By Alex Eule

The last time Barron's convened a Tech Roundtable -- in August 2023 -- generative artificial intelligence was in its honeymoon phase. ChatGPT was less than a year old. The Magnificent Seven were still just characters from a 1960s Western. And Nvidia carried a third of its current $3 trillion market value. Looking back at that moment, it's clear investors managed to both under- and overestimate how AI would change the marketplace.

Some 18 months later, AI has yet to cure cancer or drive a car cross-country. But it has upended corporate strategy and spending plans, and opened the door to an age of rapid problem solving, enhanced productivity, and machine-driven creativity.

Meanwhile, the AI hype cycle has entered a new phase, with investors looking for a payoff. As some of the early enthusiasm fades, tech stocks have entered a correction. So, where do we go from here? We put that question to four veteran stockpickers focused on tech: Felise Agranoff, portfolio manager for multiple tech-focused J.P. Morgan Asset Management funds, including the JPMorgan Active Growth exchange-traded fund and JPMorgan Equity Focus ETF; Gavin Baker, chief investment officer at Atreides Management; Tony Kim, portfolio manager and head of the global technology team in the fundamental equities division at BlackRock; and Denny Fish, portfolio manager at Janus Henderson and head of the firm's technology research.

We spoke via Zoom in the middle of March; our panelists joined from around the world. Here's an edited transcript of our conversation.

      Barron's:   The market has fallen sharply over the past month or so. Is the current selloff a buying opportunity or an opportunity to take some profits in tech and wait for a brighter day? 

Denny Fish: If we reflect on the past couple of years, coming out of 2022 and looking into 2023-24, tech [stocks] were incredibly strong, doubling during that period. The reality is, it isn't unusual to have drawdowns of this magnitude. It's natural. It's part of the cycle. Looking back, the past couple of years were a period when valuations looked unreasonable. Valuations are now starting to look more attractive. Across a number of sectors, it gives you an opportunity to lean into your best names for the next three to five years.

Many stocks are now trading at more attractive valuations, for two reasons. One is that the growth in some cases has been better than expected, so that has compressed valuation multiples. Second, the market prices have come down, and that makes things look more attractive on a price-to-earnings or price-to-free-cash-flow basis. Selectively, it is a good opportunity.

Does anyone disagree with this?

Felise Agranoff: We had some concerns over the past year, in particular, that AI capital spending was getting as good as it gets. It has been very strong, and that has resulted in substantial upside for many of the companies exposed.

We were very bullish on Nvidia and AI capital spending at the end of 2022 and early 2023. We're still bullish on the long-term prospects for AI and the impact it is going to have, not just on the tech industry but all industries. However, we see more risk that there could be a lack of upside to capital spending in the intermediate term. We need to digest the massive growth in capital spending that we have seen. Given some of these concerns, we have been paring back on the names in our funds that are directly exposed to AI capital spending.

Gavin Baker: I would take the other side. DeepSeek [the low-cost Chinese AI model] set off all this selling. For whatever reason, the stock market reaction to Deep Seek R1 came seven days after the publication of a technical paper about it.

Spot-rental availability of graphics processing units is down precipitously after the release of DeepSeek. It is much harder to rent GPUs today than it was before DeepSeek. Meanwhile, GPU spot-rental pricing is up since DeepSeek, especially for H200s [Nvidia's older-generation AI chip]. Google is raising contract GPU rental pricing, as are other hyperscalers. DRAM [dynamic random-access memory] spot prices are going up every day.

Taiwan Semiconductor Manufacturing's February revenue growth -- it reports monthly numbers -- accelerated versus January. And why are all these tip-of-the-spear indicators of GPU demand accelerating? It's because reasoning models, like DeepSeek R1 and Grok 3, consume vastly more compute than the kinds of models the world was using before them. The world got a lot compute-hungrier.

Everyone is going to spend up on Blackwell, Nvidia's next-generation GPU, because it's too big a risk not to. This is a classic prisoner's dilemma. If everyone cooperated, maybe they would all like to slow down spending. But if just one company doesn't cooperate in the great game played by Google, Meta Platforms, Microsoft, OpenAI, or xAI, that company might get a decisive advantage, and some of these companies believe that could be existential. So, they are all going to spend on Blackwell. Nvidia is trading for 22 times consensus earnings, which isn't expensive. There is a strong year of spending ahead, and companies need to see a return on investment, or ROI, from it.

How long will that spending last?

Baker: It depends how good the return is, but I think it is going to be really strong in the second half of this year, in particular. If the next generation of models is significantly better and companies are able to charge thousands of dollars a month -- instead of $20, $30, $40, or $200 -- for an AI agent that can, in many cases, do the work of humans, spending could go on for a while. And if this world doesn't materialize after Blackwell, you'll for sure see a pause.

Tony, what is your take on the recent selloff in tech stocks?

Tony Kim: This is a natural drawdown. The wild card is whether there will be a recession. If so, the selloff is justified and you will need to recalibrate.

But people aren't buying the AI trade just for this year. They're looking at it continuing in 2026 and '27. There are questions about that duration, which is contingent on ROI. As we move into the second phase of AI -- I call it moving up the stack -- a lot of companies could participate in generating that ROI.

Fish: Consider the number of software engineers and support personnel at Nvidia, Microsoft, or Google. These are key first-use cases for AI to drive productivity. The operating leverage we are likely to see, excluding depreciation from capital spending, is likely to surprise on the upside at most of these companies in the next couple of years before it bleeds into the broader economy more pervasively.

Let's take a closer look at what kinds of benefits we're going to start seeing from AI.

Agranoff: We are seeing benefits from the use of AI agents [programs or systems that can perform tasks without human intervention]. The big opportunity in the market is around companies in other sectors that are applying AI to improve customer outcomes.

To the extent that agents are right 50% of the time, it will take some time for adoption to reach an inflection point. But there are companies with vertical expertise and specific use cases that can implement agents with high levels of accuracy.

We met with Intuit, which talked about applications with agent accuracy rates of more than 90%. This is really value added. But success won't be broad-based; there will be a lot of differentiation among companies. It will be a great environment for stockpickers.

How fast will agent adoption be? When will we see the benefit of AI agents replacing our current tasks?

Baker: Everyone has to first deploy Blackwell. We have to train these new models. Then we have to see whether these models, working together, really do a lot of the work that an average human does. If they do, there will be a positive ROI in AI. If they don't, then 2026 is going to be a very different year than 2024 or '25. That goes to Tony's point about duration. There are some unknowns in 2026 based on cards that we haven't turned over.

How worried should people be about the threat of AI to their jobs?

Kim: Of the roughly $100 trillion of global gross domestic product, 55% is labor and 5% is tech spending. Even if you took just 5% out of labor, you would double all of tech spend. If that were to happen, there are two big implications: You wouldn't have enough compute in the world to do this, which should limit its progression. And there are societal implications.

Where is the net new revenue or growth in gross domestic product, versus the cost takeout? That's a bigger question mark for me. Everyone is talking about productivity on the cost side. But what about creating net new GDP? It isn't obvious to me. But if it happens, there isn't enough compute in the world.

So, the good news is that AI can't take over the world because there aren't enough chips yet. Returning to stocks, is the AI trade focused just on the so-called Magnificent Seven companies? And, how important is the health of AI to the broader market this year?

Agranoff: A whole host of companies has been correlated with the AI trade as people thought about the second-tier and third-tier impacts of this huge buildout in generative AI and the associated infrastructure to support it. As the magnitude of capital spending slows, it will be hard for these stocks to work in a meaningful way. It will be hard for AI stocks to be market leaders.

However, I see investment opportunity in the industrial sector among companies with option value in AI. For example, we own companies exposed to significant investment in the utility grid, which will support high-powered data centers. We are only in the very early stages of seeing the benefits.

So, there are opportunities within that broader AI trade. But as for the leaders of the AI trade? That debate will continue into 2026.

Fish: There is a reasonable case that the AI trade does lead the market, but in a different way than it has. Generally when a theme takes hold -- particularly a theme that is a decade-plus in the making and is probably the most profound shift we have seen in our lifetime -- a rising tide will lift all boats for a period of time. Then, as time goes on, the winners start to separate themselves.

Kim: That is exactly right. In the early days, a rising tide lifts all boats. The correlation is high. But over time, there is separation. There could be many different categories. But power laws will play out, and you will see the emergence of winners in these categories. Right now, things aren't so obvious. But I suspect that in 2026, '27, '28, that kind of Darwinian power-law concept will start to manifest as these companies start to separate more.

Agranoff: That is something we have been waiting for -- for a while -- because leadership has been narrow and there has been a high correlation among many of the winners.

When you see something as big as AI, whose impact I think we would all agree is in its early innings, it is unlikely that the winners are going to be the prior winners or the biggest companies in the market today.

Baker: Only 1% of current global internet market cap had been founded two or 2 1/2 years after the introduction of Netscape Navigator. It goes to Felise's point that we are very early. At this equivalent point in the internet cycle, Mark Zuckerberg was in middle school; so many of the iconic companies hadn't even been conceptualized, much less founded.

I can't help but note that we don't use Netscape anymore.

Baker: What is interesting about AI is the raw materials: compute and data. Today's biggest companies have the most compute and the most data. AI is existential for all of them, and that's what is creating this prisoner's dilemma. But these companies are all in on winning it, and they have the raw ingredients of compute, data, and distribution to win -- though the history [of technology incumbents] also suggests it is going to be tough.

Kim: That's spot on, Gavin. I'll agree with you again. The difference with Internet 1.0, Cloud 1.0, and AI is this very notion of the raw materials being super capital intensive, an intensity we have never seen before in the history of technology.

Agranoff: Also, the difference in this cycle versus past tech cycles is that the companies that are spending have massive amounts of cash flow. Scale really matters, as does your ability to spend. And, to Tony's point, there are very few companies that can compete against what the big tech companies are spending today -- and they still aren't spending 100% of their cash flow.

OK, let's get to some stocks. Felise, which companies will be the future winners in tech? Which companies are you betting on?

Agranoff: When we think about the beneficiaries of AI, we are interested in companies applying AI or benefiting from the infrastructure requirements needed to support AI. I would bucket the beneficiaries in a few different areas.

The first thing you need to do to get your organization ready to implement AI is to make sure to get your infrastructure and data intact. We have focused on companies that could win within the infrastructure layer. Snowflake is a name we like. It is well positioned to take advantage of the work that needs to be done in the integration side.

Cybersecurity risk is one of the key constraints and governors on the pace at which AI will be rolled out. We have also focused on cybersecurity companies that will benefit. Palo Alto Networks is an example.

Then, on the application layer, we have been focused on companies with a credible strategy to deploy AI agents. That includes Intuit, which I mentioned earlier, and HubSpot. Atlassian is another example of a company well positioned in software to benefit from implementing AI features.

AI is going to impact every single industry sector. One of the things that is underappreciated is that we will see positive impacts in areas outside of technology.

Which areas or companies are you most focused on?

Agranoff: Among the beneficiaries I would highlight is the travel company Booking Holdings. Although there is a debate about travel demand today, Booking is in a great position to put in place a more virtual experience to book your travel.

Within healthcare, we like companies such as Intuitive Surgical, the leader in robotic surgery. The company just came out with its next-generation robot, and it is at the early stages of fully utilizing the data that are gathered during surgery to improve surgical outcomes, which will result in a substantial inflection of growth in the business.

Then there are companies like Netflix that we think could benefit over time. Netflix is already a high-margin company. Netflix won in the streaming market. Now it could use its economies of scale and market leadership position to deploy AI to improve content costs, as well. We think that could result in a further inflection in profitability.

Last, there are some companies that are well exposed to infrastructure investments within the industrial sector that could benefit from AI. We believe AI is one additional catalyst for power-demand growth in the U.S. after decades of power demand barely growing. We like companies such as Quanta Services, a leader in transmission and distribution infrastructure, because the grid has been underinvested in for decades. We think the market is massively underestimating the investment cycle there, and Quanta's leadership position.

That's a good list of companies. Denny, care to go next?

Fish: Sure. I second Intuit. That was one of my names.

The way I think about investing in the AI theme is finding companies with really durable business levels -- so, companies that I would be excited about even if AI wasn't here. AI enhances the story.

One of the most important vertical software industries on the planet is electronic design and automation, called EDA. Cadence Design Systems and Synopsys form a global duopoly. Their addressable market has expanded materially over time and will likely continue to do so, due to [their role with] hyperscalers and vertical industries such as automotive, aerospace, and defense. AI enhances the value proposition in terms of speed to market for design, one of the most complex processes on Earth.

You can now get both of these stocks at the low end of their price/earnings multiple range over the past several years. Cadence has traded below 30 times forward earnings only a couple of times, and one of those was the big drawdown during Covid. It is an interesting time for that space. We like the outlook over the next three to five years.

What else do you like?

Fish: If you're a big believer in anything that we're talking about, we're just going to need more compute, and KLA is a natural beneficiary of that. It is a provider of semiconductor capital equipment in what I consider one of the most interesting swim lanes -- process control, optical inspection, and materials. The space has grown over time, and KLA has been a share gainer. We expect that to continue. KLA has the highest gross margins in the semi capital-equipment industry.

My third idea is MercadoLibre. We have invested in the company for years. It has created the most unique flywheel, from e-commerce to payments to credit to connecting consumers and merchants, and taking all that data and layering on complex logistics throughout Latin America. The Latin America region has one of the lowest e-commerce penetrations. There is a lot of room to run there.

E-commerce totaled about $150 billion in Latin America last year and is expected to grow by 50% over the next four years. Mercado has continued to be a share gainer because of this flywheel. If it continues to integrate AI deeper into its fintech platform, I suspect those credit metrics get better, as does the growth. We think the business is a long-duration grower. It is trading for about 30 times next year's earnings, and that's for a 30%-plus potential grower over the next two years.

Tony, what are your tech picks?

Kim: Felise, I agree with you on Snowflake. I agree as well on Atlassian. Denny, we have been a long-term supporter of Cadence and Mercado, so there is some overlap.

Systematically, I'm thinking through moving up the [tech] stack and the data stack. An architecture called RAG [retrieval-augmented generation] enhances text generation by integrating proprietary database information. Elastic, based in the Netherlands, is a AI-search company whose search engines sit atop the RAG vector database.

CyberArk Software is another name I'll mention. We are moving into an agentic world, or one that uses AI agents and machines more than humans, and more synthetic data than human-generated data. CyberArk, based in Israel, is well positioned in machine-to-machine identity security.

Next, I'll mention a U.S. company that is opportunistically priced because of the market selloff: Reddit. It has been unduly punished. It has a unique position as the nonsocial social network. Reddit is a bastion of human-curated content. It is going to do well as a result of AI search.

Anything else you'd like to share?

Kim: I'll bring up a few more international names that might be of interest. One is Harmonic Drive Systems in Japan. It is the world's leading gear reducer for actuators for robotics. If and when humanoid [robots] come, Harmonic is the world's leading non-Chinese component provider, and will be an integral part of the supply chain of humanoids.

Another company we like is eMemory Technology, in Taiwan. Think of it as a version of an IP [intellectual property] company selling to Taiwan Semiconductor and other foundries. But it has something new coming, called post-quantum cryptography. When quantum computers arrive, they will be big enough to break encryption. You will then need to remediate and upgrade all the encryption in the world. And eMemory has an IP standard for post-quantum encryption that I imagine will be on every chip. Every motherboard will need to have the immunization of a new encryption standard.

Next, I am torn between Alibaba Group Holding and SoftBank Group. Alibaba is the hyperscale model provider of China. With SoftBank, you are getting an investment in OpenAI, Stargate, the Vision Fund, and Arm Holdings. And Arm will play a role in Stargate, as well. SoftBank trades at a 60% discount to net asset value.

Let's include both. And let's get Gavin's picks before we bring this conversation to a close.

Baker: I really like Felise's pitch on Snowflake. It resonates with me. Snowflake has a new CEO, and the company is storming back with a vengeance. You can see that in a lot of the data. The CEO is Sridhar Ramaswamy from Google, and he was an exceptional executive there.

More broadly, my biggest fear from this panel is that everyone essentially agrees on everything. In my experience, when everyone at an event like this agrees, that isn't always the best thing. I have essentially agreed with every pitch, even if I don't own them all.

Astera Labs would be my first name. I referenced how the Blackwell upgrade cycle for Nvidia will be in full force the second half of this year. It is going to be a giant capex upgrade cycle for everyone, and everyone has to invest because of this "prisoner's dilemma" dynamic. But Blackwell isn't Nvidia's most important chip.

Nvidia's most important chip is NVSwitch. And NVSwitch 5, paired with Blackwell and some other chips that Nvidia makes, enables rack-scale computing. Picture an entire rack of GPUs -- something 7 or 8 feet high that weighs 6,000 pounds and consumes an unimaginable amount of power, something like the power consumption of 100 average American homes. NVSwitch enables the 72 GPUs to all act as one big computer, which has profound cost and performance advantages for AI.

Astera essentially enables rack-scale computing for everyone else. Astera underpins Amazon's rack-scale computing solution for Trainium [AI chips]. For now, it is the only real merchant alternative that can enable rack-scale computing for other Asics [or custom chips].

Astera is trading for 20 to 25 times our calendar 2026 earnings estimates, which are materially ahead of the Street's. It grew more than 200% last year. It will grow close to 100% this year. This is an important company -- and one we know well because we first invested when it was still private, in the series C [funding round], paying $7 a share in August 2021. We bought more in the series D in 2022. The stock came public at $36 a share. It now trades at $65, down from a recent high of more than $140. The stock is down more than 50% this year.

My next two names are both videogame companies. I expect AI to turbocharge videogame quality while also lowering the cost of creating these games.

That's a good combo. Which companies are you betting on?

Baker: The first is Roblox. After a big investment period, Roblox has gotten religion about profitability. It had 44% incremental Ebitda [earnings before interest, taxes, depreciation, and amortization] margins in calendar '24. It is doing a really good job on cost control. It is kind of back to the consistent 20%-plus top-line growth that it had before some turmoil related to a Covid hangover.

Roblox monetizes at six cents an hour; the average mobile game monetizes at 10 cents an hour. Your average console or PC game monetizes at 50 cents an hour, and some monetize up to a dollar an hour. So this is kind of the lowest-monetized videogame platform out there, with a lot of upside if Roblox just catches up to mobile games. That is nearly 60%-70% upside in their per-hour monetization. And Roblox is a platform where normal people can make games. You really get a benefit from AI making game creation easier. This is another name where we are significantly ahead of consensus on earnings and Ebitda.

My next name is Nintendo. First, you have a new product cycle -- the Switch 2. Nintendo learned its lesson with the Wii U [game console]. It tried to do too much and kind of fragmented its markets.

The Switch 2 is going to be a pretty compelling upgrade to the Switch, which is one of the best-selling consoles of all time. But more important, this is the first time in eight years they've had a new, upgraded product, which is the longest they've ever gone without a new product. It is reasonable to think there will be a lot of pent-up demand.

Kim: Over seven years, Gavin. Man, you and I should just hang out together more.

Baker: Yeah, I agree. But what is really exciting here is that the Nintendo Switch Online membership, NSO, charges $20 a year, versus PlayStation Plus at $80, and it has less than half the penetration. If Nintendo can get its online service to a PlayStation or Xbox level of pricing and attach -- and they are doing a lot to put value into it -- you could have well over 50% upside to consensus estimates. This is a really powerful product cycle.

Should investors buy the American depositary receipt or the Japanese stock, or doesn't it matter?

Baker: I don't think it matters.

Nobody mentioned Nvidia as a pick. Tell us why.

Agranoff: We still own Nvidia. Long term, Nvidia has a huge market leadership position in GPUs. However, it has a very large weighting in the benchmark that we're benchmarked against, and relative to our benchmark, we're underweight the stock today. It is really about, how much gross capital do you want to allocate to Nvidia at these [price] levels, given that we think it is at a kind of temporary peak level of capital spending, and the level of upside from here will be more muted for now.

Kim: I didn't think about talking about Magnificent Seven or Eight, because I presumed everybody already owned them.

Baker: I was trying to pick some names more off the beaten path. Nvidia is trading at 22 times earnings. There shouldn't be a debate on the multiple. What we can debate on Nvidia is the earnings. Once you get through Blackwell, you are going to have to see ROI on that spend for the spend to continue. But I think Nvidia sets up reasonably well for the second half of this year, while acknowledging that the uncertainty about '26 and '27 may keep a cap on the multiple.

We have never seen anything like this in semiconductors, where the revenue went up so much so quickly and the operating margins are so high. But at 22 times earnings, this isn't an expensive stock if earnings estimates are right. And they seem right for calendar '25, at least.

Kim: Broadcom is more expensive. Advanced Micro Devices is probably more expensive. Pretty much every company is more expensive than Nvidia on an absolute multiple. Starbucks has a higher multiple than Nvidia, and Costco trades at about twice Nvidia's multiple, or 50 times earnings.

Before we end, let's do a Mag Seven exercise. What is your favorite stock in this group, and what is your least favorite?

Agranoff: Can I say two?

In order?

Agranoff: No. 1 is Meta Platforms. No. 2 is Amazon.com. Least favorite would be Apple and Alphabet.

Baker: I would say Jensen [Nvidia CEO Jensen Huang] is one of the two most exceptional CEO founders in the world, right there with Elon [Musk, CEO of Tesla].

My least favorite is Apple. When Scott Forstall was the head of software, they rolled out Maps, and it was a disaster. He was forced to resign. Apple Intelligence has been even worse. It is buggy and doesn't work.

Apple's capex is a lot lower than its Mag Seven peers. Is that a bonus for them?

Agranoff: I think that is why the stock has done as well as it has. Similarly to Gavin, we think the innovation has been extraordinarily disappointing for Apple. However, at a time when all the other companies have been increasing their capital intensity, Apple has continued to generate strong incremental free cash flow production, despite having pretty low top-line growth. That has been a strength and is why it has worked. But without the innovation, we don't think it's going to get to the next level.

Kim: Meta is my favorite. They aren't being attacked on their core business like others, and they are net offense in new categories, either in glasses, humanoids, open source, and maybe even in search or other things. It isn't obvious what is going to replace Instagram. That is why I would go with Meta.

Fish: Keep in mind, these seven are some of the greatest companies that have ever been created in the history of mankind. And there is a base-case scenario that they all do well over a multiyear span. With that said, if we're talking about the next 12 months or so, I think Microsoft is No. 1. And if I were going to stack right down to No. 7, Tesla.

With Tesla, there is so much more uncertainty, even though the stock has come down a bunch. You have materially more electric-vehicle competition that the company has to navigate. Full self-driving: As a fairly longtime user of it, I love it, but it's still not there, like a Waymo. I don't know how quickly that transition is going to happen for Tesla on robo-taxis. Then there is a huge call option with Optimus, their robotics. I would never count Elon out. This just might not be his year. I reserve the right to change my mind at any time.

Microsoft is interesting because of all of the Mag Seven, it is the one that could actually take its foot off the gas in terms of capex and start letting the flywheel go a little bit more on free cash. And at the same time, it is starting to get more [AI] revenue contribution across the portfolio, at a time when others are still full steam ahead on model training.

Baker: On Tesla, for humanoids, it's them and the Chinese. And that is going to end up being a big market. When they ramp Optimus, we'll see, but I wouldn't bet against it. Their battery storage business has grown really fast, and is going to be an essential part of any clean-energy transition we're going to make. So, maybe I would take the other side of you on Tesla a little bit. But reasonable minds can disagree, and you are certainly a reasonable mind, Denny.

Fish: Over the multiyear, I don't disagree with you. But if we're talking about 2025, this just might not be their year.

We have four reasonable minds here. Thank you, all.

Write to Alex Eule at alex.eule@barrons.com

This content was created by Barron's, which is operated by Dow Jones & Co. Barron's is published independently from Dow Jones Newswires and The Wall Street Journal.

(END) Dow Jones Newswires

March 28, 2025 01:30 ET (05:30 GMT)

Copyright (c) 2025 Dow Jones & Company, Inc.