YT YouTube2Text

YouTube2Text → Transcripts → Top JPMorgan Strategist: 6 sectors to BUY for the NEXT AI boom!

Top JPMorgan Strategist: 6 sectors to BUY for the NEXT AI boom! — Transcript

by Phil Rosen · 7,469 words · 1,166 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00The AI bull market's been going strong
0:02for like 3 years now. Where do you think
0:04we are today in this AI cycle?
0:07>> 3 years and boy, a lot has happened,
0:10right? Now, at the risk of sounding like
0:13a broken record, okay? We are still
0:15early, believe it or not. Um and we
0:18think that across a number of different
0:20metrics, but that doesn't mean that the
0:23same investment strategy that worked 2
0:25or 3 years ago is the one that's going
0:26to work the best next year.
0:28So, how is that the case? Over the last
0:303 years, what companies have been
0:32driving this AI wave? Well, they've been
0:34predominantly the AI infrastructure
0:37beneficiaries and that US hyperscalers,
0:39right? Which are the companies that are
0:41really driving a lot of this buildout
0:42and investing in it. Now, going forward,
0:45we're not I'm not going to say that it
0:47is time to move away from the
0:49infrastructure companies. I'm actually
0:50going to say it is
0:52interesting time to look closer at them
0:54because of where valuations have reset
0:56after this year, but they no longer have
0:58a monopoly on the AI opportunity set.
1:01And I don't think phase one, phase two
1:04of buildout versus adoption are mutually
1:07exclusive. I think we're going to be in
1:09an environment where there is still
1:11increasing capital being deployed around
1:14AI infrastructure. That is going to a
1:16much broader set of companies related to
1:18the buildout, but we're also seeing a
1:20lot of evidence of companies benefiting
1:22from adoption or positioning themselves
1:24as key enablers to corporate adoption of
1:26AI.
1:27>> So, just to clarify, early in the AI
1:30cycle meaning not a bubble.
1:33>> Meaning not a bubble and I can refine
1:34that.
1:35>> Okay, tell me more.
1:36>> Yeah, well, look,
1:38one thing that we did in a recent paper,
1:41um AI conscious asset allocation, was we
1:44looked back through history at all of
1:47these capital investment booms and
1:49busts. And we were surprised that we
1:51found a pretty consistent like six-phase
1:53progression through all of this. And the
1:55bust that everyone is afraid that seeing
1:57right now is comes about this midpoint
2:00where really elevated lofty euphoric
2:03expectations hit a harsh reality.
2:06I will say there are warning signs this
2:08year that some expectations have hit a
2:11tougher reality. You're seeing that in
2:13volatility particularly in credit
2:15markets around some of the riskier names
2:16in this build out. However, what we
2:19think really grounds us in an
2:21environment that is far from a bubble
2:23bursting is just how strong demand is
2:26and in how sturdy the build out appears
2:30given its largely cash financed nature.
2:32Now, that is changing but the fact that
2:35still the predominant amount of capital
2:37being deployed right now is from highly
2:39profitable firms gives this boom I think
2:41a real air of resiliency.
2:43>> So, let's talk about the financing for a
2:45sec. A lot of these neo cloud companies
2:48Coreweave, Iron, Nebius these stocks are
2:51down like 30% in the last month or so.
2:53They've been getting absolutely crushed
2:55and I think a big reason is that people
2:57are worried about how they're being
2:58financed and their deals tied to the AI
3:02economy. Does any of that concern you?
3:05>> So,
3:06in this AI race not all balance sheets
3:08are created equal.
3:10And what we're seeing right now is with
3:12the rise in interest rates the cost of
3:14capital does matter to this build out.
3:17Now, normally when you're building out
3:20massive infrastructure and you're
3:22leaning heavily on debt a hike in
3:24interest rates can be very problematic.
3:26That's actually what led to a lot of the
3:28troubles that we had in the telecom
3:30build out, right? The good news is today
3:33there isn't as much of that reliance on
3:35debt but some individual players are
3:38more sensitive to this. So, what's going
3:40on with the neo clouds? It's not so much
3:42that all of the sudden borrowing costs
3:44have gotten to the point that these
3:45companies are going to go bust
3:47but we are seeing that higher interest
3:49rates to challenge the economics of some
3:51of these players, because these neo
3:53clouds, okay, they don't have the same
3:56kind of massive, diversified revenue
3:58streams as the US hyperscalers have.
4:01Right, these companies have massive
4:02e-commerce platforms, cloud businesses
4:05that they can rely upon to ultimately
4:07invest in all of this data center
4:09infrastructure. The neo clouds are in
4:11the business of building out this data
4:13center infrastructure, which is very
4:14expensive to do, so they have to borrow
4:16a lot to do that, and the revenue that
4:18they're that they're depending upon is
4:21more fixed in nature. So, I think part
4:23of what we're seeing is rational. These
4:25companies are riskier, the economics are
4:27looking a little less favorable in a
4:30in a rising rate environment, but there
4:32is also some reality that everyone here
4:34is just waiting for that warning sign of
4:37whether a bubble's going to burst, and
4:38the swiftness of the move that we've
4:39seen recently, I don't know how much of
4:41that is really due to deteriorating
4:43fundamentals or technical factors and
4:45sentiment.
4:46>> Okay, Steph, help me understand this. If
4:48we're seeing
4:49a lot of these AI companies, including
4:51the neo clouds, a lot of them come out
4:53and say, "We cannot keep up with demand
4:56for our products and services." At the
4:58same time, we are seeing a lot of the
4:59same companies say, "We need to tap
5:02different sources of financing." How do
5:04those two things exist at the same
5:09>> feature of this AI boom that is unlike
5:12anything that we've seen in history,
5:13okay? I mean, when I used to say,
5:14"Historically, once the tenure hits this
5:16level, things go like those rule books
5:19do not work in an era of AI." Given how
5:21strong the demand around this technology
5:24is, yes, CapEx is growing, right? But
5:27there is that information, at least,
5:29that these companies have to inform
5:31themselves. The challenge is,
5:33even if we all, okay, wanted to have
5:37hundreds of agents working for us every
5:38day, I don't know, maybe you're getting
5:40up to that point already, you as a power
5:42user,
5:43um but we cannot scale that across the
5:45economy yet, because in reality our
5:48technological ambitions can only run as
5:50fast as the physical world can. And we
5:53are in an environment where the actual
5:56build out of data centers is moving at a
5:59foot much slower pace. It's still moving
6:00really quickly, but relative to the
6:02technological frontier, it's slower and
6:04that has implications for the economics
6:05of this whole thing.
6:07>> Today's episode is brought to you by
6:09Direxion. Tesla headlines move markets
6:11on earnings, deliveries, and robo-taxi
6:13news alike. The Direxion Daily Tesla
6:16Bull 2X and Bear 1X ETFs, tickers TSLL
6:21and TSLS,
6:23seek 200% and -100% before fees and
6:26expenses of Tesla's daily performance.
6:29Bullish on Tesla's next catalyst or
6:31bracing for a pullback, trade either
6:33side with TSLL and TSLS. These are
6:36non-diversified and concentrated in a
6:38single company, and investing in the
6:40funds is not equivalent to investing
6:42directly in Tesla. An investor should
6:44carefully consider a fund's investment
6:46objective, risks, charges, and expenses
6:49before investing. A fund's prospectus
6:52and summary prospectus contain this and
6:54other information about Direxion ETFs.
6:57To obtain a fund's prospectus and
6:59summary prospectus, call 866-476-7523
7:03or visit the website at direxion.com.
7:06Now, let's get back to the episode.
7:10>> When you think about the hyperscalers
7:12who are really funding all of this that
7:14you're describing,
7:15do you see them well-positioned as far
7:18as their balance sheets, their cash
7:20flows
7:21for the year ahead? Cuz I I see
7:23companies like Amazon and Google, they I
7:26think just had negative free cash flow
7:28quarters. A lot of people see that and
7:30say, "Look, those were investment
7:32decisions they made. These are smart
7:34people at the helm leading these
7:35companies. Don't worry about it." What
7:37do you think of that?
7:38>> Yeah, it's tough because look, this
7:40year, despite the hyperscalers expected
7:43to grow earnings by 60% they've
7:46underperformed the broad market
7:47year-to-date and their valuations are
7:49about on par if not at a slight discount
7:51to the broad market. So, investors this
7:54year have definitely had an air of
7:55caution around these hyperscalers as
7:57they question not so much next 12-month
7:59earnings but longer-term competitive
8:02dynamics.
8:04And that's the thing to keep in mind
8:05here because again when we did this
8:07research historically, right? There are
8:10two different types of environments that
8:12we tend to see these infrastructure
8:13build-outs ultimately settle out at. One
8:16is a
8:17one where there's a monopoly or an
8:19oligopoly. So, a few companies that have
8:22really strong competitive moats and
8:23they're able to sustain more durable
8:25profit margins, okay? That's like the
8:28bull case for these hyperscalers. They
8:30are building out the toll roads for AI
8:32from which they expect to earn
8:35significant revenue from all of this
8:37demand that they're expecting to see. To
8:39their defense, they have a lot of demand
8:41visibly that they can see over the next
8:43few years. The combined cloud backlog
8:45for the top three hyperscalers is now
8:471.7 trillion dollars. That is compared
8:50to annual capex this year of roughly 800
8:53billion for all of the top five
8:55hyperscalers. So, a big mismatch in
8:58terms of a lot of demand there, okay?
9:00Um now, the less favorable environment
9:03that this could settle out at is one
9:05where
9:06you have a lot of competition and
9:08commoditization, right? Commoditization
9:11of AI is one of the big concerns for
9:13investors in this environment cuz it
9:16could erode the ROI that these companies
9:18are expecting to see. And so, your
9:20question is what do we expect next year?
9:22I think it's really going to depend on
9:25the
9:26visibility that these hyperscalers have
9:28to demand continuing and whether
9:31investors feel that the
9:34moats that are being established here
9:35have durability even as new competitors
9:38come online.
9:40>> Do you think that the moats have
9:41durability?
9:43>> I think so.
9:45I think for the foreseeable future,
9:47yeah. I mean, the reality is you have to
9:51be big to play in the world of AI. It is
9:53a very expensive market to be in. And
9:56so, it is really only these companies
9:58that can be in the market of building
10:01out this data center infrastructure and
10:02ultimately servicing that out. Where
10:05things get challenging is, of course,
10:07they they operate with that margin,
10:09right? So, when the costs of building
10:11data center compute begin to rise, and I
10:13think the surge in memory costs was one
10:16of those warning signs this year, right?
10:17Or if, for instance, we can't get the
10:21capacity online quickly enough because
10:23of this data center backlash, power
10:26constraints, and so forth. I think what
10:28will be interesting is among the
10:29hyperscalers, they might be
10:31differentiated based on the strength of
10:34their partnerships, the um, you know,
10:37supplier agreements that they may have
10:39in place, and their ability to supply
10:41compute in a very compute-constrained
10:43world.
10:44>> I think those are great points. Um, I
10:46want to ask you about some of these
10:48frontier labs. We have Anthropic,
10:50OpenAI, and there are a few other
10:52players, and the two big ones, they're
10:55planning these IPOs, right? They're
10:56going to totally rattle markets. People
10:58are wondering if it's going to suck all
10:59the liquidity out of stock market. I
11:01don't think that's going to happen. But,
11:03my my personal view is that they have
11:06much smaller moats than people think
11:09they do, and I think their token
11:11economics are probably going to worsen
11:13over time, and they have a lot of
11:14figuring out to do as they come public,
11:18before they go public. Is there anything
11:20about the frontier labs and maybe their
11:22token economics that worries you right
11:25now?
11:26>> So, really great question, and I think a
11:29lot of focus and anticipation around
11:31these IPOs this year, right? Part of the
11:33reality here is the business model of a
11:35model lab is completely new for us. And
11:38so, what we've seen over the course of
11:40this year with newfound competition from
11:42China, with the model labs also
11:45releasing their own cheaper models that
11:48that have also gained a lot of
11:49popularity, it's a market in pricing
11:51discovery. Like this is so nascent when
11:54we think about our usage of these models
11:56and us deciding what is our willingness
11:59and what is the price that we're willing
12:00to to pay for these models. So, I I will
12:03say with that said, when you see these
12:05model labs reduce their costs, that
12:07doesn't necessarily mean they're giving
12:09away money. They're leaving money on the
12:10table.
12:11Inference gross margins for the model
12:13labs look really solid. I mean, they're
12:14around 80% 70 to 80% based on the
12:18financials that our investment team has.
12:19So, there is room for these firms to
12:21lower costs because in doing so, what's
12:24happening?
12:25Jevons paradox, okay? It's expanding the
12:28total addressable market for model
12:30usage. If we have cheaper intelligence,
12:32businesses are going to find more ways
12:34to use it and we're still so early on
12:36that front. So, when I look at the the
12:38growth that these companies have had in
12:40a short amount of time, I think there's
12:42really really solid I mean, it is a a
12:44remarkable growth rate that we've seen
12:46there.
12:47I don't know how successful these IPOs
12:48will be. It's going to be a question of
12:49valuation. It's going to be a question
12:51of market sentiment, too, right? Higher
12:53interest rates and macro headwinds are
12:55not helping the environment right now. I
12:57think over the long term these companies
12:58do appear as well positioned at at this
13:01point in time. Um
13:03and I will say, despite the competition
13:05from China,
13:07we're not too concerned that they're
13:08going to come and steal the lunch from
13:10the US model labs because much like
13:13we've all gone to appreciate the fact
13:16that we have a world's reserve currency,
13:18we may have a world's reserve model.
13:20Doesn't have to be one model. Could be a
13:22family of models. But, I do think the
13:25ability for these model labs with their
13:27closed architecture, with the trust
13:29that, you know, they are building with
13:31all of these different enterprises, for
13:33them to really give companies that, you
13:36know, that safety, that peace of mind,
13:39um, and the accountability around how
13:42the models are going to perform for
13:43these companies, that is going to be
13:45major. Yes, people are still going to be
13:47using Chinese models. They're not
13:49leaving the ecosystem that we operate
13:51in, but what matters most is where
13:53people decide to spend their money, and
13:55it still seems like US model labs have a
13:57moat on that end.
13:59>> One quick note, everyone, YouTube is
14:01telling us that 80% of you are not
14:03subscribed to the channel. If you want
14:05to help this channel grow and help me
14:07continue to book great guests, the
14:09single best thing you can do is hit that
14:11subscribe button. I really appreciate
14:12it. Now, let's get back to the episode.
14:16That is a super interesting idea that we
14:18could have a global reserve model. I
14:21have not heard anyone talk about that,
14:23so kudos to you.
14:24I have personally gotten a little more
14:27bearish on Anthropic and OpenAI as
14:30businesses. I think they have great
14:32products, but I don't know if they can
14:34sustain
14:35how much they charge for these things
14:37for, uh, everyday customers. Maybe for
14:40enterprises, I don't know, but a lot of
14:43my friends have really been weaning off
14:45Claude and ChatGPT because they're
14:47finding things like Muse or Instinct or
14:51GrokBot.
14:52And I think that's going to happen more
14:54and more because there's a lot of these
14:55more agentic tools, I think they're a
14:58lot more useful and they're a lot more
14:59fun to use, and it feels like you're
15:01actually overseeing a team of employees
15:04instead of using like a software
15:06dashboard. That That's how I feel. And
15:09even when I look at my own AI usage,
15:12GrokBot has become 90% of it, probably,
15:15and then Claude's become like 10%. So,
15:17that used to be 90% Claude, but I've
15:20totally switched since then. Um,
15:23do you think that the almost the rise of
15:26AI agents and the proliferation of their
15:28use, is that going to change any of the
15:30economics that you're talking about
15:32here?
15:33>> So, everything that you said is very
15:34fair. Um, I found myself switch my, you
15:36know, favorite models every couple
15:37months, too.
15:39But, what we're missing from that
15:40equation is the demand story. And the
15:43reality is, with agentic AI, the amount
15:46of tokens that we're going to be
15:48processing, the size of the overall pie
15:51of demand for these models is growing
15:53enormously. Agents can consume per open
15:56router 15 times as much tokens as a
15:59human using an AI model. That is a big
16:02gap, because
16:04how much are we growing our US workforce
16:06every year?
16:08Not that much, okay? We have aging
16:09demographics, we have tight immigration
16:12policies.
16:13How fast can we grow our agent
16:15workforce?
16:16Well, there is no upper limit to that.
16:17>> Our agents will have agents.
16:19>> Our agents will have agents and have
16:20agents and have agents, which is a crazy
16:21world to think about, which could come
16:23be coming very soon, right? Um,
16:25but I think that the fact that the
16:27overall pie is growing so much, yes,
16:29there is competition amongst these model
16:31labs. We think this is healthy, but we
16:33don't think that competition is eroding
16:35the economics for all of them, because
16:36the size of the overall pie is still
16:38growing.
16:39>> Okay, so help me understand this. If
16:42these agents are using a gazillion
16:44tokens, way more than humans are, and if
16:47we're getting more and more agents every
16:49year, let's say,
16:50what does that mean for a company
16:53like Anthropic or OpenAI, one of these
16:55frontier labs? How does that change
16:57their economics?
17:00>> Well, it all goes down to the total
17:01addressable market, right? It's it's the
17:04demand story. Um, and I think, you know,
17:08we'll have to see how these pricing
17:09models ultimately evolve. Um, do you
17:12want to charge per million token, or do
17:14you give more flat kind of subscription
17:17fees? I think we're very early in the
17:19race for the enterprise wallet. But that
17:21is what these companies are going to be
17:22competing for and I think that go takes
17:25us to a really interesting point here
17:27because when we think about all of this
17:30exciting progress on the technological
17:32frontier, right? We have a proliferation
17:34of agents. I think this year is very
17:35much the year of agentic AI, okay?
17:38Despite all of that and despite some
17:40super users, you know, I'm sure many of
17:42our listeners are are Kimberly. Um,
17:44adoption in the real economy is a lot
17:46more pedestrian.
17:48Um, you know, corporate adoption is very
17:50early and there's a lot of friction that
17:52ultimately comes with companies
17:54deploying these models and that's going
17:56to be the focus of a lot of companies
17:58involved in the enabling of AI
18:00infrastructure over the coming years.
18:02But KPMG does this great report every
18:04quarter and they find that barely one in
18:0610 large US companies have scaled agents
18:10over multiple business functions. So
18:12while companies are experimenting,
18:15they're increasingly paying and ramping
18:16up to pay for AI, the actual diffusion
18:19of that across a lot of different lines
18:21of work is very early. I mean, speaking
18:23from a very highly regulated financial
18:25institution, there is a lot of hurdles
18:28that we need to surpass in order to
18:31approve and uh test and then ultimately
18:34deploy and train our our workers to use
18:37these technologies.
18:38>> It's funny because I have a lot of
18:40friends that work in big companies and
18:43they tell me they can only use something
18:45like Microsoft Copilot, which
18:48I I've no idea what the numbers say. My
18:51anecdotal evidence is that is one of the
18:52worst AI tools to be on and it can't do
18:56any of the agentic stuff. It really
18:57doesn't connect to that much in your
18:59workflow.
19:00I feel very sorry for my friends that
19:01are in these big companies because they
19:03can't use something like a Grokbot or
19:06even a muse and they have to almost silo
19:08those things to their personal life
19:10because nothing they do at work can
19:12touch these new uh uh
19:14faster, better, smarter models, which is
19:16a big bummer.
19:17>> This is such an interesting and
19:19important disconnect that, you know,
19:21when you hear from these leading AI
19:23researchers from the model labs say, "AI
19:25is going to supercharge GDP growth by
19:2810%. It's resulting in a surge in
19:30productivity. It's going to, you know,
19:32take away all of these jobs." They're
19:34living in the in this bubble of they're
19:36at the frontier, okay? But, in the real
19:38world, adoption is again far more
19:42pedestrian. That's is good news, I
19:44think, in many ways. You know, I do
19:46think it is fascinating, you know, some
19:48data shows that roughly 80% of workers
19:50in the US economy bring their own AI
19:52tools to work. Okay? So, there's some
19:54evidence of that. The reality that
19:56corporate architectures do move slower
19:59than consumers are is really
20:00interesting.
20:01Um but, it also shows that there is a
20:03lot of room to go in terms of scaling
20:07these AI use cases and then ultimately
20:09translating them into productivity
20:11gains. For anyone looking at corporate
20:13America right now and saying, "Companies
20:15are spending all this on AI and what do
20:17they have to show for it?" Well, the
20:18reality is like yes, in the data,
20:2160% of S&P 500 firms are talking about
20:24AI in earnings calls, but just 2% can
20:27point to an actual tangible profit
20:30impact. That doesn't mean AI is failing.
20:33It means we're early. There's a lot of
20:35organizational and human friction that
20:38is involved in this AI rollout that we
20:40should not lose sight of.
20:42>> A lot of the more bearish folks I speak
20:44with, they point to
20:47productivity data in the at the macro
20:49level and they say,
20:51"Where are all the productivity gains
20:53that AI is supposed to be bringing?" And
20:56I see it in my own life, my friends'
20:58lives, and my friends that are
20:59entrepreneurs, they can do a million
21:01more things today than I could a couple
21:02years ago because of AI. But, a a of the
21:05productivity data is not really
21:07reflecting that. What do you make of
21:09that?
21:10>> Yeah, absolutely. Now, this has been a
21:12huge focus for me this year.
21:14Um
21:15we just released our 2027 long-term
21:18capital market assumptions. And in that,
21:21we did a complete upheaval around how we
21:24were measuring and forecasting the
21:26long-run contribution of AI to labor
21:28productivity. That's different than the
21:31short-term, and I'll get to that in a
21:32moment, too. But, over the long term,
21:34it's going to be a function of two
21:36things, okay? One, AI makes us a lot
21:38more efficient in all of the work that
21:40we do. We estimate that the effective
21:42penetration of AI, in terms of how many
21:45tasks that are are AI-assisted, is going
21:47to grow from roughly 5% today to 40% by
21:52the end of our horizon, which is 2040.
21:54So, a huge increase there in terms of
21:56the adoption of AI. That translate, in
21:59combination with all of this capital
22:00investment, boosting our productive
22:02capital stock, translates to a long-run
22:05forecast of 2.2%.
22:07Now, that might not be as extreme as
22:10some of the AI bulls might anticipate.
22:13There's room for some upside there.
22:15Um but, the reality is, that is a much
22:17better productivity growth rate than
22:18what we had before the pandemic, which
22:20was just 1 and 1/2%.
22:22Now, today,
22:24is there any evidence of AI boosting
22:26productivity?
22:27There is, but it is much more
22:29firm-specific or sector-specific than
22:32broad economy-wide. You can look at the
22:34technology sector. Tech has been a
22:36leading sector when it comes to
22:38productivity for a variety of reasons.
22:41Um but, also the fact that that sector
22:42is very involved in the deployment of
22:44AI, and they tend to be a lot more savvy
22:46at using that technology they've built,
22:48too.
22:49We do see scope for that to widen and
22:51broaden. But, we also have to remember
22:53that this technology is going to
22:55surprise us. You know, the internet was
22:58enormously productivity-enhancing, but
23:00how many of us have to be very careful
23:02with how much time we waste on our
23:04screens, right? Um with AI, there's been
23:08a lot of work that AI has created, and
23:11I'm sure you can relate. Overseeing,
23:13orchestrating, and verifying AI is no
23:17easy task. It's very mentally
23:20exhausting, I will say. And it eats up
23:22eats up a lot of hours. So when you look
23:24at
23:25even just the task level productivity
23:27gains, they might not be as extreme as
23:31what you'd expect because over the
23:33course of a workflow, there's all of
23:34this other stuff that AI generation has
23:37created. I'll also share that there's
23:40this fascinating work that has come out
23:42of the NBER around what software
23:44engineers have experienced as they've
23:46been adopting increasingly powerful AI
23:48technologies. And you'll find that the
23:51productivity gains in terms of the
23:53amount of code generated, remarkable.
23:55Okay, nearly 1,000% increase in code
23:58generated. But is code generated
24:00actually going to the bottom line?
24:02No. When it comes down to the releases
24:05that are shipped to users, which is the
24:06more, you know, reliable source of
24:08productivity impact, that is up 20%. 20%
24:12is still remarkable, but what's really
24:13interesting is despite newer
24:15advancements from Copilot to coding
24:18agents to agentic coding swarms, right?
24:21That 20% didn't change much because at
24:24the end of the day, it's a human it has
24:26accountability when it comes to deciding
24:28what gets passed and that is where a lot
24:30of the friction still exists.
24:32>> You know, I have a buddy who used to
24:34work at Meta, and he told me that
24:37sometime last year, they essentially had
24:39to hit uh token quotas of using uh AI
24:44tools to code a certain amount of
24:46content or or I'm not a coder, I don't
24:48know exactly the words, but they had to
24:50code a certain amount each day using
24:52tokens. And he said eventually uh he
24:55would essentially oversee a few
24:57different programs going at the same
24:59time and he never wrote code again. So
25:01he completely stopped writing code for
25:03the rest of his time in meta, but he
25:05also said that he didn't feel like he
25:07was doing that much more than before.
25:09Like it just kind of felt like okay,
25:11because we have a quota to hit and we
25:13have to hit this bar on a weekly basis.
25:16I'm just going to come up with things to
25:17work on.
25:18And that maybe is where some of the
25:20productivity gains are getting sucked
25:22into and it's not super
25:25effective use of these tools here.
25:27>> Yeah, it is really interesting. I do
25:29think AI gives us the ability to try a
25:32lot of things out at once that we
25:34wouldn't have tried otherwise. I will
25:36say even in my own personal, you know,
25:37research this year it's enabled me to
25:40achieve a breath of research that I
25:42would have never been able to do so just
25:44on my own and a depth within that
25:47breath, right? Because of the ability to
25:49lean on AI.
25:50But with that said, I've run into plenty
25:52of my own frustrations or issues when it
25:55comes to really managing these
25:57hallucinations remain really quite
25:59challenging across models.
26:01And it takes a keen skilled eye to test
26:04all these, right? And reading through AI
26:06output, like I said, that can take a
26:08while and it can be exhausting.
26:10And I think the question is, you know,
26:12it's not that AI isn't contributing to
26:14maybe a higher quality of what the work
26:16that we're doing, but the way that we're
26:17measuring AI's impact is much more than
26:20the bottom line and it's a lot it's very
26:22difficult to track even across corporate
26:23America right now.
26:25>> I yeah, I think that's a it's an amazing
26:27point cuz I also work with a lot of
26:28different AI tools and with Silvia
26:31insights we're publishing AI investment
26:33research. So that is something that I
26:35look at very very closely every day and
26:37to make sure that hey, all our ducks in
26:38a row before we publish this to our
26:40readers.
26:41Um with all of this
26:44that we're talking about across the AI
26:45ecosystem, all these new tools and new
26:47use cases.
26:49If I'm sitting here as an investor,
26:52where should I be putting my capital to
26:53work as far as what you see as the best
26:55opportunities in the market based on
26:57everything we're talking about?
26:59>> So,
27:00I'm going to first take that a slightly
27:01different direction.
27:02And when it comes to how do investors
27:06invest today, I think it's very
27:08important to ensure look at your
27:09portfolio and see how much AI exposure
27:13do I have and how much exposure do I
27:15have to certain AI scenarios.
27:17This is the foundation of the work that
27:19we did in our paper AI conscious asset
27:22allocation. Is, you know, there is this
27:25reality that most portfolios were built
27:28without the assumption that AI was going
27:30to be this risk factor that permeates
27:32through all of these different asset
27:33classes, right? As the AI build out has
27:36continued to expand, it's drawing upon
27:38external capital, it's hitting broad
27:40markets, it's hitting alternative
27:42markets. Um, you know, it it's been
27:45more difficult for investors to really
27:46understand where am I currently exposed.
27:50You know, where do I already have, you
27:51know, full exposure versus where do I
27:53want to lean into more, right? So, if I
27:57would start, you know, we look at this
27:58AI risk spectrum. Higher risk tends to
28:02be associated with higher reward. And
28:03you have seen in semis,
28:06uh, large cap public equities, and, you
28:09know, private stakes in in these, you
28:11know, massive AI behemoths,
28:13have also been, you know, really quite
28:15beneficial for investors that have had
28:17access to that over the last few years.
28:19Going forward, you need to make sure
28:21that you're aware of the exposure that
28:23you have with those concentrated, uh,
28:25positions to things going a little bit
28:28wrong, right? To these companies missing
28:30certain expectations, evaluations being
28:32too lofty, whatever it may be.
28:34Now, in this middle area, this is what I
28:36think is quite exciting, okay? Within
28:39alternatives in particular,
28:41you can invest in more diversified, but
28:44still exposed to the AI theme vehicles,
28:47and private infrastructure, um and
28:49private real estate to a degree,
28:51um and even in just broader
28:53infrastructure assets in US public
28:56equity markets. So, thinking about
28:58utilities, thinking about any company
29:00that is supplying really critical
29:02infrastructure like networking or
29:03cooling equipment to this AI build-out.
29:05The reality is the infrastructure
29:08beneficiaries still seem really well
29:11positioned to monetize AI in the near
29:14term, and
29:15they can operate in slightly different
29:17wrappers depending on how you're
29:18investing in it in portfolios. So, we
29:20think there's a lot to do there, but
29:22manager selection very, very key,
29:24especially in the alternatives world.
29:27On the less risky side of the spectrum,
29:30I would really point to fixed income. I
29:33think fixed income looks really
29:34attractive these days. And I will say,
29:36look, as a younger investor, too,
29:37personally, I haven't had a lot of bonds
29:39in my portfolio, right? Especially in an
29:42AI era.
29:43But, I'm changing my mind, and I think
29:45investors more broadly are, too, because
29:48of where income levels are right now,
29:50and even though companies are issuing
29:52debt to fund this data center build-out,
29:54generally speaking, this is very
29:56high-quality debt, and even if you're a
29:59little, you know, nervous about the data
30:01center build-out, you are far more
30:03assurance that you uh greater likelihood
30:05of getting your money back as a bond
30:07investor than an equity investor in some
30:09of these names, right? So, we think the
30:11risk-reward really looks quite
30:12compelling, and that can be good and
30:14will [clears throat] stirring overall
30:15total return, but also diversification
30:18in portfolios.
30:20>> First of all, uh
30:21I'm with you that I'm not one to buy
30:24bonds or fixed income at this age. I
30:27don't think I've ever owned any of those
30:28things uh in any
30:31generational bull market or not. That's
30:32one thing. Uh second,
30:35everything you're saying, is there a way
30:37to boil that down into something like a
30:39favorite corner of the market or a
30:41favorite sector?
30:43>> So, I will say it's going to be a lot
30:45harder to going forward, but um I think
30:48technology still shines bright. I think
30:50you have to be buckled up for
30:51volatility, and I think there's going to
30:53be a lot more dispersion underneath the
30:55hood. Like what we're seeing between the
30:56neo clouds versus the hyperscalers, and
30:58even in semis, okay? Semis remains
31:01really well positioned, you know, in
31:03terms of the visibility that they have
31:04to future demand. Some of these players
31:07have very significant moats. Sem-
31:11Advanced chips are some of the most
31:13sophisticated, impressive machinery that
31:16humankind has ever created. But not all
31:18moats are created equal. What an
31:21advanced chip fab in Taiwan has versus a
31:24memory maker, very different in terms of
31:26the sustainability of those margins that
31:29that we expect. So, in our LTCMA's, for
31:31instance, when we forecast what an
31:33equity investor would anticipate to
31:35achieve in a market like Korea over the
31:38next 10 to 15 years, it's negative,
31:40okay? Because of the starting points
31:43that don't seem sustainable over the
31:44long term. Some of these markets are in
31:46far more cyclical. Others do seem to
31:49have really structural staying power.
31:51So, that's infrastructure and that
31:53piece. Beyond that, we do think there
31:56are opportunities emerging in adopters,
31:59but it's going to be a single stock
32:00story here. I'm very excited about
32:02healthcare over the long term. Health
32:04tech is just one of the most exciting
32:06things. I'm a
32:08an Oura Ring user, and just thinking
32:10about, you know, the kind of technology
32:13and the analytics that we're going to
32:14have on our own well-being going forward
32:18is, you know, wow. Really exciting. Um
32:21but it is a lot harder to price that
32:23into earnings growth in the near term
32:26for some of these companies. I think
32:28it's really going to some some firms
32:30that have an established advantage in
32:32terms of massive proprietary health data
32:35seem very well equipped, and
32:38particularly in highly regulated
32:39industries, those that can use that to
32:42their advantage also seem well
32:44positioned. So, within health care,
32:46within financials, I think there's
32:47opportunities amongst those adopters.
32:49And then finally, I will say software
32:51has been such an interesting story over
32:53the course of this year. We've gone from
32:55SaaS-pocalypse
32:57to actually a few of these names now
33:01establishing themselves as critical
33:03enablers to corporate adoption. And I
33:06think we're in the very early innings of
33:08that. You know, software is one example
33:10of an industry that first people yelled
33:12destruction, and now we're realizing
33:15disruption can actually breed new
33:17winners. Incumbents can reinvent
33:18themselves, or new players can come in
33:21and ultimately disrupt the marketplace.
33:23And I do think there's a lot of exciting
33:25growth stories there when it comes to
33:28orchestrating agentic AI and helping
33:31secure corporate defenses in
33:32cybersecurity.
33:34>> Okay, I'm going to try to
33:36remember and say these back to you.
33:39Okay, I think we heard AI
33:40infrastructure,
33:42uh software,
33:44industrials, is that
33:45>> As a part of AI infrastructure, yeah.
33:48>> Okay.
33:48>> I would bucket that there.
33:49>> Um and did I say semiconductors already?
33:52Is that So, semiconductors, software,
33:53industrials, infrastructure,
33:55uh financials,
33:57health care.
33:58>> Mhm.
33:59>> That's it That sounds like a ton of
34:00opportunity.
34:00>> There are There's a ton of
34:01opportunities. But again, a lot of
34:03dispersion underneath the hood. Right?
34:05Stock selection very important in this
34:07market. We think active managers are
34:08well positioned to take advantage of
34:10that.
34:11Um but that is really going to be key.
34:13And then again, just making sure your
34:14portfolio overall is built to withstand
34:17these periods of volatility because
34:18we're seeing technical factors at play,
34:21sentiment very much in the driver's seat
34:22of markets these days.
34:24>> How do you think about separating
34:26between
34:27US-focused stocks versus non-US?
34:31>> Mhm.
34:32Well, we've talked a lot about the US
34:34markets, but we cannot be remiss to
34:36mention that international has
34:38outperformed the US for now 2 years
34:40running, pretty meaningfully so.
34:43Now, internationally, there is a lot of
34:46compelling reasons why I think investors
34:47should be looking to some of these
34:49markets. On the one hand, the AI
34:51infrastructure boom has gone global. We
34:53are portfolio managers estimate that for
34:55every $1 of hyperscaler CapEx in the US,
34:59$0.60 of that goes to Asia. All of these
35:02different companies integral to that
35:04tech buildout, okay? We're super, super
35:06key there. Beyond that, for investors
35:08trying to diversify a little bit from
35:10this AI theme, we think Europe and Japan
35:13scream really attractive. So, in Europe,
35:16you have other themes going on here, you
35:18know, those economies are now investing
35:21a lot more in defense. They're also
35:23investing a lot more in energy. Some of
35:25that's AI related, other part of that
35:27is, you know, there's been a whole
35:28different geopolitical world that we
35:30live in, com- countries are making sure
35:32that they have some degree of resiliency
35:34around these geopolitical shocks. So,
35:37that is bolstering sectors in those
35:39markets. We think value, if you're a
35:41value, if you want to add some value in
35:42your portfolio, international value is
35:45the place to look.
35:47Beyond that,
35:49I would say Asia, Europe,
35:52um
35:52elsewhere in emerging markets, it's more
35:54challenging, you know, there aren't the
35:56same kind of structural themes at play
35:58here. So, I would balance that Europe,
36:01Japan, and kind of Asia AI story with
36:04again, just a really strong and
36:05diversified US side of your portfolio.
36:08>> So, I I just spoke to an investor this
36:10week who
36:12is super bullish on Latin America, which
36:14I think is a really interesting story,
36:15too. I've been in this one ETF called I
36:18think it's the ticker is IOLF, and it's
36:21just a Latin America, I think it's an
36:22iShares ETF. Anyway, it's up like 30%
36:25this year, maybe 28%. Totally crushed
36:28the US benchmark. That's been pretty
36:30good. I think that's a really
36:32interesting international story.
36:34South Korea, I have kind of taken a
36:36beating on beating on the last month,
36:39but I've been in in this EWI ticker, the
36:43South Korea ETF for like 2 years almost.
36:45So, it's had an amazing run, but now I'm
36:47getting a little concerned and you also
36:49just said your team doesn't think South
36:51Korea has
36:52that optimistic of future. Did I get
36:55that right or
36:55>> Yes, but I will say
36:57I was looking at a very long-term
36:59horizon. In the near term, I mean, you
37:00know, the bottlenecks
37:02that persist across memory are really
37:04significant and we are in a market where
37:07the demand is just growing as we've
37:09said, you know, magnitudes on
37:11magnitudes. So, you know, these
37:13companies do still seem well positioned
37:15in the near term to monetize a lot of
37:17that demand. It is also notable that
37:19valuations in international markets and
37:21even in those Asian AI beneficiaries are
37:23not all that demanding. Okay, I think
37:25the South Korean forward PE is something
37:27like four to five times compared to the
37:30US market around 19 times. So, you know,
37:33these are high momentum, highly
37:35concentrated markets,
37:37but the reset that we saw really over
37:40the course of the summer around these
37:42high momentum semis names has left us in
37:44we think like healthier positions.
37:46>> Do you think that buying into, let's
37:49say, the AI trade today,
37:51if you had zero exposure, is it
37:54not So, not too late then? Cuz a lot of
37:56people still look at the market and they
37:57say valuations are too frothy,
38:01another dot com, whatever the fears may
38:03be,
38:04but it sounds like if we're still early,
38:07you could still buy in at current
38:08levels, let's say.
38:09>> Yeah. And I would say, look, market
38:12timing, we we always say from our
38:14business it is very tough. It is very
38:16tough. But, you have we have to anchor
38:18ourselves in how much real earnings
38:21power is associated to this AI boom?
38:24We're not investing on hopes and dreams
38:26here, okay? This year the market is
38:28expecting to see a 30 plus percent
38:30annual earnings growth, which is
38:32remarkable. We're talking about record
38:34high profit margins, and over the next 2
38:36years we're also looking at consensus
38:39mid double-digit earnings expectations,
38:42which we think are probably the low
38:43water mark on what we could anticipate
38:45to see. AI infrastructure this build out
38:47is still driving a lot of earnings power
38:49across the market. So even in a year
38:52where you have seen the forward PE on
38:55the market contract by four times, we're
38:58still near all-time highs. We've had
38:59this war in the Middle East, we've had
39:01rising rates, right? We've had this a
39:04significant backup in long-term yields,
39:06and yet we're where we are today because
39:09of that earnings power behind us. So
39:11when you say is it a good time to
39:12invest? I think so because the outlook
39:14for AI driving productivity, economic
39:17growth, and ultimately sustainability in
39:19that earnings growth still shines
39:21bright. You just want to make sure
39:22you're prepared for the bumps along the
39:24way.
39:25>> What I think one of the biggest stories
39:27this year has been like this Teflon
39:28market, right? It can totally ignore and
39:31survive any of the geopolitics, any of
39:33the macro headwinds, and you know, bond
39:36yields pretty much did nothing to the AI
39:38trade. My view is that
39:40any of the macro data that people are
39:41worried about is pretty much overwhelmed
39:44by the strength of the AI story.
39:47Is there any scenario you see where
39:50something overtakes the AI story, like
39:53whether it's geopolitics or maybe
39:54there's macro data or the Fed? I don't
39:57know. Does any of that concern you?
39:59>> Yeah.
40:00You know, it's a common saying how like
40:02the stock market is not the economy, but
40:04I always like to think about, you know,
40:05the stock market it can be having a
40:07grand old time, a grand party, but if
40:09someone takes the rug out from under
40:11you, like party's over. That's very much
40:13how I think about the the macroeconomy.
40:15And if something were to take the
40:17ground, you know, the carpet out from
40:19under us,
40:20um things would get very difficult here
40:23because so much of this CapEx boom is
40:26ultimately under the assumption that
40:27demand is going to continue to grow at
40:29these really robust levels. And the
40:32reality is the recessions of the past
40:35have all been the kind that have been
40:36impossible to forecast. We don't know
40:39what we don't know. And there is that
40:42reality that, you know, you you do want
40:44to make sure as an investor that you're
40:45prepared for some of those outcomes. You
40:48have that insurance, especially in an
40:49environment where you're getting paid to
40:51own that insurance. So, that's how I
40:53would frame it.
40:55>> I think that's uh
40:56very diplomatic and very uh safe to put
41:00it in those terms. Steph, where can we
41:02send people to find more of your work?
41:05>> Um yes, so on our website,
41:07jpmorgan.com/aihub
41:11has all of our insights around
41:13artificial intelligence. We have a
41:15presentation deck, and we have links to
41:17a lot of the really great research that
41:18I've been talking about today. So, our
41:20white paper from our team called SIAG,
41:22AI conscious asset allocation, is on
41:25that landing page, as well as our newly
41:27released long-term capital market
41:29assumptions. And then, of course, I'm
41:31always sharing my thoughts on LinkedIn
41:33as well, so you can give me a follow
41:34there.
41:35>> You are a super LinkedIn influencer, so
41:38everyone should definitely go find you
41:39there. Thank you very much for your
41:40time, and we'll do this again soon.
41:41>> Thanks so much.

About this transcript

This page contains the full transcript of Top JPMorgan Strategist: 6 sectors to BUY for the NEXT AI boom! by Phil Rosen, generated from the public captions YouTube serves with the video. The transcript has 7,469 words across 1,166 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.