By Alex Eule
Artificial intelligence promises to remake economies, supercharge productivity, cure cancer, discover new drugs, and solve climate change. It is also said to be destroying jobs, privacy, profit margins, and the search for truth. One piece of data triggers optimism, followed just days, or hours, later by apocalyptic dread.
The battle is playing out in real-time on Wall Street. Nvidia, once the surefire way to invest in AI, has been dead money for six months. Software, formerly viewed as the perfect business model, is now seen as a relic, outdated by automated agents that can solve business problems on the fly. And once asset-light tech platforms look increasingly like indebted industrials, forced to spend billions of dollars to keep up with AI's substantial financing demands. Tech investing, put simply, has been turned on its head.
While the volatility is painful, it is also creating big opportunities for investors, and it's a perfect time to be picking stocks. With that goal in mind, Barron's recently brought together four tech specialists to make sense of the 2026 tech investing landscape: Glen Kacher, chief investment officer of Light Street Capital; Brent Thill, Jefferies senior software analyst; Morgan Samet, managing partner and co-head of Lingotto Innovation; and Ben Reitzes, head of technology research at Melius Research.
They joined us via Zoom earlier this month. Here is an edited transcript of our 2026 Tech Roundtable.
Barron's : Given all the uncertainty in the tech world, let's start with the things we know. What, if anything, seems certain to you about AI and the tech space in the coming year?
Glen Kacher: The emergence of AI as an architectural computing change is driving the industry forward. The largest companies with the most capital to spend are saying this is the most important thing happening in our industry. And they're spending billions and billions of dollars on capex [capital expenditures]. So, for me, the thing that's most certain -- and what we hang our hat on -- is that semiconductor demand is extremely strong, and the demand for AI compute cycles is still very, very strong.
Does anyone want to challenge that?
Morgan Samet: I would potentially pile onto it. The market potential for what AI can do in not just the digital economy but also the physical economy is massive. And it is wildly underappreciated. The market gets caught up in certain debates about the timing of cycles or whether the return on investment is real. It is short-term-oriented, focusing on niche parts of the value chain, such as software. That misses the big picture.
It isn't software that is being disrupted; it's labor. The labor markets are on the order of $10 trillion to $20 trillion, whether in the digital or the physical economy. When you see things through those lenses, it changes the incentives of the decision makers involved.
It takes a while for these things to get ramped up, for the pipes to exist, for the decisions to get made. Trust and security are incredibly important. But if you take what these tools can do in the consumer market without bounds, and then apply that to what can be done at the enterprise, the opportunity is massive.
I don't want to spook everybody when I say it is labor that's being disrupted, because the pace of the labor deployment and the friction indicate how much time we have to create new jobs and reorient labor into higher-value tasks.
Ben Reitzes: The certainty is that the rate of change will be accelerating. I see more disruption ahead, not less.
Brent Thill: Two things are certain. AI will be globally adopted by every organizational board. We didn't see boards move to the cloud. We didn't see boards think about Bitcoin, the move to Web3, or many of these other tectonic shifts. With AI, multiple board members I've spoken with have said that they're all in, and they're reorienting their companies across every industry. All the field work we do points to absolute global adoption starting at the board level -- not the IT level, where we usually saw past tech trends start.
The second area is infrastructure. Everyone in the infrastructure market, including Microsoft, [ Alphabet's] Google, and Amazon.com, is seeing incredible demand that they can't keep up with.
Are board members being dragged into this, or are they going on the offense because they already see the benefits of AI adoption?
Thill: They are acting offensively. When JPMorgan Chase said it had reduced the total operational spend for support but put that money back into revenue-generating roles, that's a board-level topic. If you can cut costs and improve revenue, that is an offensive move.
Kacher: What is amazing about AI is that the industry has built these tools that are incredibly productive for the single user. It is very approachable for a CEO and tangible for all of us. Everyone has had an individual personal experience with this technology, helping them to create content. And then, they're hearing about the efficiency of it writing code. That is an amazing advancement for individual productivity. The gap we are all struggling with is that it takes time to build the tools for teams. That is where the doubt comes -- in terms of applying these tools for large numbers of people to use together.
Does the shortage of infrastructure delay things significantly?
Thill: In some cases, it already has. CoreWeave had to push out $8 billion of capex in the fourth quarter, and had issues with getting physical buildings up and live. The demand is exceptional, but the schedules are shifting materially. It can come down to [a lack of] PVC piping that could delay a project by months. Things are getting pushed out, but that doesn't mean the intent to spend isn't there.
The big question is return on investment. When do you see returns, or monetization of this technology? When will AI have a meaningful impact?
Kacher: Every time we have seen a major architectural change filter through the tech industry, it takes a decade or more for it to be widely adopted. Look at software-as-a-service, or SaaS, for instance. Salesforce was founded in 1999. It took 15 years for roughly 25% of computing cycles to shift to the cloud model. It took another eight years to get the next 25%. So it takes time to create these applications for large numbers of people to use.
Today's software isn't going away. You will see AI adopted on top of existing applications at the core of running an enterprise, and incumbent software players will build some of those applications. There will also be native AI and LLMs [large language models] they have to compete with.
Samet: You are seeing this in consumer spaces first, because there are no barriers to entry in a lot of products. Some people are willing to take bigger risks than I am to test them out on their devices.
Then, as you think about the enterprise, Glen made a good point: It is stitching it all together that is difficult. Executives are operating from a place of both fear and opportunity. People act faster when you pair fear and opportunity.
I now have subscriptions to Claude, Gemini, and ChatGPT, and wonder every day which one to use. Are enterprises dealing with this question as well? How do we know which winners will emerge here?
Thill: Multiple AI vendors are going to win. Intuit uses more than 20. It understands that Gemini's better for this, ChatGPT's better for this, Anthropic's better for this. Same thing at Microsoft. Same thing at Snowflake. You will see many companies try to find a framework or governor of LLMs, because they don't want to just use one.
For example, Claude is really good for development, and you may not use Gemini for development. But you want the power of all these LLMs working for you. Everyone thought they would become commoditized, and now they are becoming more specialized. The embrace of many of these models inside a system that can help orchestrate their use is the preferred route from many of the vendors. One AI company coming in and crushing the entire organization is probably not going to happen.
You all make it sound so clear, but in the markets, things are chaotic. How should investors think about all of this?
Reitzes: This news cycle is like nothing I've ever seen, and I was an analyst in the dot-com bubble. Investors need to be aware of the narratives that have taken over the market. A few years ago, we said AI was eating software. We have been surprised by the degree to which this narrative has picked up steam. Given the rate of change, investors need to ask, what is the narrative that could take over negatively, as well as to the upside? There is a lot of shoot first, ask questions later, because we don't really know how things are going to change.
So you need to be aware of the fundamentals, and the narratives, as well.
Kacher: There is a paradox here. Ben is right that the rate of change is high. There are many narratives going around, and the one I find most interesting or conflicting is from the tech bears who say that AI is so powerful, it's going to destroy all of software. But at the same time, they say there is no return on investment coming from AI, and therefore semiconductor demand is going to crater. You can't destroy an industry of incredible value if the product doesn't do something incredibly valuable. So it is a very bizarre conflict.
In that case, in what areas do you invest?
Kacher: For me, it is an incredible time to be an investor in the semiconductor industry.
AI writes code; it writes content; it matches ads to content, and users to content; and it improves advertising yields. We have seen massive ROI for Facebook, Google, and Amazon on the advertising front. It is able to execute a lot of tasks more cheaply than humans, and arguably better, in certain cases.
Reitzes: The tokenization of labor is occurring. It is scary to many, but AI is going to augment labor and explode productivity. The economic returns are huge.
This laborer wants to know what you mean by labor becoming tokenized.
Reitzes: It's labor becoming software or digital labor. AI has the ability to be a Ph.D. in your pocket, a doctor in your pocket, an analyst in your pocket, an accountant, and whatnot. It will augment a lot of these positions, and make people doing these jobs much better. That is like a tokenization of labor, using AI tokens to produce output. The unfortunate thing is that we have gone through many cycles of labor disruption that take place over 10 to 20 years, and this one feels like it is taking place over 10 to 20 minutes.
But we heard earlier today that there will be time for labor to right-size itself.
Reitzes: With all due respect to what was said before, I don't think we have a lot of time. You're already seeing a pause in hiring for knowledge workers.
Samet: To clarify, it is going to happen much faster than people think. But there is also far less friction in learning something new or executing on something with different levels of knowledge and lower barriers. You will see, especially among younger people, more entrepreneurs launching smaller businesses or flatter structures as a result of these tools.
Kacher: Every one of these transitions took 20-plus years, from horse-drawn carriages to gas-powered trucks. The iPhone came out in 2007. By 2013, smartphones had reached 50% penetration. We always overestimate the speed at which these transitions disrupt existing industries. The Armageddon of employment is probably overstated in terms of speed.
Reitzes: Many of the software companies I talk to don't realize how quickly they may need to move to consumption models, rather than charging per seat for access. That is what we call a negative mix shift.
What does all of this mean for specific software companies and stocks? Brent, would you like to start?
Thill: From a software perspective, Snowflake is a huge winner. It is growing by close to 30%, with widening margins. It has the best margin expansion of any large-cap software company. It has one primary competitor in Databricks, which is doing extremely well, so it is a two-horse race right now. We continue to see that you can run every AI technology in Snowflake's platform and do it safely.
Explain what Snowflake does, please.
Thill: Snowflake is a data platform. It can analyze all your corporate data to help you make better decisions about which products to launch, what expenses to cut, which trendlines competitors are seeing, and so forth. It is effectively a watchtower or control tower for all your data. It has its own AI agents, which users can deploy. It also partners with all the big cloud vendors, including Amazon and Microsoft, and recently launched on Google. You can run all kinds of agents inside the framework on the Snowflake platform. The hyperscalers like Snowflake because it brings more workloads onto their core compute platforms.
Samet: The control-tower narrative is spot on, and you're going to need a few of those to be able to guide your systems internally. Never in an enterprise do you ever want to be relying on just one company.
We are investors in ServiceNow, and we own Databricks in the private market. But the most underappreciated company in this area is Cloudflare. I say underappreciated although the price/earnings multiple is rich, but when you look at the way the stock trades, it is clear that the market views this as a security play.
Between Cloudflare's content delivery network and its security, it sits in front of approximately 20% of the web. But most valuable now is its physical infrastructure--its servers that sit closer to users, or what's known as the edge. As you think about the transition to the edge and the move to agentic AI, latency and geography matter. And as you have more bottlenecks to data-center deployment, Cloudflare already has a broad physical footprint that is spread out, helping move data with less delay.
We have seen revenue accelerate materially in the past four quarters.
Kacher: We like businesses on the infrastructure side of the software industry -- those that provide the picks and shovels and have business models that monetize usage rather than seats. The company we like best there is JFrog, which runs a system of record or artifact-management platform. It keeps track of the inventory of code in each release of software and the development process, and is used by 80% of Fortune 100 companies and many leading AI labs.
We think it will generate more than two bucks of free cash flow per share in 2027. Free cash flow is growing by 40% a year. The stock is likely to trade at 30 to 35 times free cash flow, which implies 50%-plus upside. The market has incorrectly identified this business as at risk from AI. In fact, the inflection and rerating will occur because it is an AI winner.
Reitzes: We take an Nvidia-centric view of coverage, and broaden it out. But if I had to pick a stock in the software space, it's IBM. There is a perception that IBM's mainframe business could be threatened by AI. We don't think so. In a world where we're short of compute and CPUs [central processing units], it may benefit on-premises compute, which could see a boost due to AI.
The majority of IBM's software follows a consumptive and instance-related pricing model, not SaaS, and we think that is a better place to be. Long term, there is a quantum computing call option. A lot of folks think that is pie in the sky. We have gone to their labs and been impressed by what IBM is working on in quantum.
Quantum and classical computing will have their moment at some point in the early 2030s. IBM is the stock that we're sticking with that is probably the closest thing to software. We're more cautious than Wall Street on names like Microsoft and Adobe.
There are likely to be a lot of chip factories opening up starting the end of this year and into 2027 and 2028. What is the opportunity there for semiconductor equipment stocks?
Kacher: We're big fans of ASML Holding. The company is absolutely necessary in making advanced GPUs [graphics processing units] and XPUs [other AI chips], and all the chips powering the buildout of AI.
We see the greatest incremental demand in the packaging space. One company we like a lot is MKS. Over the next two years, advanced semiconductor packaging -- how silicon wafers are turned into chips -- should continue to grow faster than the front-end wafer-fab industry. That is driven by AI accelerators, high-bandwidth memory, and then chiplet-based architectures.
We think MKS will have better than $13 of earnings-per-share power in 2027 on a modest recovery scenario, and could trade for 25 to 27 times earnings. That number gets you a $350 to $400 stock, or 60% to 80% upside. The company is well-positioned. It is a key supplier to TSMC [ Taiwan Semiconductor Manufacturing], and that is a huge tailwind.
Ben, what are you thinking about in hardware?
Reitzes: A lot of the AI semi stocks have stalled, from Nvidia to AMD [ Advanced Micro Devices] to Broadcom. Those are the elite AI semi names. The stock prices have taken a breather in the past couple of months because there is a fear of peak capex as software becomes increasingly tokenized, and as labor becomes tokenized.
To make these stocks work, we need to see upside in cloud revenue at the major hyperscalers. Based on our work, we feel that can occur in 2027 and 2028, with significant upside to revenue. That will eventually make investors feel better about the sustainability of cloud capex heading into 2030, and that could help those stocks.
We can't talk about hardware without discussing Nvidia. It is trading for 21 times this year's expected earnings, which is even with the broad market, despite much faster growth. Why is the stock so cheap, and is it time to buy?
Samet: It is astonishing. Nvidia continues to rerate down. It makes no sense, and it has nothing to do with the fundamentals. It has more to do with market structure right now, since 7.5% of the S&P 500 is Nvidia. That pushes limits in terms of how much certain funds can invest. And it is unclear whether that will change. At some point either the numbers will improve or there will be an announcement about an expansion into a new area with near-term revenue and near-term agreements that will lead to another gap up in the stock.
Kacher: If you look at the consensus estimates, you see 75% growth in earnings per share expected in 2027, and 30% to 35% expected in 2028. Arguably, there is the potential for 40% earnings growth in '28. The stock is trading for 17 times consensus 2028 estimates.
Compare that to Apple, which is struggling to grow earnings by 10% a year and is trading at a mid-20s multiple. It doesn't make a whole lot of sense. It again ties back to this faulty idea that there isn't much return on investment for AI compute. That is false.
We own Nvidia and like it a lot. We also own Broadcom for similar reasons. It is another competitor in the inference space, and is an incredibly well-run company. It isn't as cheap as Nvidia, but it is arguably positioned well as we move toward more inference computing. That's the way we're playing it.
Reitzes: I'm at Nvidia's GTC conference right now. [Nvidia CEO] Jensen Huang gave an order number for $1 trillion in hardware revenue that was phenomenal. It looks like there is meaningful upside in every quarter through the end of 2027.
There is concern about peak capex and return. But I have been through these things before. I had recommended Apple starting in 2004, in that prior cycle. When it started to hit $100 billion in market cap, there were a few unofficial market-cap walls. People forget, but there was a perception that Apple was just a hardware company, and that eventually hardware companies always get disrupted. Apple was able to make that transition to services and massively expand its multiple.
Samet: Perhaps the market misunderstands the level of Nvidia's software play. You can't build a robotics company or an autonomous-vehicle company without them.
We haven't talked much today about the rest of the Magnificent Seven stocks, which last year's Tech Roundtable discussed obsessively. What is your view?
Reitzes: The Mag Seven are already under assault by the market rotation under way. The overwhelming issue is the lack of free cash flow for many of them. The free cash flow for Alphabet's Google, Amazon.com, Tesla, Microsoft, and Meta Platforms is very much in doubt. Companies of that size are generally relied on for some level of stock buybacks and dividend growth. So in addition to the rotation, this lack of cash flow for many of the companies is something investors are wrestling with.
Then, on the horizon, we have Anthropic, OpenAI, and SpaceX, at least, that could come public. There could be the Mag 10, or the Mag 11. How are investors going to make room for that? Which stock do you sell to make room for new names like these in a portfolio, or which do you not buy to make room? The specter of that decision is hanging over the whole group.
Are there other stocks you'd like to share with us that you find particularly interesting?
Thill: Meta is really interesting. It will earn more than $30 a share in the next year. Put a multiple of 25 or 30 on that, and you have a materially higher stock price. To Ben's point, when you think about what gets crowded out, Meta's world is so immersive and so different from what anyone else is creating that we think it is going to have a seat at everyone's consumer table. Our worlds will be incredibly richer because of what they are doing with AI.
The cheapest name right now is Amazon. The death of AWS [Amazon Web Services] is greatly exaggerated. We saw Alphabet trade at the beginning of last year at 11 or 12 times Ebitda [earnings before interest, taxes, depreciation, and amortization]. Amazon is now at the same multiple and below the average multiple of Walmart on a next-12-month enterprise value-to-Ebitda basis. That one just seems oversold.
Thanks, Brent. Morgan, any more names from you?
Samet: The Chinese battery maker, CATL [ Contemporary Amperex Technology]. The cost differential and the quality differential between CATL and the rest of the world is massive, and even Tesla is using CATL's cells. The market is focused on geopolitical tensions, which are real, but in the short term there is no solution without CATL. China's automation of the physical world and the rest of the world outside of the U.S. involves very large markets in terms of geography and things that can move and need energy.
Thill: One beneficiary in software is going to be vertical applications. Procore Technologies sells software for construction management. The company partnered with Nvidia to bring AI to the construction space. Construction is the most manual, highest-waste industry on the planet, and AI can bring better efficiency. We think Procore is a great mid-cap story that is vertically focused on the world of global construction.
Reitzes: Can I add one? I talk a lot of love for AI semis, which most people think is related to Nvidia, Broadcom, AMD, and even Marvell Technology. I want to give a shout out to Intel. We need more foundries. We need more advanced chip-making capabilities. Intel obviously has those assets.
[CEO] Lip-Bu Tan is quite an amazing executive. This is a big change, vis-à-vis what the prior leadership was all about. He is revered by many of his competitors. And he possesses the skill set to do what Lisa Su did with AMD: fixing the product business and then creating optionality for their foundry business. Given that the U.S. government is an Intel shareholder and he has President Donald Trump rooting for him, and that we need foundry assets, there is a chance Intel could make this happen. The company also has some hidden packaging assets that are doing really well right now.
Glen, any stocks you'd like to mention that we haven't yet talked about?
Kacher: Brent had a good theme in vertical software. Our pick in that category is Cellebrite, ticker CLBT. It is the dominant player in digital forensics -- a provider of software to law enforcement all over the world. It sells the key technology used to crack a cellphone open digitally and analyze its content. It is going to generate a dollar a share of free cash flow in 2027, growing free cash flow at 30%. We think the stock has 50% to 75% upside.
Cellebrite just introduced an AI product that helps law enforcement officers look at the information and communications that have happened across the entire suite of applications on the phone, and make sense of those communications. It solves crimes and increases the speed at which we can close investigations and come to a conclusion in the court system. It is a tremendous company and a dominant player.
Let's end this roundtable by asking what will we be talking about at this time next year.
Samet Two things. One is more about the consumer. Outside of ChatGPT and AI assistants, we haven't seen a major unlock that affects the everyday American. We are bound to see a new platform, a new product, and a new service that become pretty mass.
Second will be more technology in the physical economy. It will roll out faster than most people think. Autonomous vehicles is an area that we have been focused on. The thing I feel most certain about is that the technology is solved, and it is happening faster than people think.
Thill: Mine is, why is capex continuing to go materially higher? That is going to stunt potential software growth. As capex keeps increasing, it is going to benefit infrastructure and semis and hardware. It isn't going to accrue to software.
How do you think the market will react?
Thill: Investors will just keep buying energy, data-center infrastructure, and hardware stocks. This is a broken record, right? It has been happening every year, and everyone says, well, at some point it will end and capex will decline, and we will see a shift from infrastructure to software. We haven't seen it. Semis keep outperforming.
Reitzes: We are going to be talking about data. One of the big surprises is how much data all this is going to take. Everyone gets that we're going to need a lot of semis, a lot of tractors, a lot of construction. We are going to need a lot of data centers. But we are going to need physical data storage.
We're going to need more flash, more hard drives, and more data storage systems both for the cloud platforms and even on-premises, and on our devices. AI makes your data more valuable. Don't throw away those old emails.
Glen, your thoughts?
Kacher: The debate in another year is going to be, can we keep spending at this rate? The numbers are going to get larger for the AI buildout, and the percentage of free cash flow that the Mag Seven are spending will become a much bigger issue. Those doing the spending will get tougher questions asked of them.
There is a lot to think about here. Thank you, everyone.
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 27, 2026 01:00 ET (05:00 GMT)
Copyright (c) 2026 Dow Jones & Company, Inc.
Comments