Why it’s not too late to start buying red-hot AI stocks
Late. Late. Late. That’s how I felt when I saw the Roundhill Memory ETF — symbol DRAM, of course — take in more than $5 billion in a month. Super late when I saw it took in $1.1 billion on Thursday alone, all to play what the press is hailing as the memory supercycle. But then I look at its core holdings like SK Hynix, Micron , Samsung, Sandisk , Seagate , Kioxia and Western Digital , I say to myself this is a cocktail of high octane the world has never seen before. And a fantastic way to get exposure to the makers of memory all over the globe, including some not listed on American exchanges. The Korean companies are on fire. The Japanese firms are on fire. I want in. That billion that was invested on Thursday, maybe that was the smart money. Rather than late, I think that if you own this ETF you have everything that represents the incredible bottleneck of ever-rising prices where the customers can’t say no thanks to the demands of AI computing. How can you say no to those kinds of investments? How do you say no to the insatiable? Late. Late. Late. I am reading my old friend Herb Greenberg’s Substack newsletter, Red Flag Alerts, and what does he focus on this week? Modine Manufacturing . I know it is a smoking-hot company from its gorgeous stock chart and I know it has a tremendous reputation in heating and cooling. Modine says it has built really strong relationships with the hyperscalers. But Herb talks about how it hasn’t been exactly straightforward with its transformation and how risky it’s become. It does seem a little headlong. But the more I look at it, the more I think it hasn’t really been discovered yet. I dig deeper and I can’t believe how much these hyperscalers seem to love the company, even as Herb says that they might not even have any of these hyperscalers as clients. I’ll take the other side of the trade. Modine Manufacturing is a buy. It sounds as good as Vertiv or Club name Eaton , two data center suppliers, except a little less discovered. Certainly better than Carrier , which has too much residential HVAC exposure. Plus, Modine got rid of all the slow-growing industrial. What a management team. I am grateful to Herb for tipping me off, even as the chart says I am not early. Late. Late. Late. Micron rallies over 200 points in a week without me, going from $542 a share to $747. Then again, it doesn’t matter what kind of company it is, or how special its high-bandwidth memory (HBM) chips are. No company can have a stock that goes up that much in such a short period of time. It’s absurd. But then I remember Micron is trading at 9 times the next 12 months worth of earnings, according to FactSet, and it may turn out to be even lower given how much the company can raise prices. So, you are buying one of the cheapest quality stocks in the universe. It’s like when Melius Research’s Ben Reitzes was bold enough to initiative coverage of Micron and Sandisk with buy recommendations in an April 27 note, one of the latest attempts from Wall Street to capture some memory upside. It turns out that you caught a 50% gain in Micron in just a couple weeks if you listened to Reitzes. It wasn’t the latest, as in, “are you kidding me, you want to come in this late?” It was the latest, as in, “the latest, greatest way to play memory.” It wasn’t rash. It was conscientious. And it’s just all too tantalizing. You can’t read about these companies without wanting to buy them. I spent a lot of time with Director of Portfolio Analysis Jeff Marks this week, just kicking around trying to get into Micron if we just had a couple of down days. And why not? We know the demand is there. We know it’s an easier way to play this supply-demand imbalance than buying the stocks of Qnity or Element Solutions . Or maybe not? Those are sleeper cell memory plays. They make the materials you must have to turn silicon wafers into chips. At least we own Qnity by virtue of fellow Club name Dupont spinning it off in the fall. Late. Late. Late. We are now buying the tertiary plays, the incidentals, because the gains that have been made already by memory, by cooling and by construction — why not Nucor ? — are way too far gone and known and we have never seen any profits like this before. Ever. But then we think of the price-to-earnings multiples. They are still so small in many, many cases. We think of things as stupid as people selling Taiwan Semi on Friday because Apple is going to do business with Intel , even as it was well-known Apple was considering using Intel’s foundry in some capacity. I even asked outgoing Apple CEO Tim Cook about it a couple of weeks ago for heaven’s sakes. There was nothing revelatory here. The winner was actually not Intel , which we figured would get this business, but Taiwan Semi. Why? Taiwan Semi can take the production capacity it was going to allocate to Apple at rock-bottom prices — because Apple negotiates so hard — and give it to someone else for much, much higher prices. It’s time to raise numbers Taiwan Semi. The chance to use the discount to buy lasts for just a few minutes. Late. Late. Late. Did you see the “collapse” in CoreWeave’s stock Friday after it reported on Thursday night ? Sure, it had record revenues but it’s not making any money and shouldn’t it be? When is all its borrowing going to generate profits? But then we have CEO Michael Intrator on “Squawk on the Street” Friday, and he points out his earlier tranches of debt were rated junk and now his bonds ratings have gone much higher and he’s borrowing at around 6%. How can he not take advantage of the “free” money he can get to put up the best AI data centers, the ones Nvidia CEO Jensen Huang swears by? How can CoreWeave, the best builder of data centers, not build as many as possible? He certainly has the expertise as a former bitcoin miner and a financier who is everyone’s favorite site builder in the industry. These are complex buildings, billion-dollar works, and there is a finite number of builders, many of which secured investments from Nvidia. I admit to never having bought a stock because the company’s bond rating has improved. But when you think of the opportunity, maybe, just maybe it makes sense to do exactly what Intrator is doing. Jensen has helped Intrator tremendously, including saving CoreWeave’s initial public offering last year with a $250 million anchor investment when it looked like the deal would have to be pulled. At the time, Nvidia already owned about 6% of CoreWeave, a so-called neocloud. Now it’s up almost 13%, according to FactSet data. That includes another $2 billion investment in January of this year. Jensen is up big on CoreWeave, as he is on so many other investments . They are joined at the hip. Nvidia has also helped neoclouds Nebius and Iren — two fringe companies that he’s made non-fringe with a couple of investments. Nvidia participated in Nebius’ $700 million round in 2024 and then invested $2 billion in March . Nvidia’s deal with Iren announced Thursday : a five-year warrant for up to 30 million shares, which amounts to a $2.1 billion investment. Plus, a $3.4 billion cloud contract for Iren to manage GPU cloud services. I met the Nebius folks two years ago at Nvidia’s annual GTC event, a showcase for all things AI in San Jose, California. It was an incredible moment because I went to their booth because Jensen said I ought to swing by and say hello to them. That’s just what I did. I felt bad for them. Nobody else wanted to spend time with them. Who knew they could turn out to be so important? I don’t feel bad for them now. All three companies saw their stocks take off because of these deals. But they haven’t been uniformly stellar performers of late. Maybe because it’s not as special if all three got a deal? Perhaps because the market has fallen in love with central processing units (CPUs) made by Intel and out of love with graphics processing units (GPUs) made by Nvidia. The fickle nature of buyers now makes the whole move feel very late, too. But as you can tell from my confusion, for every data point that makes me feel that it’s late—including the inclusion of the curse word supercycle in the lexicon— I think of a Nebius or Iren just sitting there ripe for the pre-Nvidia buying. I am conscious that I am running a Charitable Trust, and these companies feel like hedge fund “names” with a finite shelf life. You buy them and you don’t forget them. But then again, how do you know they aren’t the next Vertiv, which came public through a Goldman Sachs blank-check company in 2020 to little fanfare with an enterprise value of $5.3 billion ? The power and cooling company has market capitalization of $130 billion now. You just don’t know. VRT .SPX 5Y mountain Vertiv’s stock performance over the past five years versus the S & P 500. And that’s the real problem. Look at it from my point of view. We run a charitable trust where I like to go with established companies that hopefully don’t have a shelf life. They have a long life. I have been trying to select companies with an eye toward value, not high growth because we have enough high growth. But what happens when I do so? I end up starting positions in Johnson & Johnson in the upper $230s and Cardinal Health in the upper $220s. For now, that’s been so wrong, so painful. Yet both had been among the most reliable of all companies for so long. They are just solid growth stories, albeit out of favor right now and getting even more out of favor. The same way that Becton Dickinson and Abbott Labs have gone out of favor. The irony should not be lost on you. J & J and Abbott are quintessentially safe stocks that have gotten dangerous. Micron, Sandisk and Nebius are supposed to be the high-risk names and they turn out to be high-growth stocks that seem to have no top. As someone who runs a public portfolio, where all our moves under scrutiny, I don’t want to buy a Micron or a Western Digital because I remember how they have ended up before. Disastrously. For so long, memory stocks were the ultimate cyclicals. But now I see them on the “right” days as nothing more than magnificent longs. So you try to compromise. Maybe you should buy that DRAM ETF? Maybe I should buy some Micron. Maybe the lack of a split makes it seem like it’s all a little too much. It’s an uphill battle to even convince many, if not most, people that the data center at this point is an investible thesis. The stocks have moved so much. There is too much money chasing too small a profit at best and a huge amount of losses at present. Bottom line Let me leave you with this. A week ago I was at Amazon for an interview with CEO Andy Jassy . I met with people who know far more than I, a generalist, know about what can happen with the data center. Who knows how much compute Anthropic and Open AI need? Who knows what third major player will emerge needing the same parts and providers? Can you imagine that? All I can tell you is these very bright people are not inexperienced kids hoping to make a big score. These are longtime warriors who have examined all the parameters and know the market better than any hedge-fund manager. They understand what will be needed over multiple years and they know what the market wants. How can I disagree with them? Who am I to disagree with them? So when I hear the doubters, those who see the now empty cash flows and the hundreds of billions that “must” be spent, I wonder what they see? Do they talk to Jassy? Do they talk to the people he surrounds himself with? Do they know as much as Jassy’s crew? Do they know how early, early, early Jassy thinks he is? Do they know Jassy sees the profits starting to flow as soon as next year? Do they understand the short and long games that Amazon is playing? Do they know Amazon doesn’t want to be Bing’d by Microsoft? It doesn’t want to be AOL to Alphabet’s Google. Maybe I have been won over by those who measure the time to win over a horizon that will be decades and we are in the second year of the first decade. Whatever, I sure as heck wished I had bought Micron over Cardinal or Sandisk over J & J. It’s hard right now to even imagine the data center stocks going down knowing the budgets these hyperscalers have and how much they expect to make starting next year with their data centers. None of the so-called really smart investors I speak to and watch on television are talking about the profits that are about to be made. They are talking about the losses being generated now. I have presented the two sides of the same story in barebones fashion. In the end, I come out with Jassy. Got to go buy some Micron. (Jim Cramer’s Charitable Trust is long NVDA, CAH, JNJ, MSFT, AMZN, GOOGL, AAPL, ETN, DD and Q. See here for a full list of the stocks.) As a subscriber to the CNBC Investing Club with Jim Cramer, you will receive a trade alert before Jim makes a trade. Jim waits 45 minutes after sending a trade alert before buying or selling a stock in his charitable trust’s portfolio. If Jim has talked about a stock on CNBC TV, he waits 72 hours after issuing the trade alert before executing the trade. THE ABOVE INVESTING CLUB INFORMATION IS SUBJECT TO OUR TERMS AND CONDITIONS AND PRIVACY POLICY , TOGETHER WITH OUR DISCLAIMER . NO FIDUCIARY OBLIGATION OR DUTY EXISTS, OR IS CREATED, BY VIRTUE OF YOUR RECEIPT OF ANY INFORMATION PROVIDED IN CONNECTION WITH THE INVESTING CLUB. NO SPECIFIC OUTCOME OR PROFIT IS GUARANTEED.
