The AI trade has left the hyperscalers in the dust. What will it take for that to change?
We are beginning to see the real weaknesses in these hyperscalers. Amazon , Alphabet , Microsoft and Meta Platforms may have the money, but they have run into a brick wall in this stock market. That brick wall is hardware. I should have realized how acute the shortage is when we saw how the memory chip stocks galloped to much higher stock prices. The shortage of chips, stemming from just a handful of players in the high-bandwidth memory (HBM) category — SK Hynix with roughly 60% share, followed by about 20% apiece for Samsung and Micron — is a bottleneck they have not been able to overcome. HBM is a specialized kind of dynamic random access memory (DRAM) crucial for AI computing. We know that because Apple had to own up to price increases , squeezed by the memory makers shifting more of their capacity to HBM from consumer-grade DRAM. The stocks of another class of memory chips — Sandisk , Western Digital and Seagate , which all focus on long-term data storage — seem to have no cap on them either. They are doing their best trying to deliver innovations . I wish they were expanding directly with new fabs, but that seems to be the wrong approach according to these storage companies. The opaque nature of the cost of these chips in a business-to-business context is undeniable. We can’t open up online chats to find out. It just seems like one big black box. Perhaps that’s why we didn’t know how onerous the penalties have become for these hyperscalers shelling out billions upon billions in capital expenditures. We do know that Microsoft and Meta both called out higher component pricing on their earnings calls as one factor behind their big capex numbers. All four hyperscalers have seen their stocks decline over the past month, while the tech-heavy Nasdaq is up almost 1%. A basket of memory stocks , meanwhile, has a one-month surge of 41%. Of the four hyperscalers, only Meta is almost entirely exposed to the consumer with an advertising model. This reliance on ad budgets severely restricts the company in the mind of the market. Meta needs a web services business , just like the other three have. If Meta had one, I could see its stock doubling and many — including us — are hanging on for one. Having a cloud business would make it easier for Meta to show a clear return on investment from all its AI capex. Meta is down 12.55% year to date. MU YTD mountain Micron’s year-to-date stock performance. I had thought the HBM tightness would be alleviated by additional chip fabrication plants, known as fabs. But they can’t get online fast enough, or they can’t get more out of their existing machines quick enough. That’s because the real intellectual property in this supply chain is not the hyperscalers, nor the memory chipmakers, but the capital equipment companies: Applied Materials , Lam Research and KLA Corp . We need more of what they produce, but we aren’t going to get it in time to sort out which hyperscalers can win. The capital equipment companies are all dual and triple oriented, which is why they are regarded as more dangerous than a Micron or a Sandisk. I think that’s probably false, though, because Applied Materials CEO Gary Dickerson told me last month that the company has “unprecedented visibility” from customers because demand is so strong, so I don’t think they are going to have shortfalls versus Wall Street estimates any time soon. This whole memory complex has thrown a monkey wrench into the hyperscalers’ growth plans. It is no longer whether they have enough Nvidia chips, as we saw at earlier stages of the AI boom. The hyperscalers have tried to tackle the Nvidia stranglehold by teaming up with Marvell Technology and Broadcom to co-design custom AI chips. We picked Broadcom to run with, and we’re nearly three years into our ownership. Still, I can’t believe how high Marvell has gotten this year, with shares more than tripling. It’s interesting to see Nvidia CEO Jensen Huang embracing Marvell, first with a $2 billion investment in March and then, earlier this month, calling it the next trillion company . Why is it so curious? Because Marvell is working with Amazon to defeat Nvidia and build its own semiconductor business. Already, Amazon says that if its chip business was a standalone entity, it would have a $50 billion annual revenue run rate . It’s incredible to see how the stock of Broadcom has collapsed, even as it continues to work with Alphabet’s Google to try to break the Nvidia stranglehold. While I don’t think there’s a weakness in that partnership, Broadcom’s last conference call confounded us. The stock was at $479 a share before earnings on June 3. It’s clawed back some of its 22% post-earnings slide over a few sessions, but still ended Thursday at $411. Thankfully, we booked some profits on June 2 at roughly $480 because the stock had a parabolic move into earnings, and my long-held discipline tells me to trim parabolic moves. Believe me, as a money manager, I wish I had bet on Seagate, Western Digital and Sandisk, or even an exchange-traded fund tracking the Korean market, where Samsung and SK Hynix are by far the biggest names. Woulda, shoulda, coulda. SK Hynix is planning to list in New York , in a bid to broaden its investor base and boost its overall profile, but the stock has had such a run it feels foolish to chase. Then again, when you consider the passion-versus-rigor argument I laid out Wednesday during our Monthly Meeting, you can buy it. I also regret not owning Applied Materials or Lam Research, as we had them on “Mad Money” and they are extraordinary companies. Of course, buying Arm Holdings this year has been a total home run for us. But with these names in the memory and semi-cap equipment space, we were hemmed in by our decision to embrace the underperforming hyperscalers. They are underperforming precisely because of these shortages and the cost of Nvidia and, most importantly, materials, labor, and siting . It’s becoming obvious to me and everyone that, right now, these companies, plus Nvidia, are just in too much of a battle to figure out who is going to win. In the meantime, the memory-and-storage semis, which used to be considered total commodities, are now different enough that they can’t be matched to drive down prices. Despite their runs, they are probably better buys than the hyperscalers … for now. I feel the same way about Club names Corning and Qnity Electronics , two around-the-edges winners of the AI trade. Corning’s fiber is coveted inside data centers for its data transfer speeds. Qnity’s specialty materials are essential to the making and packaging of chips. Both stocks have more than doubled this year. If I had my druthers, I would pick one or even two more of these semi companies we don’t have in the portfolio, but that would involve selling Salesforce , which we want to give one more quarter to, and perhaps Microsoft , which is being assassinated by sellers who think that as much as 50% of their business can be disrupted. Here, I’m talking about Microsoft’s enterprise software business and its reliance on a seat-based model. It’s the same overhang on Salesforce. Stay or go? So, the question becomes why stay with the hyperscalers? There’s a couple of reasons. One of these, or maybe two, are going to blink on AI spending. That will cause the other two to roar. I don’t think Alphabet will blink because it just raised money and it has a prized partner in Apple, which is using Google’s Gemini models to give Siri a much-needed infusion of AI capabilities. Microsoft badly needs to merge with OpenAI. I am beginning to believe it is the only way out for Microsoft, as far-fetched as this idea may seem. Meta needs to build a cloud business, or it becomes pretty much irrelevant in this competition. Amazon is too competitive and won’t stop spending. Anthropic, assuming it follows through on plans to go public , will be a huge company filled with hubris, which would normally cause a downfall. It hasn’t yet. That makes for a battle among Amazon, Alphabet, Anthropic and OpenAI. Four. Anthropic will have a similar war chest to Alphabet. OpenAI will need far more capital than it will get from an initial public offering because it has more of a consumer-focused revenue base compared with Anthropic, which is heavily skewed toward business-to-business. OpenAI wants to get closer to a 50-50 revenue split , but for now, Anthropic is more loved by the investor class. The key to Amazon is to get its AI business to profitability as soon as possible, something they think can happen next year. Of course, the missing link for all of these companies is precisely that: profitability. It will be the capital markets that determine which of the four will win, as I don’t think we need them all. During this period, I no longer see how any of these companies’ stocks can outperform the suppliers, which is the intent of this piece. I am laying the ground for the jettisoning of a couple of techs to pick up something in the incredibly lucrative food chain when we get some sort of general market decline caused by the president’s erratic behavior. I probably won’t have long to wait. It’s tough to admit that you are on the wrong horses, even as I thought the Broadcoms, Cornings and Qnitys were enough. That’s what made the recent Broadcom decline so painful. Any one of these —Sandisk, Seagate, Micron, Applied Materials, Lam or KLA, and Marvell — were superior to what we own. We need to reposition, and we need to do it shortly given that the hyperscalers are still spending like mad —or until two of them blink and then the pace of building will slow down. I am betting that this slowdown will not occur for sometime, though, because of the ease with which the stock market absorbed the Alphabet stock sale and SpaceX’s IPO. As you know, I was very concerned that there wouldn’t be enough demand for this new supply of stock. I don’t know if SpaceX was sui generis — there was money for a special situation such as another Elon Musk venture to own — or that we are more endowed with cash than I thought. I think the coming Nasdaq 100 admission for SpaceX will be an interesting tell and still one more millstone around Nvidia’s neck. The good news is that right now, only the looming IPOs of Anthropic, which will be wildly oversubscribed, and OpenAI, which will be far more tepid, stand in the way of the next run for the market, as the coffers seem to be replenishing. Why am I not worried about the impact of a potentially more hawkish Federal Reserve Chairman Kevin Warsh in all this? Don’t interest rates determine things? I think they have helped determine the prices of Nvidia, Amazon, Microsoft, Google and Apple. But not as much as the others mentioned here, all of which are about shortages and price increases. The biggest one to worry about is the one I have the most allegiance to: Nvidia. If the stock of Nvidia is going to go higher, it will have to adopt the way of Apple , buying back its own stock by the fistful, as there is just too much supply. I think a hedge fund would abandon Nvidia, Microsoft, Alphabet, Amazon, Meta or even short their stocks on any lift. Just putting it out there. A trust stays the course. (See here for a full list of the stocks in Jim Cramer’s Charitable Trust.) 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.