Is A.I. Spending Sustainable?

Data is everywhere, but is it ready for consumption? Morningstar developed the language of global investment data so you have the right ingredients to help you shine. Morningstar, where data speaks.

One of the questions I get is, can this go to zero? So my answer is there's absolutely no way the spending right now is going to go to zero.

Hello and welcome to the Barron Streetwise Podcast. I'm Jack Howe, and the voice you just heard, that's Venu Krishna. He's the head of US equity strategy at Barclays, and he's talking about AI infrastructure spending. It's not going to zero. It could go down. He doesn't necessarily think it will go down from here, but it could, and he's going to tell us what that would mean for stocks. Don't be alarmed about that. Venu is, I would say, broadly bullish. I, on the other hand, I'm going to quickly run through a few warning signs for investors right now before we come to that conversation with Venu. This is not because I think you should sell your stocks. It's impossible to try to spot the top of a bull market. It's because, I don't know, once in a while I just like to work myself into a frenzy about stuff to worry about and then decide in the end to just leave everything the same and see what happens. It's part of the fun. Listening in is our audio producer, Alexis Moore. Hi, Alexis. Hi, Jack. The S&P 500 is up a frisky 15% so far this year. Hooray. That's nothing, by the way, compared with Bitcoin, which is up 30%, and gold, which is up more than 50%. Folks increasingly are calling it a debasement trade. That means if you're worried that the federal debt is getting away from us, reaching escape velocity, or if you're worried that the Federal Reserve is going to lower interest rates to a point where it spurs inflation, maybe you're doing things to move cash into assets that are disassociated from the dollar. That's the backdrop. Maybe some of those people who are worried about inflation are buying stocks too. The S&P 500 does look expensive at 25 times earnings. But historically, there is very little correlation between current valuations and short-term returns. That's another way of saying that just because prices are high doesn't mean that they're not going to go much higher. They might. You could sell a chunk of your stocks and raise cash in an effort to protect yourself from what you think is an impending market downturn. That reduces your risk of capital loss in the near term, but it doesn't reduce your risk that You're wrong or early and your fully invested neighbor ends up doing better than you do through the rest of the bull market. There should be a high finance term for that risk. Suburban deviation, maybe? I'm pretty sure it's a driving force of late-stage upswings. So, I'm going to run through what I think are glaring warning signs that I have cherry-picked. Let's all agree up front that after I run through them, we will immediately ignore them and stick with our carefully planned allocations. Number 1, I'll call AI open relationships. If that sounds mildly creepy, then I've done my job. Are you familiar with polycules? A who? A polyhood? A polycule? Do you know what a throuple is?

I don't.

Is that something you use an antibiotic for? I don't know. It is. A polycule is when you're in a relationship with multiple people. 4 or more, and you're all in a relationship together, a.k.a. the polycule. I see. And so thruple would be 3? Correct. But that— but the term was not big enough to contain everything that's going on. They needed something bigger than thruple. And they said, let's not go to quad— well, I guess quadruple was already taken. So I don't know what you would call that. They said, you know what, let's just go to poly. And then we're covered no matter how high it goes. All right. Look, I'm learning things. It does sound a little bit like this Morgan Stanley report I read recently. This thing has a chart. Look, the report is titled AI Mapping Circularity. Morgan Stanley is attempting to map the industry's circular financial relationships. And I would describe the result as jarring to look at, even if the participants seem to be having fun. Chart includes the big companies we're all familiar with— NVIDIA and AMD and Oracle and Microsoft and CoreWeave and OpenAI. We talked about CoreWeave on this podcast that we heard from one of the co-founders. That was earlier this year. That's a company that used to be in crypto mining and they changed over to basically buying and deploying NVIDIA AI chips in data centers. And it's an odd arrangement because NVIDIA is also a key investor in the company. And there's some vendor financing involved there. Something similar goes on with Oracle too. We hear about a lot of these relationships. Last month, NVIDIA agreed to make this massive investment in the creator of ChatGPT, OpenAI, to help it build out its computing infrastructure. Then this past week, OpenAI signed a partnership with NVIDIA's rival, Advanced Micro Devices. There are so many solid lines and double lines and dotted lines and curvy lines and different color lines connecting all these companies that it's hard to make sense of who exactly is involved with whom and how. Morgan Stanley argues that more disclosure would be helpful to investors. It calls out something called remaining performance obligations or RPOs. That's a measure of future contracted revenue. It points out that OpenAI accounts for about two-thirds of RPOs at Oracle and 40% at CoreWeave, meaning that both of those companies are quite dependent on OpenAI's success. It also says that across the industry, quote, new innovative finance structures and off-balance sheet partnerships are making it, quote, challenging to evaluate the risks. Keep in mind that AI companies have accounted for pretty much all of the S&P 500's rise since the public launch of ChatGPT. That means that even if you're just an ordinary index fund investor, you have plenty of exposure to the AI poly whatever it was that Alexis, what was it? Polymolecule?

Yep, exactly.

Polycule. Polycule. PolyCube, the, uh, you have more exposure than you might have thought to the AI PolyCube. So I'll call that warning sign number 1. And number 2, I'm going to label this refried memes. Everyone remembers the GameStop hoopla. This was several years back. The heavily shorted shares of a company that is mostly in the fading business of selling video games on discs through stores. Those shares suddenly soared, and they did so largely because stock flippers who look for ideas in a Reddit chat room called WallStreetBets, they decided that sending that stock higher would be funny and profitable. And there were some other unlikely stocks that jumped around the same time. There was a struggling movie theater chain. There was a company that had Zoom in the name, not the Zoom that you're thinking of, but another totally unrelated Zoom. In other words, They sent the wrong Zoom hire. There was another company that was a holding company for assets of the old Blockbuster Video and so on. And so in December 2021, there was an exchange-traded fund that launched. It was called the Roundhill Meme Stock ETF. And in a turn of events that no one saw coming, and by no one, I mean absolutely everyone, that fund turned sharply lower. Almost immediately. It closed after less than 2 years. And as of this past week, it's back. Roundhill has relaunched its meme stock ETF. So I guess since the first launch marked a speculative peak, maybe this relaunch is a harbinger. A bad harbinger. A canary in the crypto mine.

Whoa.

Wait a second. Write that down. I'm using that for a headline for something. But I'll just point out that if this is a harbinger, it's a tricky one. The original fund used short-selling activity and online chatter as gauges of meme status, but the new one uses an active stock picker. And if you look at the top holdings, you see quantum computing, fuel cells, crypto, artificial intelligence, and rare earth metals. A lot of the stocks are recent high flyers with distant visions of profits. For example, We talked in the past about quantum computing and we might have mentioned Rigetti Computing. That's the top holding this fund. It took every ounce of thinking power I have to try to describe what quantum computing is the first time. I would have to limber up before trying again, so I won't do so now. But suffice it to say that Rigetti does less in revenue than a well-located liquor store and it isn't expected to turn a profit this decade, but the stock is up. More than 6,000% in a year. And so I think that would be an excellent candidate for a momentum fund or for a venture fund. A manager of that fund could try to tell the spicy growers from the pipe dreams. But Rigetti and these other companies I see in the fund, they lack what I think is the only necessary defining characteristic of a meme stock. To me, a meme stock must have a punchline. If you knowingly buy the wrong Zoom stock and then you convince others to do likewise, maybe that's not exactly funny, but I think on some level you're trying to do something funny. If on the other hand, you're just buying quantum computing stocks because they're going up a lot, I think you're being sincere either about the business prospects for quantum computing or just the trading momentum. So if a once fallen meme fund relaunches without the meme stocks, is that a warning sign? I don't know. But if the company that relaunches the meme fund doesn't really seem to get what a meme is, maybe that's a meme of its own. Okay, the third and final warning sign. I mean, this is a little obscure, but I would just point out that not only are Bitcoin and gold beating the stock market this year, but convertible securities are beating the stock market too. The iShares Convertible Stock ETF has returned around 23% year to date. You all probably know what convertible securities are. They're, they're kind of hybrids between bonds and stocks. Corporate bonds and stocks are born when companies want to raise money from the public. And so to do so, they agree to either pay an interest rate, that's a bond, or to give up part ownership in the business, that's a stock. Some companies agree to do a little of both, and that's a convertible security. The companies that issue them, I think, give up kind of a lot. So what kind of companies are these? Well, the top holding in that iShares fund is a company called Strategy, whose business model is to raise funds and use them to hoard Bitcoin. Crypto and fuel cells and AI cash burners are well represented in the fund. So the fact that convertibles from these companies are paying off means that the stocks are going nuts. Not that some neglected pocket of the fixed income universe is suddenly getting its due. Is that a warning sign? I say maybe. You might see other warning signs that you're worried about. If you do, drop us a line. You can record yourself on the voice memo app of your phone and send it to jack.hough@barons.com. Maybe you'll hear yourself on a future episode of the podcast and we'll dig into your warning sign. Time for a quick break. And when we come back, we're going to hear from Venu Krishna at Barclays about AI spending and why you shouldn't be too worried. He also has an interesting hedging idea for investors who are concerned. That's after this quick break. Data is everywhere. When orchestrated properly, it sings. At Morningstar, we analyze and enrich data, making it actionable and powerful for you. Morningstar, where data speaks. Welcome back. CapEx by AI companies, in other words, capital expenditures, spending on infrastructure, it's near a record high as a percentage of revenue. A lot of these big companies have tons of cash flow, so they can afford these investments, but other companies are borrowing. And since this spending is fueling profit growth and the stock market, it makes you wonder whether it's sustainable. Venu Krishna, the head of US equity strategy at Barclays, has dug into that topic recently. I spoke with him about that, about what would likely happen if there were a decline in CapEx spending by AI companies. What would it mean for stock market investors? About what investors should buy if they're looking for a little more diversification. And he also mentioned, you'll hear this near the end of our chat coming up, hedging strategy that involves options. Here's Venu.

We are in a sort of transformative tech cycle broadly defined by AI right now, along with cloud. But we have the same concerns to some extent that the market is getting ahead of itself. And certainly there are pockets of the market where we think that concern is real. And that's why even in the note, where we wrote in AI, we are specific to suggest how you should think about hedging. I think we've done a lot of analysis and comparison, so we don't think it's comparable in along many different metrics with the dot-com bubble. Though, as we continue in this phase of investments, we will start getting some contours which might resemble what you saw in the dot-com period. And that's the reason why, frankly, the focus of our note was essentially what can go wrong. By some estimates, you know, almost 70% of first half GDP was driven just by the AI-related CapEx spending, right? So that does give some concern as what if it can slow down?

Well, I was gonna ask you, how do all this money that's being spent on infrastructure, is this infrastructure being put to immediate use? In other words, are we building any capacity that's sitting out there unused? Are we building ahead of demand Or are we still trying to catch up with demand?

Right now, demand is outstripping supply. So when you talk about demand, it's a demand for compute, right? And right now, AI, for example, is already being used by the biggest spenders themselves in their core businesses. And that's the reason why these big spenders are not only spending, and mind you, from their own operating cash flows largely, but their net profitability last quarter actually improved by almost 200 basis points. What they're doing is already deploying some of these. So that's the reason why there's a dichotomy. You can have a Microsoft, you know, laying off thousands of people, and yet the profitability is improving because software productivity has improved anywhere from 20 to 40%. Company after company is telling you. So if you're a software native or a software-centric business, then you're already seeing the benefits of AI and monetizing it. What has not yet happened is enterprise-level utilization. Right now it's at the consumer level where OpenAI, for example, I think already has close to 800 million users. And that's where most of the compute capacity is being used up to handle those queries.

Consumer side is people kind of tinkering or doing things with ChatGPT. They go in there, they ask a couple of questions out of curiosity, or they've built it into their regular everyday work. But you say we're only scratching the surface of companies figuring out how to put AI to work on their larger datasets.

Yeah, and obviously different companies are at different levels. Some early adopters or people who have been more aggressive have found better ways to sort of use it sooner rather than later, but it's an ongoing process. But that doesn't mean to your core question that, you know, when people talk about the dot-com bubble, the question really comes down to is this our dark fiber moment? And what they mean by that is will the efficiencies improve so dramatically that it leads to a critical underutilization of infrastructure assets?

In this case, dark fiber was all this fiber optic cable back in the early days of broadband that people, that companies laid, and there wasn't yet demand for it. So it was never turned on or it wasn't immediately turned on and it became a glut.

Not just that. I recollect that period very well. Not just that. The companies laying it were telling you, giving you all kinds of physics sort of analogies and citing the Moore's Law to say how exponentially demand is going to increase and why you need significant capacity and that it gets eaten up, right? And the question is all about timing. So I think they were right, but they were not—

It was true. They were just off by what, 15 years or something like that?

Exactly. So the companies really invested. Ultimately, they did not benefit truly from it, but internet did change the world. Did change business, did increase productivity, did lead to multiple new business models. And so eventually the capacity got utilized, but there was clearly an overbuild. So that is the concern right now. And some of the companies, you know, who are spending have acknowledged that, you know, you could mistime it, but it's a race now because in technology, what the last two decades have shown is there is a first mover advantage. And there is a platform phenomenon where, you know, you control stuff if you are good at it. And so none of these companies wants to miss out on that race. And so you can, one can argue, and that's a legitimate concern, is that this arms race within the AI landscape is what might lead to a potential misallocation or mistiming of capital spending because the scale of it, even if you inflation adjusted, is far, far greater than what you saw during the dot-com period. Right. But, you know, like we discussed in our note, fundamentally it's very different because the companies spending then were in a far weaker sort of starting point financially compared to the companies doing it right now. But going forward, if they keep spending at this rate, then even their cash flows might not be enough.

We hear more and more announcements of these circular financial relationships between the AI giants. These companies seem to be their own best customers. Companies are providing financing to their partners and investing in them and selling to them. So among these companies, NVIDIA and AMD and OpenAI and Microsoft, there just seems to be this very complex web of close relationships. I mean, it makes me wonder, is there any portion of this spending that we need to be concerned about because of these circular relationships?

I think it is worth tracking and noting. I don't think the investment per se is something negative, but I did find, to your point, something very strange where, for example, in the AMD OpenAI deal, you know, OpenAI is actually going to be buying chips from AMD, but then AMD is giving them essentially warrants. Or effectively an equity stake to buy chips from them. So that did surprise me. And what that tells you is that in this case, looks like perhaps OpenAI had a stronger hand because AMD is a distant number 2 in the AI chips sort of provider. And so it's also in AMD's best interest to lock in substantial potential demand. And if that means partly giving some equity stake, So be it, because they're intertwined.

It seems harmless enough, but like, if I'm buying a massive number of chips from AMD, my goal as their customer is to stick it to them on the price. I want to get as low of a price as possible. I want a deal where I'm walking away whistling and they're crying about the price that I paid. But if I'm now an equity holder or a warrant holder, or I'm an investor who has— who participates in the upside of AMD, I don't want AMD to get beaten too bad on price. I want them profiting richly so that my shares or warrants go up. So it puts me on both sides of the relationship.

It's a trade-off, right? That's why if you see all the big tech companies, they do stakes, stakes. You know, for example, big consumer companies take stakes in startups, right? One for market intelligence to see where is the next wave of sort of innovation coming, and then potentially having the business knowledge and the option to acquire it. And build it. So I think in your case you raise an interesting point. At the same time, it's also in the interest of the people who are buying chips to not have one company have 85% market share, which is Nvidia right now. And that's the reason why in Nvidia volumes are up, prices are up, margins are up, right? That's because demand is so outstripping supply.

So you don't see this as a bubble. You believe that the spending right now is warranted considering the demand and the growth potential and so forth. You mentioned this might change and where we could see CapEx slow or decline or whatever. What, what are those things that could happen that would be problems for this buildout of AI?

I think there are 3 or 4 broad things to be aware of. Top of the list is the power constraint or what we call the power wall, because the growth in data centers right now, which is needed to keep them working, is not keeping pace with it because there's just way too much demand. And by the way, demand also affects the broader pricing for the average consumer for their power bills. So there's a political aspect to it as well. You could end up in a situation where there's simply not enough power to make these data centers run.

People used to say, I don't want a nuclear plant in my backyard. Now they say, I don't want a data center in my backyard. It's going to double my electricity prices. Bring me a nuclear plant to bring those prices down.

I mean, I, in fact, interesting you raise that point. I read an article in your Barrons, which I was not aware of, the IPO of Fermi, which one of your colleagues wrote about. And that's a pretty amazing story. Its market cap is $16 billion. It's a single asset REIT. There's no revenue right now, and it's going to supply power for data center operators. And for example, nuclear reactor was the first one. You know, the article suggested it's going to take them 6 years they are thinking of getting it online. Meanwhile, it could take typically 15 to 20 years given the approvals, the process, and, and everything involved in that. So power constraint is real. But stepping back, that is, I think, the single biggest constraint which can slow data center growth. The second is just what if we hit a demand wall? In other words, the demand for compute is significant in the pre-training phase where we are right now. As you move from pre-training to what is called inference, the demand for compute goes down. Along with that, if you have more efficiencies in models, for example, you could have a situation where demand and supply can be more in balance, in which case you don't need as much capital spending, for example. The third one is if you keep spending at this rate, then funding will be a bigger issue. I think as the scale of funding keeps rising, then how you fund it and whether the next round of funding is on as sound a financial footing as the previous cycles have been will be something worth watching.

I want to return to something that you mentioned early in our conversation, which is ways that investors can hedge right now. You know, when we think about all these risks, the risk of a downturn in the stock market, I always think to myself, you know, if there's a downturn in the stock market, we all lose together, right? There's some comfort in that. But if you sell early and you sit there while your neighbor is getting richer than you, like that to me feels like a big risk. I think that that's, emotions that drives, you know, the sort of late stages of a bull run like this is people don't wanna miss out. So if you're this person who doesn't wanna miss out, but you wanna do something to protect yourself, what can you do to hedge right now?

Couple of things. One is that, like I said, the AI story is way beyond just the hyperscalers. So what is happening is when you, you know, so much investment goes into data centers, then the providers of connectors, electrical equipment, cooling systems, turbines, the just the building material. The list goes on and on. So there are subsectors within energy, industrials, power, utilities, and the rest of tech outside of big tech who are all what I would call auxiliary beneficiaries of this CapEx spend. But what we find over there is that the exuberance is beginning to show. So even though the AI-related revenues are small right now and are growing at a healthy clip for these companies, but in a lot of cases, their multiples have moved up very, very fast.

To your point, I saw— you can probably remember not so long ago that hard drive stocks used to trade at single-digit price-earnings ratios. It used to be the most boring part of tech. But now suddenly I read that there's close to a year waiting time for these high-capacity drives for AI data centers, and the, and the hard drive companies, they're trading at closer to 20 times or thereabouts earnings. So that's one example of a related tech stock that's really moved up in price. I'm sure there are many others.

That's an excellent example. I would add parts of utilities. There are some power utility companies trading at 35 times forward earnings. For example, our firm, my derivatives team launched a, what we call the Equity Euphoria Index. And what we are trying to do over there is try to gauge the level of euphoric behavior in equity markets by getting information from the derivatives markets. So we look at liquid options of over 1,500 stocks and see what percentage of the stocks are exhibiting euphoric behavior. And right now that number is close to 12.5%, which is coming to peak levels seen over the last, call it, 30 years. And the second phenomenon is this exuberance is more among smaller companies, and there's a good representation of these AI auxiliary beneficiaries within them. One of the things we suggest people is to look at these beneficiaries and to look at where there are cheap options over there. And for example, buy protection and different structures in which you can do that. Alternatively, some of the clients are looking at companies which are trading at very high multiples which are inconsistent with their long-term pattern given this AI hype. And what they're doing is if the stock is already at, let's say, 52-week high, they're selling out-of-the-money call options and buying protection, effectively costing them nothing.

Selling out-of-the-money call options means that you're gonna make a profit if the stock does not go above that level. You're gonna pocket the premium that you raise. And if the stock does skyrocket, you're gonna have to deliver your shares to someone. You're still gonna make some money,, but you won't make quite as much upside as you would have.

Yeah. In fact, what they're doing is the income they pull in by selling those options, they're buying protection with that. They're buying puts with that, right? So if the stock were to go down, effectively what they're saying is that this stock is at exuberant levels. So the probability of going up is lower than the probability of it going down in whatever the timeframe potentially. So that's just a hedging tool, for example. So I think it is prudent to think about what if CapEx slows, and, and hence it's prudent to protect yourself. And I think a very good example of this AI or dark fiber moment was during Deepseek. Remember Deepseek was just 6 months ago where the concern was that are we just overspending because the Chinese companies, Deepseek, could do it at a fractional cost to get similar performance. So what that led to was, one, a big sell-off in the hyperscalers, but it also led to a big sell-off in all the auxiliary beneficiaries in energy, industrials, and power. At the same time, interestingly, we saw a rotation into other more defensive names in the market, more rotation towards quality, towards value, and things like that. And in fact, 70% of S&P stocks were actually up during that massive sell-off, right? So that tells you that, you know, it gives you some sense of to think about where you want to seek protection. And so I think that's one way to think about it.

Thank you, Venut. Did everyone understand the option strategy he was talking about? I'm not gonna go into a whole thing on options. I'm sure I've done that in the past in this podcast. I'll just quickly say, and Alexis, you can play me off with Oscar music if this gets outta hand, you can spend a little money to place a downside bet on a stock or an index. That's called a put, and that could be a hedge. But if it doesn't pay off, you can lose the cost of the bet. And even though the cost might be small, if you're constantly placing and losing these bets to protect yourself from a fall in a stock or an index, it could subtract from your returns over time. One way to pay for those puts is to sell covered calls. When you do that, someone else wants to make a bet that your stock, a stock that you own, is going to go higher. You're selling a bet to that person. You put the proceeds in your pocket. If the stock does go higher, you're going to miss out on some of the upside. If it doesn't, you keep the cost of the bet. You use it to fund the purchase of your puts. Is that violins I hear in the background? All right, that's enough. Thank you all for listening. Alexis Moore is our producer. You can subscribe to the podcast on Apple Podcasts, Spotify, or wherever you listen. If you listen on Apple, write us a review. See you next week. Data is everywhere, but is it ready for consumption? Morningstar developed the language of global investment data, so you have the right ingredients to help you shine. Morningstar. Where data speaks.