech companies including
[24:28] Apple, Arm, Astera Labs, Broadcom,
[24:32] Cisco, Facebook, Google, Nvidia,
[24:35] Salesforce, and Uber. In addition, they
[24:38] worked as a trusted advisor and executed
[24:40] thousands of M&A transactions for some
[24:43] of the most industry-defining deals in
[24:45] technology.
[24:46] Throughout the years, Morgan Stanley's
[24:48] tech investment banking team's success
[24:51] has been achieved by dedication,
[24:53] consistency, and a North Star purpose of
[24:56] helping its clients to maximize their
[24:58] opportunities and overcome their
[25:00] challenges. Now, let's get back to the
[25:02] episode. When we look back 1 year, what
[25:05] headline do you think is going to be the
[25:07] most significant? Like, the the Deep
[25:09] Seek headline was in January, um
[25:12] last year.
[25:21] Will still be having an impact?
[25:23] Will it be Will it be this cloud moment
[25:25] or
[25:25] it's this I think it's this code
[25:26] generation is just very different. And
[25:29] the question is, how do we leverage
[25:30] that?
[25:31] Um it's hard for me to believe, for
[25:32] example, I'm it's not my expertise. I I
[25:34] I specialize in semiconductors, but it's
[25:36] hard for me to believe that every
[25:38] software company is going to not have a
[25:39] business in the future.
[25:41] So, I think they'll be able to leverage
[25:43] AI tools and capabilities to make their
[25:46] businesses that much stronger. But, you
[25:49] know, we will see. Like, here's a, you
[25:51] know, one of my favorite things that I
[25:53] learned in, you know, the last 37 years
[25:56] of covering this was from Andy Grove.
[25:58] Um you talked about paradigm shifts. He
[26:00] talked about this is basically in the
[26:02] early 1990s about how paradigm shifts
[26:05] can basically transform companies or
[26:08] sectors at a strategic inflection point.
[26:11] And that's where you have kind of legacy
[26:14] suppliers that don't kind of get the
[26:18] transition at that strategic inflection
[26:20] point, at that paradigm shift, and their
[26:23] businesses head south.
[26:25] And then other companies have, you know,
[26:28] a a much, you know, brighter future. And
[26:31] so, this is like, you know, examples of
[26:33] going from CPUs to GPUs as an example or
[26:37] from cell phones to smart phones.
[26:39] There's been many examples of this and,
[26:43] you know, who knows if the cloud moment
[26:44] is going to be part of that for
[26:45] software. So, it's hard to know, but
[26:47] really important to watch. So, that's
[26:49] the lens I use when I try to figure out
[26:51] who are the winners and it's hard, as
[26:52] you know, in semiconductors to figure
[26:54] out who's going to be a really
[26:56] successful company. Who's going to be
[26:57] the next Astera Labs? Who's going to be
[26:59] the next, you know,
[27:01] Marvell, Broadcom, whatever.
[27:03] Um and I do it through that lens of who
[27:05] has,
[27:06] you know, innovation, durable businesses
[27:09] that are going to be able to take
[27:10] advantage of these paradigm shifts and
[27:13] strategic inflection points.
[27:16] So, when you guys when you
[27:18] are advising or talking to boards or
[27:21] CEOs,
[27:23] how are they looking at AI? Are they
[27:24] looking at it as
[27:27] a productivity tool? Are they looking at
[27:29] it maybe to transform actual head count
[27:32] or margin expansion? Are they just
[27:34] trying to do whatever they can because
[27:36] they need to compete?
[27:37] I mean, how are they viewing AI within
[27:39] their own company?
[27:41] Well, I think within semiconductors, um
[27:44] it it's pretty clear that they're going
[27:46] to drive it. So, it's the foundation.
[27:48] It's a lot of the technology behind how
[27:50] you're going to basically support uh the
[27:53] compute need and feed it with memory and
[27:56] there's obviously, you know,
[27:58] companies doing things where you put uh
[28:01] you know, compute in memory or very
[28:03] close to memory. So, it's really
[28:05] important for those companies. For the
[28:06] broader world, it needs a little bit
[28:08] different.
[28:09] about what about how they're utilizing
[28:10] it? How are um whether it's a chip
[28:12] company or others actually within
[28:14] whether it's for the chip design or the
[28:17] workflow? There's a so, Datadog's a
[28:19] company that we we we spent a lot of
[28:20] time with. Datadog talks about on their
[28:22] earnings transcript. They have they talk
[28:24] about AI in two ways. They have Datadog
[28:25] for AI, which is what Datadog is doing
[28:28] by putting intelligence into the
[28:29] products that they sell to their
[28:30] enterprises and their customers. Then
[28:32] they have AI for Datadog, which is how
[28:35] is Datadog actually consuming AI both
[28:38] from improving workflows, improving
[28:40] productivity, enabling them to do more.
[28:43] And so, it's a combination of those two.
[28:45] Most companies, and this is Datadog's a
[28:46] software business in the data
[28:47] infrastructure side. And they're they're
[28:50] looking at it I want to be able to serve
[28:52] my companies with the best AI tools. And
[28:54] so, then I also want to be able to make
[28:56] sure that I'm leveraging those tools to
[28:58] make my business more important. The
[29:00] question around like head count is a
[29:01] really, you know, you know, tough one.
[29:03] It's an it's an interesting discussion,
[29:05] right? On one hand, you have Jack Dorsey
[29:06] who just
[29:07] decided to let go of about 40 plus
[29:09] percent of his workforce.
[29:11] Um and a lot of it saying we can do we
[29:13] can do basically same with less
[29:15] effectively or maybe a little bit more
[29:17] with less. Um but I think a lot of
[29:19] people are thinking about it as kind of
[29:21] in that like I I kind of simplify to I
[29:24] can do the same with less because AI
[29:26] makes me much more productive.
[29:28] I can do more with a little bit less or
[29:31] a little bit more with a little bit less
[29:32] or I can do more with same or even more
[29:35] with a little bit more.
[29:36] And like I think those those are the
[29:38] four different outcomes that companies
[29:40] are thinking about. And I think many
[29:41] companies are not as dramatic as Jack
[29:44] Dorsey. He's in kind of the same with
[29:46] less. And then on the other end of the
[29:48] spectrum, if you just look at like the
[29:51] wrecks for hiring across the software
[29:53] industry or across even the AI
[29:55] companies, the large language models,
[29:57] they're all still hiring very
[29:58] aggressively. So, you know, or they're
[30:00] looking for talent. Even in an Amazon,
[30:02] which goes through regular layoffs,
[30:04] still has tons of e-wrecks out for new
[30:06] people. And they I think they even said
[30:08] one of their their reductions in force
[30:11] that we expect many of the people who
[30:13] are leaving these jobs to be hired in a
[30:15] different job within Amazon. And so, I
[30:17] think
[30:18] that's kind of the question they're
[30:19] trying to answer is like cuz I think
[30:20] most companies if you're if you're built
[30:22] for profit, you'd rather do there's a
[30:25] one-time change, which if you do same
[30:27] with less, which is I can be a lot more
[30:28] profitable once that's done. Like again,
[30:30] what drives business ultimately, right,
[30:32] is, you know, growth and free cash flow
[30:34] or growth and profitability. That only
[30:35] affects one of those axes, which is
[30:37] profitability. If you can do more
[30:39] revenue
[30:40] with better profits, that's ultimately
[30:42] the ideal. I don't know if they've
[30:44] titrated yet the right formula of
[30:48] what level of head count, what
[30:49] investments in head count do I make,
[30:50] what investments in AI do I make. But I
[30:52] think for most people, it's going to be
[30:54] more that way, which is can I can I grow
[30:56] my business better with AI, can I be
[30:59] more productive, can I be more efficient
[31:01] as opposed to just wholesale like I'm
[31:03] taking heads out.
[31:05] Cuz I've heard I mean, actually I heard
[31:06] it from an investment banking firm that
[31:08] said this will be as large as we ever
[31:09] are.
[31:10] Yeah, it's it's quite possible. I mean,
[31:12] but it may be that within that
[31:14] population it may shift around.
[31:15] Okay. Um and some investment banks are
[31:20] built in with the biggest banks in the
[31:21] world, they're probably some of the
[31:22] largest law firms in the world. Right,
[31:24] if you imagine the number of lawyers
[31:26] that sit inside of JP Morgan, it
[31:28] probably rivals a top Wall Street law
[31:30] firm. Okay.
[31:32] The question they would be asking
[31:33] themselves is do you actually need that
[31:35] number of people now with AI? Can I get
[31:38] more of that analysis done, you know,
[31:40] because, you know, AI is, you know,
[31:43] there's most of most of law is now sort
[31:45] of digitized and sort of analyzable and
[31:47] run through models. It's text-based. I
[31:49] can get all the data I need. Do I need
[31:51] as many people doing that? That's a good
[31:52] question to ask. But do they take that
[31:55] and shrink the head count or do they
[31:56] redeploy that into what they can grow.
[31:59] And so, I I still think you're going to
[32:00] have net growth for some of these
[32:03] companies for a while, but it's going to
[32:04] be smaller at the margin than it than
[32:06] you would have originally planned.
[32:08] So, if you did have a slowdown on the
[32:11] enterprise adoption of AI because maybe
[32:13] someone isn't seeing that return on
[32:15] investment.
[32:16] Mhm. Um
[32:18] that could be very very disruptive. I
[32:20] mean, do you
[32:22] are you worried about that? Um how
[32:24] quickly do people have to see that ROI
[32:27] in order for that not to be an impact?
[32:28] Yeah, I mean, that's that's definitely
[32:30] Look again, the when we started, I think
[32:32] in my opening comment I said the the big
[32:34] question is on the return, right, for
[32:36] the spend that you're doing. It kind of
[32:37] met to the enterprise, it's the same
[32:39] thing. You know, if you're taking part
[32:41] of your budget, you know, that may have
[32:43] gone somewhere else and now you're
[32:44] putting it into AI infrastructure and AI
[32:46] tools and you're not getting the
[32:48] benefit,
[32:49] um then I think for sure that's a
[32:51] question mark, you'll slow it down. I
[32:53] think the I think it's less about us
[32:55] getting a return if I think about us as
[32:57] a customer Morgan Stanley, it's more
[32:59] about can we get the tools that actually
[33:02] work within our regulatory, our
[33:04] compliance, and our security
[33:05] architecture that we need
[33:07] [clears throat] as opposed to getting
[33:08] returns. I think the returns within our
[33:10] organization, you can just think of so
[33:12] many ways to find the returns.
[33:14] Um you know, one would one simple one
[33:17] would be expense management. How many
[33:19] expenses are reviewed across a bank like
[33:21] Morgan Stanley or across other
[33:23] enterprises
[33:24] are actually reviewed today by a human?
[33:27] Probably more than I would say the
[33:28] answer is I don't know the number, but
[33:30] it's probably more than it needs to be.
[33:31] Okay. Um that's a pretty easy thing to
[33:34] kind of figure out how to kind of put
[33:36] rules in place that can solve a lot of
[33:38] the basic questions when you're doing
[33:39] those reports and figuring out expenses.
[33:41] So, I think there's going to be ways to
[33:42] find it. I think the bigger challenge is
[33:44] making sure it's fits within the overall
[33:46] architecture. [laughter]
[33:47] You know, as we as an
[33:49] regulated entity, think about health
[33:51] care the same or others, like
[33:53] the you can't just put something in the
[33:54] wild and let it go. You actually have to
[33:56] be kind of has to be sort of set up and
[33:59] managed in a way that actually is
[34:00] compliant, and that takes some time. So,
[34:02] I think I'm less worried about the
[34:04] returns in the immediate near term, but
[34:06] just worried about the obstacles of
[34:08] making sure that what we put into the
[34:09] plant is actually
[34:11] the right the right sort of software
[34:13] tools that that based on sort of what
[34:15] our needs and needs and sort of wants
[34:18] are.
[34:19] Is there a I mean, is there some point
[34:23] that is the most fragile in AI that
[34:27] could cause a big disruption? Again, I
[34:29] said asked about enterprise adoption, if
[34:31] that slowed down, that would certainly
[34:32] be a
[34:33] disruption. What what else do you worry
[34:35] about that could disrupt this? I mean, I
[34:37] think that's probably the big thing is
[34:38] are people getting the returns that they
[34:40] expect. And I think we're still early on
[34:43] in this to be able to determine where
[34:45] that's going to be. For companies that
[34:47] don't make the investments, they're
[34:48] probably going to be left behind. So,
[34:49] they do need to make, you know, some,
[34:51] you know, got to be some leap of faith,
[34:53] if you will, that there's going to be
[34:55] productivity that happens. And like
[34:57] that's the data that we're seeing today.
[34:59] I mean, we see it for sure inside what
[35:01] we do. We're way more productive with AI
[35:03] tools than without them. And so, I think
[35:05] that's where where things will basically
[35:07] head. I think it'll be be a continuum
[35:09] of, you know, some companies that just
[35:11] will take the approach that just says
[35:13] I'm going to do, you know, the same with
[35:16] a lot less people, but there's going to
[35:17] be others that will basically plow it in
[35:19] and find ways to to get, you know, more
[35:21] things that will basically build and
[35:23] deliver over time. Because I think that
[35:25] productivity argument that, um you know,
[35:27] for example, um oh, we can get rid of,
[35:30] you know, 200 people um that that
[35:33] typically answer a hotline for customer
[35:35] service that you turn over to AI.
[35:38] Okay.
[35:39] If you got rid of those people, yes,
[35:41] that's a that's a savings, but a lot of
[35:43] these productivity gains are just maybe
[35:47] making customer service better, but
[35:49] they're not really I mean, you don't
[35:51] really see it at the bottom bottom line.
[35:53] I mean, there isn't that return yet. So,
[35:55] I think that the the people,
[35:57] you know, certainly
[35:58] on their minds. I mean, I think the the
[36:00] the things that give me a little bit
[36:01] more confidence, if you look in like
[36:04] businesses like Meta and Alphabet, I
[36:06] think are good examples of businesses
[36:08] where maybe
[36:10] it's hard to disaggregate, but if you
[36:12] look at their business from a growth
[36:13] perspective and kind of profitability
[36:15] perspective over the last couple of
[36:16] years as they've in a way been using AI,
[36:20] you know, into their core businesses,
[36:21] whether it's the algorithm that serve
[36:23] ads and other things like that, the
[36:25] businesses have actually I mean, if you
[36:26] look at their businesses, it's
[36:27] accelerated. And so, now again, you
[36:30] can't like there's no line item that
[36:32] says this is the challenge, right?
[36:33] There's no line item that says AI
[36:35] revenue, you know, to show that, but if
[36:38] you just look at how fast and how
[36:40] dynamic those businesses have been over
[36:41] the last couple of years as they've this
[36:43] scale to be accelerating at scale is
[36:46] is really impressive. And so, that's at
[36:48] least one area. So, it's it's helped
[36:50] their business, for sure. Is it driving
[36:52] more margin? Yes, it is. Are they
[36:54] spending more because of it? Absolutely.
[36:57] And so, I think as you get more examples
[36:58] like that, and then maybe that's not as
[37:00] clear-cut as, you know,
[37:03] a simple sort of like line item that may
[37:05] someone may have like like if you're
[37:06] Salesforce or you're ServiceNow, here's
[37:08] my AI revenue skew and what it's
[37:10] growing, but I think it's very powerful
[37:13] and certainly been a driver to their
[37:14] stock price performance. And so, if more
[37:16] of that starts to manifest itself, I
[37:18] think you start to answer that question.
[37:20] But it could be, you know, it could take
[37:22] another 12 to 18 months to show that.
[37:24] But I those are good those are great
[37:26] examples of companies that
[37:28] I think most people acknowledge that
[37:30] that
[37:31] AI has really helped their business.
[37:32] Right, but that's all happening at a
[37:33] time when global GDP is below trend
[37:36] line. So, we've got a 120 trillion
[37:39] dollar global economy, expectations like
[37:41] low 3% type of growth here. That's below
[37:45] historical trends. So, you can imagine
[37:46] if we get to a point where we're growing
[37:48] at trend line or better, what does that
[37:51] sort of mean for all the
[37:53] you know, the growth that will happen
[37:54] for these companies by themselves?
[37:57] Okay, so when we a lot of times when
[37:59] people talk about AI, it's just it's one
[38:01] thing, right? And it's or it's one
[38:03] company that people have in their mind
[38:04] that
[38:05] whether it's OpenAI or whomever, but
[38:08] but of course, AI is an entire
[38:10] ecosystem, right? So, help the listeners
[38:13] to understand sort of every layer in the
[38:16] the stack, like who plays in the top
[38:19] layer, who plays in between, all the way
[38:21] down to applications. And so, kind of
[38:24] help us
[38:25] draw that mentally. And then maybe where
[38:28] the value is being accrued today.
[38:32] Well, Like I I can start with the
[38:34] foundation. Yes. Which is obviously
[38:36] going to come from semiconductors and
[38:38] platforms that people build around
[38:39] those. So, you ultimately have to do
[38:41] compute, got to have memory cuz you need
[38:44] to basically be able to feed. You need
[38:45] to have data that's going to go feed and
[38:47] get processed. You need to be able to
[38:49] have interconnections and connectivity
[38:51] within that. And that's for, you know,
[38:53] for
[38:54] data centers are going to be driving
[38:57] what's called scale out technologies,
[38:59] which today is all largely copper based,
[39:01] some optical. Um in the future, we'll
[39:03] see more scale up. So, you'll see more
[39:05] of a transition to optical as a result
[39:08] of that just cuz we'll run out of the
[39:10] physical limits basically of what you
[39:11] can do with with copper,
[39:14] uh you know, wiring.
[39:15] Um so, there's a lot that happens.
[39:17] to just be in the data center in order
[39:19] to be to benefit from AI for a chip
[39:22] company.
[39:23] Oh, yeah, exactly. Yeah, for sure. And
[39:25] AI is going to, you know, we see what's
[39:26] happens with Waymo as an example. Or
[39:29] we're just very early days in robotics,
[39:32] like super super early. And that's going
[39:34] to have, you know, all kinds of
[39:36] semiconductors that go into that, right?
[39:38] Sensors, motors, all those kinds of
[39:41] things. It'll be pretty significant. Um
[39:44] I suspect bigger than the automotive
[39:45] industry as an example. It'll take time
[39:47] to get there, but that's a whole 'nother
[39:50] trend that's going to happen as well.
[39:51] So, you've got semiconductors I think
[39:52] will be the bedrock foundation for a lot
[39:55] of stuff that happens. And then from
[39:56] there, we'll move up the layers up the
[39:59] stack, if you will, into software and
[40:00] applications and things like that
[40:02] that'll basically, you know, have all
[40:05] the all the value drivers.
[40:06] are the hyperscalers there?
[40:08] In that chart? Well, they're they're
[40:10] they're providing all the services
[40:11] behind it. And, you know, that's they're
[40:13] also I think trying to capture some of
[40:15] the foundational technologies, which is
[40:17] why they're all effectively trying to do
[40:19] their own semiconductor development, but
[40:23] it's hard. Mhm. Um so, you know, to me,
[40:26] if you can if you can design
[40:27] semiconductors and you can use that
[40:29] across lots of companies, like what a
[40:32] Broadcom does, I think you just have a
[40:34] better ability to basically leverage
[40:36] your R&D, your capabilities. Being
[40:38] vertical can be very good. You know what
[40:40] the workloads are, but it's just it is a
[40:42] bit harder.
[40:42] Right. But they're you know, they're
[40:43] providing services and and the
[40:45] platforms, if you will, to then enable
[40:48] the rest of what's happening in the
[40:50] other the other layers.
[40:51] Yeah, the way sort of I some sometimes
[40:53] joke about it is like I it's called
[40:54] follow the gross margin. Okay. You know,
[40:56] in the old day in the cloud world, the
[40:58] gross margins many of the gross margins,
[40:59] the best gross margins were actually in
[41:01] the application layer.
[41:02] Right? So, the biggest software
[41:03] companies were earning 80 85% gross
[41:06] margins in that business. And, you know,
[41:08] as you move to AI, actually the best
[41:10] some of the best gross margins are
[41:11] actually at the, you know, chip provider
[41:13] level of that technology. So, it's the
[41:15] it's the Nvidias, it's the AMDs of the
[41:17] world, etc., that have some of the best
[41:19] margin. And then it kind of goes down
[41:20] from there. And so, you know, you've got
[41:23] if you think about it, you've got the
[41:25] software application that's coding on
[41:26] top of one of the big LLMs. They're
[41:28] actually running their compute on a
[41:31] hyperscaler or a data center that's and
[41:33] then that is buying chips and
[41:35] infrastructure and year from that whole
[41:37] layer of what Mark has covered for the
[41:39] last 30 uh 39 years. That's where a lot
[41:42] of the value has accrued today.
[41:44] You're starting to see a change it'll
[41:45] change a little bit for sure because,
[41:47] you know, as you get to scale, you know,
[41:49] the layer in the middle of the LLM layer
[41:51] and the application layer will start to
[41:53] sort of see improvements in the gross
[41:55] margin, but it's been a tough business
[41:57] for them because they've had to pay out
[41:59] so much along the way.
[42:01] Um but I think it's starting to change
[42:02] because you're now seeing I mean, coding
[42:05] as an example, right? Coding that
[42:07] business went from probably de minimus
[42:09] revenue 2 and 1/2 years ago to now
[42:12] multi-billion dollars sort of ARR
[42:14] business. So, it's it's gone from
[42:15] essentially zero to this to a very large
[42:17] number. As you get to those large
[42:19] numbers, it starts to become very
[42:20] profitable with very good gross margins
[42:22] on top of it. And you've gotten to scale
[42:23] with it. And so, that'll start to happen
[42:25] and will roll through all all all of the
[42:28] sort of
[42:29] LLM and the applications that are
[42:31] written on top of it as well. And so,
[42:33] again, what you're seeing already I
[42:35] think in the VC community and the early
[42:37] stage startup community, like the amount
[42:39] of dollars it takes and the amount of
[42:40] people it takes to bring a bring a
[42:42] product to scale vastly different than
[42:44] it was 10 years ago, 15 20 years ago in
[42:47] the cloud environment. You can do it
[42:49] with probably
[42:50] 25% of the people and probably a lot
[42:52] less with a you know, a little bit less
[42:54] capital. So, then these businesses can
[42:55] scale. Right? And then the capital's
[42:57] going to compute, it's going to, you
[42:59] know, to consumption. And you you're
[43:00] seeing these businesses again, like the
[43:03] number of new company creations and the
[43:05] the number of like new apps that are
[43:07] being put in the app store, like I think
[43:09] I don't know I won't I'm probably not
[43:11] they're not all-time highs, but they're
[43:12] at near highs, like after a big lows.
[43:14] And so, you're starting to see this
[43:16] already, this environment start to be
[43:17] created where, you know, there'll be a
[43:19] little bit more diversification of how
[43:21] the value and the gross margins get
[43:23] split up across the chain. Okay.
[43:26] So, Mark, we were talking earlier about
[43:28] that cyclicality of the semiconductor
[43:31] industry. And when when we did start
[43:33] GSA, one of the thoughts was that if you
[43:36] have a fabless foundry business model,
[43:38] you that the cycles are going to be
[43:40] tempered because you have fewer people
[43:42] making decisions. We said that earlier.
[43:44] Yeah.
[43:45] So, does that really mean though that
[43:48] now the semiconductor industry won't be
[43:50] cyclical? I mean, you look at memory as
[43:52] an example and it's I mean, the most
[43:54] cyclical part of the the industry.
[43:56] But, you know, right now memory is um
[44:00] you know, the demand is huge and the
[44:02] capacity is small. So, is that just an
[44:05] issue of well, as soon as capacity ramps
[44:08] up, it'll change once again or they'll
[44:11] over people will overbuild on the memory
[44:12] side?
[44:14] Yeah, unfortunately, I don't think we
[44:15] can ever eliminate cyclicality in this
[44:17] industry. It's just it's just not
[44:19] possible. But we can make it better and
[44:22] I think we can change the amplitude of
[44:24] some of it over time. Memory's
[44:25] particularly difficult because you've
[44:27] got three big players by and large in
[44:29] DRAM. You've got more in uh in NAND. And
[44:33] so, it is hard and it's difficult. You
[44:35] can't just bring on
[44:37] capacity in small increments. It's
[44:39] pretty chunky. And typically, cuz it's
[44:42] just human nature, you're not going to
[44:44] build plants until you've got lots of
[44:46] demand. And ultimately, you're just
[44:47] going to probably get to a point where
[44:49] supply is going to over, you know,
[44:51] overshoot the demand. And if that
[44:53] happens during an economic slowdown, it
[44:56] just amplifies the problem. So, I think
[44:58] there'll be sectors that are always
[44:59] going to be quite cyclical. That's why
[45:01] unfortunately, I look at Micron or SK
[45:03] Hynix or Samsung and what they do, and
[45:05] it's extraordinary. It's so hard what
[45:08] they do.
[45:09] hard. And yet Micron
[45:10] trades at 10 times earnings. That is
[45:12] less than half of what the S&P 500
[45:14] trades at. So, it's crazy to me. I think
[45:17] ultimately that will will change, I
[45:19] believe, cuz um it will it's just
[45:21] there's not going to be another DRAM
[45:22] company, I don't think,
[45:24] um that can do what they do. So,
[45:26] ultimately I think they'll get more
[45:27] pricing power and it'll be less, you
[45:30] know, less um
[45:31] problematic, but it's always going to be
[45:33] somewhat cyclical because of the factor
[45:35] I gave you. Now, the rest of the
[45:36] industry, when you look at the fabless
[45:38] companies, I do think they manage
[45:40] inventories well.
[45:42] And they've typically done a good job.
[45:44] COVID was just uh I think a very bad
[45:46] example because it was just such a
[45:49] massive hit to the supply chain. And
[45:52] everybody began to understand the
[45:54] importance of semiconductors when the
[45:55] golden screw wasn't available. So, you
[45:57] couldn't buy that car, you couldn't get
[45:59] that washing machine. And so, people
[46:01] just said, well, great, let me just get
[46:02] all I can. I'll I'll basically over
[46:04] inventory. And we all know how that
[46:07] movie ends. I mean, we all have been
[46:08] doing this for a long time. We just know
[46:10] there's double triple ordering. And then
[46:12] ultimately, um when things slow down,
[46:15] you just
[46:17] cancel things. Right. Um so, I think
[46:19] that's that's an unusual time. I don't
[46:22] know that we'll see something like that
[46:23] in the future. So, for the rest of, you
[46:26] know, the sector, I do think it will be
[46:27] less cyclical, but we won't ever
[46:29] eliminate the cyclicality entirely. The
[46:33] good news though is that for these
[46:34] fabless companies, they get variable
[46:36] cost business models.
[46:37] Right. So, you don't have all the under
[46:39] absorption that you do if you've got a
[46:41] manufacturing facility. So, the Nvidias
[46:44] and Broadcoms and Marvells and Mediateks
[46:47] of the world, they're just going to be
[46:48] it's going to be different for them in
[46:50] this next when we get to the next
[46:52] ultimate downturn. Um cuz their gross
[46:54] margins will hold up. And the challenge
[46:56] is just how much of revenues fall
[46:58] relative to your opex you need to
[47:00] basically try to cover. So, um
[47:02] some cyclicality, but not terrible. And
[47:05] I think largely it'll be kind of tied to
[47:06] the economic environment.
[47:08] I mean, do you think that it's possible
[47:10] that it that a different model
[47:14] emerges from from AI? Because the demand
[47:17] is going to be so huge, can it really
[47:20] just be TSMC and a a few others
[47:24] that are supplying this now $1 trillion
[47:28] industry? I mean, is that possible? Or
[47:29] do or you think that fabless companies
[47:32] will begin to make investments in
[47:35] capacity differently than they do today,
[47:38] which would affect the the margins?
[47:40] Well, TSMC is an amazing company and
[47:42] quite frankly, I don't know that our
[47:44] industry would be where it is without
[47:47] TSMC. So, they've been incredible.
[47:51] Uh they can drive technology
[47:53] extraordinarily well. You know, you hear
[47:56] all these companies talk about first
[47:57] time right silicon. I mean, so the whole
[47:59] industry is just so much better managed
[48:01] today and TSMC's been a big part of it.
[48:03] But we do need, I think, to have broader
[48:07] manufacturing. I'd be very surprised
[48:10] outside of maybe there be a couple of
[48:11] examples. I don't think that most semi
[48:13] companies that today are just design
[48:16] firms will go and and basically build
[48:18] capacity themselves.
[48:20] But you've got GlobalFoundries, UMC,
[48:24] Intel. So, you've got other centers of
[48:27] excellence that can basically build over
[48:30] time. And then there's other parts of
[48:31] the industry like, you know, within
[48:33] analog, it's just a very different model
[48:34] for TI or analog devices and whatnot.
[48:37] Just the capital intensity is just very
[48:39] different. But for big digital,
[48:41] um it it it's just it's very hard. Um
[48:44] but I do think we'll have other people
[48:46] than just TSMC. We kind of have to.
[48:48] Well, it's definitely when you you know,
[48:49] you talk about the kind of the supply
[48:51] demand. I mean, there are definitely
[48:52] real choke points in this AI story that
[48:54] make it a little bit different than
[48:55] other kind of bubbles or other things
[48:57] that may you might think would be a
[48:58] bubble.
[48:59] Um
[49:00] TSMC is a choke point, right? The you
[49:02] know, all the silicon has to go through
[49:05] TSMC mostly for what's being done at,
[49:08] you know, whether it's AMD or
[49:10] uh it's it's Nvidia or others. And then
[49:13] you have, you know, memory as a choke
[49:14] point and that may get solved. You have
[49:16] power. The availability of power in the
[49:18] US is an issue. Um it's not easy to do
[49:21] it. It's not easy to bring online. There
[49:22] are, you know, real political and social
[49:25] costs of of power. Um you have the lack
[49:28] of
[49:29] infrastructure from a workforce
[49:30] perspective to be able to put these
[49:31] together. Not easy to go find much
[49:34] people to put these data centers
[49:35] together, right? You're seeing slowdowns
[49:37] in many of these companies in terms of
[49:38] their build. And so there's some of the
[49:40] governors which kind of mean that the
[49:42] you know, the compute just the supply
[49:44] demand for compute will probably still
[49:46] stay kind of
[49:48] under supplied Okay. and over demand
[49:51] um for a while because these choke
[49:52] points aren't easy choke points
[49:54] Right. to solve. Like it's not as if
[49:56] Hynix or Samsung or Micron can snap
[49:58] their finger and put a lot more capacity
[50:00] in. Same for the for the fab uh for
[50:03] TSMC. Same to find workers and train
[50:05] them up. Same to be able to get power
[50:06] through it. So it's there are some
[50:08] natural things that probably put some
[50:10] governors on
[50:11] the pace at which the supply demand
[50:15] balance can shift dramatically. And I
[50:17] think that's one of the reasons you feel
[50:19] more optimistic about there not being
[50:22] you know, sharp you know,
[50:23] cycles or things can you can you're
[50:25] looking around the corner for something
[50:26] that could go wrong. It doesn't feel
[50:28] like those are going [clears throat] to
[50:29] be those are going to be helped those
[50:30] are going to govern like the pace of Cuz
[50:32] we're because they're constraining
[50:34] They're constraining the
[50:35] the acceleration, yes. I mean every
[50:37] company we talk to when you ask them
[50:38] about their demand for compute will tell
[50:41] you they can't get enough of it.
[50:42] Right.
[50:42] Right? I mean you watch what's happened
[50:44] to token tokens have gone kind of
[50:46] vertical. And you know, you look at what
[50:48] they where they were 12 months ago you
[50:49] look where they were two years ago.
[50:50] People just can't get enough of it and
[50:52] so there's just not enough
[50:54] infrastructure to put into the ground
[50:55] fast enough to solve some of those
[50:57] issues in the next couple of years.
[51:00] So one of the things that's different um
[51:02] today than
[51:04] you know, for sure five or 10 years ago
[51:07] is that
[51:08] the government governments are very
[51:10] involved in semiconductors. Okay? And so
[51:13] there's
[51:15] semiconductors are now what everybody
[51:17] emphatically cares about, right? Um if
[51:19] you're if you're a government US
[51:20] government specifically and
[51:24] so it's it's changing I I think changing
[51:26] the entire industry. It's certainly
[51:28] changing the personality of the
[51:29] industry. And I think that there's a lot
[51:32] of risk to that um including that
[51:36] companies are going to be driven by
[51:38] something other than market forces in
[51:40] the decisions they make.
[51:42] So how much do you guys worry about
[51:43] that? What do you tell investors that I
[51:45] mean is that a downside or is that good
[51:48] for the industry? Is it is it a good
[51:50] market driver because you know that um
[51:52] semiconductors are important?
[51:55] Uh
[51:56] I mean I think we're
[51:58] I think the the focus again I I would
[52:01] maybe on the US itself. Um I think the
[52:03] focus on the US is
[52:05] again they want to
[52:07] reshore critical manufacturing to the
[52:09] US. You know, or at least a portion of
[52:11] that manufacturing to the US. Um
[52:14] I don't think it's um and that's going
[52:16] to be in areas like semiconductors. It's
[52:18] going to be in areas of AI. It's going
[52:20] to be in areas of you know,
[52:23] mission critical whether it's rare earth
[52:25] minerals, whether it's
[52:27] um parts and supply for
[52:29] you know, the defense infrastructure
[52:31] etc. They all want that to be built
[52:32] here. Um
[52:34] you know, I don't think that's not a bad
[52:36] thing to reshore some of the critical
[52:38] critical parts of that manufacturing and
[52:39] the high value of that manufacturing. I
[52:41] think it's actually very positive for
[52:43] the economy. I think it's actually when
[52:44] you think about the job
[52:46] the job what we just talked about from
[52:48] earlier about the job loss potential.
[52:50] There's actually a ton of jobs that will
[52:52] be created. Um
[52:54] hopefully it's efficient. Hopefully the
[52:55] investments make sense from an economic
[52:57] perspective. Certainly this
[52:58] administration feels
[53:01] that there's a there's a real desire to
[53:03] make sure that it gets done and gets
[53:04] done well with kind of buy-in from
[53:05] industry and it gets done on economic
[53:07] terms. Um but it is a risk if it
[53:11] you know, I think it's a risk for us
[53:12] geopolitically if it doesn't actually
[53:14] happen. Um
[53:16] so I think that's I think it is an
[53:17] imperative and I think there's a lot of
[53:18] momentum behind it. You you see whether
[53:21] it's I think they all have different
[53:22] versions of an American resilience,
[53:24] American dynamism. There's all this talk
[53:26] about that. And I think that's an
[53:27] important part of the next you know,
[53:30] decade for the US as it builds back some
[53:32] of that critical infrastructure.
[53:34] And certainly AI and automation is a
[53:36] huge part of that.
[53:37] Right.
[53:38] It just seems that one of the things
[53:40] that has that the semiconductor industry
[53:42] has been so successful is in efficiency.
[53:45] Okay? I mean it's just it has to be such
[53:48] an efficient industry. And so I just
[53:49] think that too many governments getting
[53:51] involved I mean everybody's going to
[53:53] want their own ecosystem and it just I
[53:55] mean I think that it there's a lot of
[53:56] risk there.
[53:58] Yeah, you could have a rebuild risk. Um
[54:00] it is very good in my opinion that we
[54:02] will reshore some of the stuff here in
[54:04] the US so we can kind of control it. The
[54:05] good news is that you know, this
[54:06] industry is still very much
[54:09] um led by the innovations coming out of
[54:12] the US companies.
[54:14] And so that's not always the case. You
[54:16] know, there's really interesting data as
[54:18] well. If you look at just this is this
[54:20] is just a broader sort of question
[54:22] around tech and what's driving
[54:24] innovation out there. If you take the
[54:26] 100 most valuable companies in the world
[54:29] so the 100 global most valuable
[54:32] companies
[54:33] um
[54:34] two thirds of those are here in the US
[54:36] and they're 80% of the value. Yeah.
[54:38] Incredible.
[54:39] Within technology um they're basically
[54:42] about kind of like the same percentage
[54:43] of you know, tech in the S&P. It's about
[54:46] 36% of the total.
[54:48] And so it's pretty it's pretty amazing
[54:51] um how that plays out here uh for our
[54:54] companies in the US. It's just um so 36%
[54:57] is basically of those you know, the the
[54:59] 100 most valuable. That's
[55:02] and 80% of that number are US companies.
[55:06] And so that that's that's very valuable.
[55:08] So I guess the risk is that you see
[55:10] other countries that look to do things
[55:12] create some of these inefficiencies. I
[55:13] agree with you. When the world was flat
[55:15] and you could just have supply chains
[55:17] that just were logical to be in
[55:19] different parts to leverage the
[55:20] capabilities that was a good thing.
[55:23] Um I think we'll just need to see
[55:24] incentives to make sure that happens you
[55:27] know, on our soil here and just make
[55:28] sure that we've got the critical
[55:30] capabilities to make sure that we just
[55:32] continue to lead in whatever is next for
[55:35] technology and for the you know, for the
[55:37] the world.
[55:38] So do you worry about the
[55:41] semiconductor companies in the US being
[55:43] shut out of the China market and
[55:44] therefore just losing um what is a huge
[55:47] market opportunity?
[55:49] You know, um you look at the at the
[55:51] global economy as I said before it's a
[55:52] $120 trillion.
[55:54] Um we're about 22% of that and China's
[55:58] like 17. So between the two of us we're
[56:00] 40%.
[56:01] So it would be
[56:03] unfortunate if we can't have access to
[56:05] the China market. Um I tend to think
[56:08] though that we will and so it may be
[56:10] portions that are excluded but I think
[56:13] in general we're going to probably find
[56:15] that there will still be an ability to
[56:16] basically go do business in other parts
[56:19] of the world. Um and you know, we're
[56:22] we're kind of all dependent on one
[56:23] another. And at the end of the day
[56:25] innovation and execution becomes mission
[56:28] critical for how our companies will
[56:31] continue to evolve and will we still be
[56:33] the biggest economy in the world a
[56:35] decade, two decades, 50 years from now?
[56:38] Um I hope so. I'd like to believe we
[56:40] will because we've got still incredibly
[56:42] bright, innovative, hard working people
[56:44] that want to continue to to change the
[56:47] world around us. But you know, we've got
[56:49] it's it's a big world and there's lots
[56:51] of competition out there. And planned
[56:54] economies have in some sense an
[56:57] advantage but they also have a lot of
[56:59] challenges, too.
[57:00] Yeah, because I mean you know, this
[57:02] isn't a planned economy, right? In the
[57:03] US and as you said we own 80% of the the
[57:07] value. I mean so that's it's pretty
[57:08] incredible we did that without any
[57:10] government involvement at all. Right.
[57:12] You know. Very little. Exactly. And so
[57:15] And the government involvement doesn't
[57:16] have to be just money. Right? You know,
[57:18] I think back like China was very
[57:20] deliberate about how it crafted cuz it
[57:22] came out of a you know, totally
[57:24] eventually 100% of the GDP was was owned
[57:26] by China. It was owned by the government
[57:27] at one point. And they were very
[57:29] deliberate and I've watched them go
[57:30] through these different phases. They
[57:32] wanted to create big companies that
[57:34] could compete get it on a global scale.
[57:36] Right? So they took critical industries
[57:38] petrochemicals, oil and gas. They kind
[57:40] of took it down to three companies.
[57:42] Right? They had an offshore oil company.
[57:43] They had two petrochemical companies.
[57:46] They did the same in telecom. They
[57:47] wanted to create a kind of real winner
[57:48] that could compete with
[57:50] the petrochemicals wanted to compete
[57:51] with Exxon and the Shells of the world
[57:53] and the BPs of the world. The telecoms
[57:54] had to compete with the AT&T, the
[57:56] Verizons, the Vodafones, the British
[57:58] Telecom etc. They they're trying to do
[58:01] the same in technology. And I think in
[58:02] the US we've had the
[58:05] because of
[58:06] the foundation of our business our
[58:08] capitalism and and how we've we've built
[58:10] these big businesses but and they've
[58:12] kind of grown organically but I do think
[58:13] that the government support doesn't have
[58:15] to just be money but it has to be
[58:17] enabling them to be able to make
[58:18] investment decisions that actually make
[58:20] sense in this country from whether it's
[58:22] a regulatory perspective, whether it's a
[58:24] fees and permitting perspective. You
[58:26] know, there's all these things that can
[58:27] be helpful to these businesses to to
[58:30] want to incentivize them to be able to
[58:31] put money into put sort of steel in the
[58:33] ground, put boots on the ground to be
[58:35] able to build these industries. So I
[58:36] think I I'm very optimistic about that
[58:38] piece of it. Um cuz we've shown we can
[58:41] compete. These industries can compete.
[58:44] You know, we're competing with somebody
[58:44] who may have a lower cost. Right. But I
[58:47] think we can compete and we can allow
[58:49] that to happen here because we'll be
[58:51] probably the we should be the best at
[58:53] automating. We should be the best at
[58:54] leveraging artificial intelligence and
[58:56] putting intelligence into our workflows
[58:58] and processes. That's going to enable us
[59:00] advantages. And I think as long as the
[59:02] government is they don't have to
[59:03] necessarily
[59:04] [music]
[59:04] seed it with capital Mhm.
[59:07] they can kind of create that sort of
[59:08] environment in a number of different
[59:10] ways that doesn't necessarily I think
[59:12] lead to inefficiencies in the way the
[59:14] businesses are run. Yeah, when when uh
[59:16] you know, macro point there is that I
[59:18] think it's it's fundamentally hard for
[59:20] governments to pick winners. And so what
[59:22] you just described Yeah. and maybe
[59:24] that's the challenge. It's hard to pick
[59:26] a winner. You can maybe do it today but
[59:28] things change pretty quickly. So you go
[59:31] back and look at what's made the United
[59:33] States so great and why we got 80% of
[59:36] the value of those companies here.
[59:38] I think it's because you've just got
[59:40] this innovation and because capitalism
[59:42] is kind of ruthless. Right.
[59:44] And so if you don't execute, if you
[59:45] don't continue to innovate, you get run
[59:47] over.
[59:48] And so look at, you know, just take a
[59:50] look at how our industry has evolved.
[59:52] Companies like
[59:53] Digital Equipment Corp, Compaq.
[59:56] Like there's reasons why those companies
[59:58] don't exist anymore, but they got
[59:59] replaced by other things.
[01:00:01] Right. So again, you take like that
[01:00:04] example I gave you of 2010 versus now
[01:00:07] within compute, HP was on that list in
[01:00:10] 2010. They're no longer. It's replaced
[01:00:13] by Dell.
[01:00:14] And within semiconductors,
[01:00:16] you would have had Intel, Qualcomm, and
[01:00:19] TI on that list in 2020 of the five most
[01:00:22] profitable companies in semiconductors.
[01:00:24] They're not in the list anymore today.
[01:00:27] And so are they great companies?
[01:00:29] You know, maybe, but things change and
[01:00:31] so if you have a forcing function of
[01:00:34] like a government entity that's going to
[01:00:36] try to dictate who is going to succeed,
[01:00:38] what's the national champion, I don't
[01:00:40] know if that's great, especially in
[01:00:41] technology where the rate of change is
[01:00:43] so fast.
[01:00:45] Yeah, I mean we've you've seen it in
[01:00:47] China. You've seen, you know,
[01:00:50] despite the fact that
[01:00:52] that they supported SMIC for example was
[01:00:55] a state champion and you know, never
[01:00:57] quite I mean you know, now SMIC is doing
[01:00:58] fine, but the point is is they're not
[01:01:00] going to compete. Yeah.
[01:01:02] Exactly. Exactly. Okay, so
[01:01:05] we'll end on a couple of questions. One
[01:01:07] is what what can change in this next 6
[01:01:11] to 9 months that would change your mind
[01:01:14] about AI and maybe how enthusiastic you
[01:01:17] are?
[01:01:19] Yeah. Well, I'll I'll just say it's if
[01:01:21] we don't get the returns. If we can't
[01:01:23] get the things that are basically
[01:01:26] driving the investment, the usage model,
[01:01:29] that would obviously change my view, but
[01:01:32] I'd be very, very surprised cuz again, I
[01:01:33] think we're very early on in terms how
[01:01:36] that goes.
[01:01:37] Um I guess you could have intervention
[01:01:39] that says we're not going to allow you
[01:01:41] to do certain [music] things with it.
[01:01:42] That could be an issue well in terms of
[01:01:44] how regulation goes on that. But
[01:01:47] otherwise, 6 to 9 months, I mean we're
[01:01:49] going to see still a lot of change.
[01:01:51] Um but I think it's it's hard to unseat
[01:01:55] it in that period of time. Yeah. I'm
[01:01:58] probably most like the question is more
[01:02:00] like what's the risk. I think the upside
[01:02:02] would be the new models that are being
[01:02:04] trained on Blackwell will be coming out
[01:02:06] over the next three to four or five six
[01:02:08] months. So I actually think you're going
[01:02:09] to see a lot more interesting things
[01:02:11] that are really supportive to the sort
[01:02:14] of curve that we're on, the exponent
[01:02:16] that's curve of
[01:02:17] performance continue to stay. I think
[01:02:19] the models that are trained on Blackwell
[01:02:21] will be phenomenal, right? Like because
[01:02:24] again, what you see each time there's
[01:02:25] been this, yeah, you trained on you
[01:02:27] trained on Hopper, then the TPUs, the
[01:02:29] TPU and now it's going to be Blackwell
[01:02:31] and these the improvements that they
[01:02:33] continue to make and the sort of
[01:02:34] efficacy and the performance and the
[01:02:36] throughput is going to be I think again,
[01:02:39] it's going to lead to again staircase
[01:02:41] functions and the ability of these
[01:02:42] models to deliver value whether it's on
[01:02:44] coding or whether it's on, you know,
[01:02:46] agentic commerce or e-commerce or other
[01:02:48] things like that. So I actually think
[01:02:50] that when you start to see these results
[01:02:52] of these models start to flow through
[01:02:53] probably like summer late second quarter
[01:02:55] summer time into the fall, I think
[01:02:57] that's going to be actually another
[01:02:58] catalyst as opposed to something that's
[01:03:00] negative. Yeah. So do you worry about
[01:03:02] memory or energy slowing
[01:03:05] or capacity? Those again, those three
[01:03:07] those were the three things that our
[01:03:08] CEOs
[01:03:10] put up flags about. I mean
[01:03:13] those have to be solved, right? They
[01:03:14] have to be solved. I mean again, I think
[01:03:16] it's, you know,
[01:03:18] it's hard to solve that those are those
[01:03:19] are not easy issues to solve. And so
[01:03:23] I think it'll we'll still be
[01:03:24] constrained,
[01:03:26] but I think you'll probably see a
[01:03:27] reallocation of how where they where
[01:03:29] people will put resources to what level
[01:03:31] of compute. Um you know, people are
[01:03:33] finding ways to do things like creative.
[01:03:36] You're starting to see all these
[01:03:37] companies originally they all wanted to
[01:03:39] be kind of on the grid. Now they're
[01:03:40] actually going behind the meter, right?
[01:03:42] They're trying to find different ways to
[01:03:44] kind of get power. They're trying to be
[01:03:45] creative with, you know, managing the
[01:03:48] power. They're I mean there's a data
[01:03:50] center business that's actually
[01:03:51] leveraging, you know, kind of old Tesla
[01:03:53] batteries as kind of backup for their
[01:03:56] power because, you know, if you're
[01:03:57] behind the grid, you need to smooth the
[01:03:59] power. And you can't you know, cuz you
[01:04:01] can't burn the chips with like huge
[01:04:03] fluctuations in power. So there's all
[01:04:05] kinds of things that are being created.
[01:04:06] Including coal by the way is coming
[01:04:08] back, right? Yeah, modular nuclear. I
[01:04:10] mean they they talk about everything,
[01:04:11] but again, power is still I think you
[01:04:13] the constraints are still there, which
[01:04:14] is again,
[01:04:15] it will probably I don't think it
[01:04:17] necessarily impedes performance cuz
[01:04:18] people are going to dedicate the leading
[01:04:20] edge of the model like the resources of
[01:04:21] leading edge of the model. It may have
[01:04:23] an issue for other things, which may
[01:04:25] mean there's opportunities for the
[01:04:27] inference chip providers and other
[01:04:29] people that are coming out to be able to
[01:04:30] take on some additional sort of
[01:04:32] resources, but it's still we're going to
[01:04:34] be in a world where we're still
[01:04:35] constrained for the next series of
[01:04:36] years. It's just not
[01:04:38] at least in the US for sure. And that'll
[01:04:40] I think, you know, there'll be parts of
[01:04:42] the ecosystem that are going to be more
[01:04:43] challenged. So I think because AI is so
[01:04:46] powerful, so important, that's where
[01:04:48] people are going to resource. And so
[01:04:51] when you get into periods like we're in
[01:04:53] today and probably will be for a while
[01:04:55] where things are more constrained,
[01:04:57] there'll be other segments that end up
[01:05:00] being kind of you know, pushed aside if
[01:05:02] you will, so elbowed out. That'll be one
[01:05:03] of the issues I think for other big
[01:05:05] memory consumers historically like
[01:05:07] things like PCs and handsets and maybe
[01:05:09] consumer items. Those will be hard to
[01:05:11] probably go and, you know, and get. So
[01:05:14] probably means that prices go up. You
[01:05:16] could see, you know, issues in terms of
[01:05:18] just demand for like if a if a
[01:05:20] smartphone's going to become more
[01:05:21] expensive as an example, does that slow
[01:05:23] demand? Okay. So that doesn't impact the
[01:05:25] AI situation, but that is just maybe
[01:05:29] some of the Slow down. Yeah, exactly. So
[01:05:31] that'll be one of the issues. And the
[01:05:32] other thing just to kind of be watching
[01:05:33] out for I think is that because
[01:05:35] um the growth here of this AI wave has
[01:05:37] been so significant and we talked before
[01:05:39] like the industry is heading to a
[01:05:41] trillion dollars. We thought that was
[01:05:42] going to happen in 2030. It's 2026.
[01:05:46] So the rate of change has been pretty
[01:05:48] significant and one of the things that
[01:05:50] happens unfortunately with investors is
[01:05:52] second derivative of that. So it's not
[01:05:55] enough just to keep growing. Uh for a
[01:05:57] lot of investors that are momentum
[01:05:58] oriented, it becomes the rate of change.
[01:06:01] And so when that second derivative
[01:06:02] starts to change, that's where it
[01:06:04] becomes more problematic. So as an
[01:06:05] example, just look at the last 3 years
[01:06:08] of just CapEx for what's happened in AI
[01:06:11] infrastructure.
[01:06:13] It was about 60% in 2024, about 65% last
[01:06:18] year. This year is about 70% growth.
[01:06:19] These are huge numbers now. $800 billion
[01:06:22] to spend this year. Is it going to grow
[01:06:25] another like 75% in 2027? Maybe, but
[01:06:30] that's been driving obviously a lot of
[01:06:31] the economics of the semiconductor
[01:06:34] companies. And so for all of tech what
[01:06:36] we have to really kind of focus on is
[01:06:38] when we start to reach that inflection
[01:06:39] point where the growth starts to slow.
[01:06:42] So it's still growing, but it starts to
[01:06:43] slow. That's where you might have some
[01:06:46] issues that it'll be kind of hard for
[01:06:47] people to understand. Is that because
[01:06:49] there's a demand problem coming or just
[01:06:51] because things got a little bit over
[01:06:53] inflated when the broad market trades at
[01:06:55] the P multiples that today? That'll be
[01:06:57] kind of a big thing to focus on and I
[01:06:59] suspect that's the discussion we'll be
[01:07:01] having, Okay. you know, towards the end
[01:07:03] of this year and 2027 because,
[01:07:05] you know, I just don't know. Like the
[01:07:06] semi industry this year is going to grow
[01:07:07] like 45%.
[01:07:08] Yeah. I doubt it's going to grow 45%
[01:07:10] again in 2027. I'd love it, but it seems
[01:07:13] like it won't. Right.
[01:07:16] Okay, so last question. What are we
[01:07:18] underestimating about AI?
[01:07:22] Hm.
[01:07:25] I mean I think it's going to be just the
[01:07:27] the pervasiveness and durability and the
[01:07:31] things that
[01:07:33] today we can't still contemplate just
[01:07:35] like you couldn't have contemplated 5
[01:07:38] years ago what it would look like.
[01:07:39] Right. And so um and that's, you know,
[01:07:43] that could be scary um cuz you don't
[01:07:45] know how it's going to go. It won't be
[01:07:47] just, you know,
[01:07:49] people driving it. It'll be,
[01:07:51] you know, the the AI driving something.
[01:07:54] So
[01:07:55] that'll be one of the things that I
[01:07:56] think will be important to think about.
[01:07:58] I think it's when you think about like
[01:08:00] we all kind of grew up with all these
[01:08:02] great you know, great things about
[01:08:04] technology, you know, we were kind of
[01:08:05] all grow up. We remember the moon
[01:08:07] launches and all these things like that,
[01:08:08] right?
[01:08:09] And you know, we all thought we'd be,
[01:08:11] you know,
[01:08:12] traveling in supersonic plane. You know,
[01:08:14] I I'm a big like you know, I'm not about
[01:08:16] that stuff like technology. And so what
[01:08:17] I think that we're probably when I think
[01:08:19] about AI, what I think it really will
[01:08:20] enable that all the great things about
[01:08:23] how we live our daily life, things will
[01:08:24] probably get better. But I think it's
[01:08:25] actually when I think about it, I think
[01:08:27] like a lot of it is like the
[01:08:28] infrastructure of the future. Right?
[01:08:30] It's going to enable so many different
[01:08:31] things whether it's in robotics,
[01:08:35] automation, or autonomous, whether it's
[01:08:37] transport, or whether it's delivery of
[01:08:39] goods. Um it's going to enable, you
[01:08:42] know, as you think about space travel.
[01:08:44] Like the the need to have that
[01:08:46] intelligence like go beyond just what we
[01:08:48] do here. I think there's tons of
[01:08:50] opportunities. So I think it's like it's
[01:08:52] to me like the it's the next like the
[01:08:55] it's the infrastructure upon which we
[01:08:57] build the next kind of kind of growth
[01:08:59] segment of the economy.
[01:09:01] If you think about like things were
[01:09:02] built on railroads, they were built on
[01:09:04] steel, they were built on semiconduct
[01:09:06] this is just AI is now the next layer of
[01:09:09] that. And so I think, you know,
[01:09:11] again, it's probably not something you
[01:09:12] see over the next couple of years, but
[01:09:13] like my personal view is in the next 20
[01:09:15] years, it's going to be we're going to
[01:09:16] look back. It's going to be a little bit
[01:09:18] like I remember like we had a research
[01:09:21] analyst Mary Meeker who said Amazon's
[01:09:23] going to be this amazing stock. It's
[01:09:24] going to be worth like $400 a share.
[01:09:26] Everyone's like, you're nuts. It's never
[01:09:27] going to get there." Turns out 400 was
[01:09:30] like really, really low. And so I think
[01:09:32] the the it's hard to think that big, but
[01:09:35] I do think this could be something that
[01:09:37] really changes the fabric
[01:09:40] of what we know. And so I think it's
[01:09:42] pretty and it's pretty exciting to think
[01:09:43] about cuz hopefully I'm we're all still
[01:09:45] alive and you look back 25 years from
[01:09:47] now and it's going to be wow, we we were
[01:09:49] underestimating how much it changed the
[01:09:51] way we live and changed what
[01:09:54] society is. Well, you left out
[01:09:55] healthcare, so I think the odds are good
[01:09:57] if we live long enough and the
[01:09:58] innovations continue, we will actually
[01:10:00] definitely live longer, I think. I think
[01:10:02] it's pretty I think it's going to be
[01:10:03] fascinating. Fascinating to watch.
[01:10:05] Well, when I started out I said that
[01:10:07] after I interviewed these CEOs of these
[01:10:10] AI chip companies that you get so
[01:10:11] excited about the future opportunities
[01:10:13] and so you didn't
[01:10:16] you got me really excited again. So I
[01:10:19] think that
[01:10:20] great great days ahead. So appreciate
[01:10:22] you guys being here Thank you. and
[01:10:24] sharing your thoughts with the with the
[01:10:26] listeners.
[01:10:26] Appreciate it. Thank you very much. It's
[01:10:27] going to be an interesting future. That
[01:10:28] will be.
[01:10:32] [music]