Latest / UBS Trending / The age of intelligent machines

Transcript

The age of intelligent machines · UBS Trending

0:00
We know who's selling the picks and shovels of the AI revolution.
0:03
But the bigger question is
0:05
who ultimately captures the value
0:07
from semiconductor stocks to a genetic
0:09
AI to the industry's most likely to be transformed over
0:12
the next decade?
0:13
We're going to cover that
0:14
and a whole lot more right now with Al Crawford on
0:18
UBS trending.
0:30
Hi everyone, and welcome to UBS trending I'm Anthony Story.
0:33
So nice for you to join us today
0:35
and happy to be in the studio with my guest
0:37
anchor Crawford again from Alger anchor.
0:39
Nice to have you back.
0:40
It's been a little bit since we've had you in.
0:42
Yeah. It's always fun to be here.
0:43
Yeah. Good. Good to have you.
0:44
So, you know, we've got a lot
0:46
I can talk about
0:47
so many different things
0:48
with you and especially around the AI revolution and evolution.
0:51
I need to talk to you
0:52
about semiconductors, however,
0:53
and we've been talking about this here
0:54
at UBS and our chief investment office.
0:56
We're looking at some declines
0:58
in the semiconductor index
0:59
over the last month or so, roughly 20%, give or take.
1:02
And there's been a lot of questions coming up about,
1:05
you know, is this an overcrowded trade?
1:07
Is there something fundamental
1:08
that's happening that maybe investors should be aware of?
1:11
And obviously a lot of it
1:12
is because of or surrounding the AI story?
1:16
What are you thinking about when you're looking
1:18
as an investor yourself at semiconductors
1:20
and this bit of a decline we've seen?
1:22
Okay.
1:23
So, before we talk about
1:25
before I answer your question,
1:27
I think I would like to give you some history of semiconductors.
1:29
Yeah.
1:30
And I feel like I'm in a good position to do it
1:33
because I was our semiconductor analyst at one point.
1:36
So when I was an analyst,
1:39
there were 50, 70 different semiconductor
1:42
companies that made different widgets.
1:45
And it was an incredibly crowded trade.
1:48
Or in crowded, landscape.
1:50
There was seven memory companies.
1:52
There were, you know, six companies that competed
1:55
for for selling ten cent parts.
1:59
There were four FPGA companies.
2:02
There were two CPU companies.
2:04
There were, you know, so there was basically a very,
2:08
kind of a crowded, competitive landscape.
2:10
But what happened over the last two decades was that
2:14
every generation was more and more difficult to make,
2:18
and we call those nodes.
2:19
And so as you shrink a semiconductor, it was more expensive.
2:24
And that expense was so high
2:27
that there was a forced consolidation.
2:31
Because not everyone could actually cross the chasm and build
2:35
the next iteration of a semiconductor chip
2:38
and still price it competitively.
2:41
So it forced this consolidation in the industry,
2:45
and effectively made
2:47
many parts of the semiconductor industry from capital equipment
2:50
all the way to the end product to to the chip end product.
2:55
Effectively, I would say duopolies.
2:59
So you marry that history with what is happening today.
3:04
The one thing that we need that is essential for
3:08
AI is compute and intelligence and memory.
3:15
And these are the affected
3:18
industries that don't really want to add supply,
3:22
you know, into a demand signal that looks like a hockey stick.
3:27
If we had infinite capacity in the supply chain,
3:30
maybe we would have created a bubble.
3:33
Maybe we would have created,
3:35
you know, too much AI capacity on the ground.
3:38
But this is actually a thing that we should appreciate.
3:42
And it helps multiples because in fact, it's
3:47
capping the growth of the industry of the overall
3:50
AI industry, and it gives us a longer duration.
3:54
Right now,
3:55
you know,
3:55
there are more and more complicated chips and compute
3:58
that are going to be needed over time.
4:00
We're already seeing that. Now.
4:01
Does that give you
4:02
any kind of concern
4:03
about the growth of these semiconductor companies?
4:06
And by the way,
4:07
we're seeing massive growth in the emerging market side
4:10
Taiwan, Korea, they are exploding because of their
4:13
manufacturing capabilities
4:15
in the semiconductor and microchip space.
4:17
But do these more complicated
4:19
compute needs get in the way of growth?
4:22
No, I think look, we
4:25
the semiconductor industry
4:26
is actually at the forefront of technology.
4:28
They always have been.
4:30
So these are engineering problems, not science problem.
4:34
And I think that I would worry if we have a science problem
4:37
because that that is really being limited by the laws of physics,
4:41
not by the laws of engineering.
4:42
Engineers will figure out ways to engineer around an issue.
4:47
The laws of physics cannot be bent.
4:49
So no, I don't I don't worry that the complexity
4:52
is going to hinder the growth of the industry.
4:57
Could you have delays here and there? Sure.
5:00
I worry more so that,
5:02
you know, could do we not have enough capacity such that
5:06
it hinders the growth of AI because the demand for AI
5:09
and those tokens are just immense, right?
5:13
Okay.
5:13
So switching gears just just a little bit,
5:16
when you look at the hyperscalers,
5:18
their CapEx is expected to exceed
5:21
half $1 trillion, half $1 trillion this this year alone.
5:25
And I saw you in an interview from a little while ago
5:28
where you talked about
5:29
how you expect CapEx to continue to rise through 2028,
5:33
given everything that's happening now and the continued,
5:37
sort of evolution, but also astronomical
5:40
evolution of AI and CapEx, is 2028 still your call?
5:43
How are we doing there?
5:44
Yeah.
5:45
So, you know, so I made that call in 2024,
5:49
that we would basically have rising CapEx through 2028.
5:52
And I would say as of last night,
5:54
we got our first data point on 2027,
5:58
where, you know, one of the
5:59
hyperscalers came out and said that CapEx in 27
6:02
would rise significantly.
6:04
So we're kind of 75% of the way there.
6:07
On this call, when I made the 2028
6:09
call, that wasn't necessarily the peak.
6:13
It was just how how long I was willing to look out.
6:17
And it also is a function of how much CapEx gets spent.
6:20
So if we put in $3 trillion of CapEx into the ground in 2028,
6:25
then my view on 29 and 30 would be amended.
6:29
But really, I do think we have a long runway for CapEx,
6:33
ahead of us, and in part because we are at the very early innings
6:37
of how AI is going to impact us and how it is going to be used.
6:41
So,
6:43
you know, we're going to
6:44
need all of this compute to support our demand, right?
6:48
Yeah.
6:48
And I guess you'll have to keep looking at the earnings
6:51
reports and information
6:52
that's coming out from these hyperscalers to determine
6:55
how far down the road do you want to go past 2028 for CapEx?
6:58
I mean, it could continue to go through the 2030s.
7:01
Yes, it absolutely could.
7:02
But again, I think it's going to be a function of how much CapEx
7:05
we put in the right,
7:06
because there is going to be a level of CapEx
7:09
that is too much for society
7:11
to absorb, i.e.,
7:13
if we had put in $5 trillion of CapEx into the ground last year,
7:17
we're just not ready to absorb that.
7:20
Even if we had no supply supply chain constraints,
7:23
we had no electricity constraints.
7:25
We simply could not absorb it.
7:27
Yeah.
7:27
I mean, there's like change functions,
7:30
for society, like it takes some time to truly change.
7:34
Yeah.
7:35
So I think it does just
7:36
depend on, on how CapEx gets put into the ground
7:41
and when
7:42
interestingly, when you look at the different phases
7:44
of the AI evolution,
7:46
obviously CapEx is extremely important in this evolution.
7:50
Phase one was really about infrastructure
7:52
and the advent of large language models.
7:54
And now phase two is more about a genetic I do you
7:57
what is the phase three?
7:58
I mean, who knows how many phases there will be ahead of us.
8:01
But do you predict what phase three is,
8:02
or are you maybe even seeing phase three starting to take shape?
8:06
So what I would argue
8:08
is that
8:09
we may be going through these phases in terms of the impact,
8:12
but it doesn't mean that phase one is still not benefited.
8:17
Right?
8:17
So phase one, yes, is like the early infrastructure,
8:22
that is getting benefited from what is happening.
8:24
Phase two is it could be the genetic.
8:28
But I would argue that
8:30
that in that agency that we will have using a genetic
8:34
AI has a very long runway.
8:37
Right.
8:38
So, so agents are going to impact our lives
8:41
and impact what we do for forever.
8:46
And so instead of thinking about it as phases,
8:48
because phases makes it looks like it like there's a peak
8:51
and then there's a drop off. Right.
8:53
What I would say it's like steps that build upon one another.
8:58
Versus
9:00
versus something that that peaks and comes back down.
9:04
But look, agents are a very big deal.
9:06
It's changed
9:07
the way that we actually compute this idea of the CPU
9:11
orchestrating agents versus a GPU, has been,
9:16
something that's driven a lot of different stocks.
9:20
The next phase, I believe,
9:21
is the actual implementation of artificial intelligence..
9:26
After that,
9:27
it will be about robotics and autonomous and physical.
9:32
That is
9:33
truly the next phase that I think comes out.
9:36
And I again, I'm not gonna use phase the next step,
9:40
that comes probably at the end of this decade
9:43
and it is going to be game changing.
9:46
Wow.
9:46
So imagine I mean, I have a I have an 18 year old.
9:50
She doesn't want to drive.
9:51
And I'm kind of thinking, well, maybe she doesn't ever
9:53
have to learn how to drive, because in 4 to 5 years,
9:57
I do think we'll have kind of a ubiquitous,
10:00
maybe not 4 or 5 years by the end.
10:02
By mid 2035 or mid 2030s,
10:05
we'll probably have some sort of a more ubiquitous
10:09
robotaxi network
10:11
where, you know, we'll question whether or not
10:13
we need cars, right?
10:14
I mean, they're testing them more and more in major cities LA,
10:17
San Francisco, Texas, various cities there.
10:19
And you know, to, I think mostly success.
10:23
That's right.
10:23
People are getting used to seeing driverless cars
10:25
just coming down the street waiting to pick them up.
10:27
And I was just in Austin.
10:29
I in one of them, by a company called AV ride.
10:32
And it is mind blowing.
10:34
Wow. It is truly mind blowing.
10:36
And beyond that you have robotics.
10:38
And so having a robot robot in your home
10:41
that kind of looks and feels like like you to some extent,
10:45
it can do
10:45
everything that you do
10:46
in the confines of your home, in your office.
10:49
I think that will become the norm.
10:51
So that is that is how I see the steps
10:54
of AI building upon upon each other.
10:57
Wow. So much, so much highway.
11:00
Oh AI roadway ahead of us with this evolution, I want to ask you
11:04
before we get to wrap up, though,
11:06
you know, I it's it's obviously transformative.
11:08
It's on par
11:09
with the internet or even electricity,
11:11
you know, from 150 years ago.
11:13
So as we move
11:14
beyond sort of this early excitement
11:16
and the talking of where AI is going,
11:19
what which
11:20
industry application
11:21
do you think will be sort of the next
11:23
has the greatest potential maybe to
11:25
to really see some success or potential greatness
11:29
with these steps that you're talking about as we go into
11:33
the future applications of artificial intelligence?
11:35
Okay.
11:35
So I actually think that
11:38
every industry is going to be transformed.
11:40
So choosing choosing the one that's going to be impactful
11:43
is really a function of which one has the most impact on GDP.
11:47
And I think it's health care.
11:49
That will really be a massive beneficiary.
11:51
Big opportunities
11:52
there, big opportunities to bend the cost of health care.
11:57
You know, and even if you take it down to robotics and,
12:00
and what if we had robots actually having a one on
12:03
one kind of you had a one on one robot for every senior,
12:07
in our,
12:08
you know,
12:08
every senior person who is, you know, you have a companion,
12:13
you know, that that takes care of them.
12:16
But I think as we look out to kind of like
12:18
2030 into middle of next decade,
12:22
what I am most excited about is how
12:25
robotics is going to transform our economy.
12:28
If you think about robots, one of the early reasons
12:33
that we outsourced not only call centers, but,
12:37
you know, all of our manufacturing that
12:39
I mean, in the 60s through the 80s, 25% of our workforce
12:44
had to do something with manufacturing, right?
12:47
Because we had such a strong manufacturing base.
12:50
And that came along with, you know, a,
12:54
a, a huge emphasis
12:56
on engineering and all of these, these skill sets,
13:00
whether it was like building and, electricians and plumbers.
13:05
I think there's all of these skill sets that, over time,
13:08
have become whittled away in our society because we no longer
13:12
have that manufacturing footprint,
13:13
because we've outsourced so much of it to to other countries.
13:17
And the reason we did that, the sole reason was labor
13:22
costs, right?
13:23
Of course, we have cheap electricity.
13:25
We have the consumer right here.
13:27
We don't have to pay taxes and tariffs and transportation costs.
13:31
If it was made right here.
13:33
However, the cost of labor was so like the
13:37
that arbitrage was so big that there was no choice.
13:42
Well, I argue that as we introduced robotics
13:47
and the cost of robots becomes, you know, quite cheap,
13:52
we will be able to be, equivalently,
13:55
as competitive as where we outsource to,
13:58
which will allow us to bring manufacturing back to the US.
14:03
And it truly begins our industrial renaissance.
14:07
So I am really excited about the long term
14:11
for for the US economy. Wow.
14:14
I think that's a perfect way to end our conversation,
14:17
because I think that takes us into the future
14:19
and who knows where we're going.
14:21
But it sounds like there's a lot of optimism
14:23
from where you're sitting.
14:24
So thanks, anchor. Great to have you back in the studio.
14:26
Good to be here.
14:27
Good to see Augur Crawford from Alger.
14:30
Great. Great conversation.
14:31
And thank you all for sticking around.
14:32
And we hope that you look for more of this content
14:35
on our website.
14:36
Ubs.com/views, tons of insights from our great partners
14:40
like Alger,
14:40
as well as our own chief investment office here at UBS.
14:43
Plus, you can follow us on social media.
14:45
We're on LinkedIn, Facebook,
14:47
and we also have a dedicated Instagram channel for the studio.
14:50
Follow us at UBS. Trending. Lots of content there.
14:52
And as always,
14:53
if you have any questions about what
14:54
Arthur and I spoke about today, please continue this conversation
14:57
with your financial advisor.
14:58
Until next time, I'm Anthony Pastore.
15:00
We wish you a great rest of your day, and remember
15:02
to keep your eyes on what's trending.
15:04
We'll see you soon, everybody. Thanks.