Why the crazy spending of this AI boom isn’t like the dot-com bubble
If I hear one more hedge fund oligarch on television tell me that the data center buildout/artificial intelligence story is just like the dot-com bubble, I am going to tear someone else’s hair out because I don’t have enough to demonstrate my full frustration. Many of the people opining and whining about how this moment could be as bad or worse than the 2000 bust are uninformed or ahistorical. The buildout’s blockbuster numbers come from companies that can handle blockbuster capital projects without a problem. Has it been worth it so far? That’s the biggest stumbling block to processing what’s really happening right now. We look at this mega construction project with huge spending by companies, and we ask ourselves, “Is that all there is?” We can’t believe that all these Nvidia platforms loaded with hardware and software don’t do more than they currently do. I get that. It should be self-evident by now that AI is more than just an advanced code writer that can compile things on the hop. For example, we should have beaten some cancers. OpenAI CEO Sam Altman claimed AI can do it in a recent blog post entitled “Abundant Intelligence,” but noted the need for a massive increase in computing power. Many of us are becoming skeptical of the medical claims, especially when the pharma CEOs I talk to off the record are surprised that the people making these kinds of claims aren’t working with their companies. Given all the money that’s been spent, I would have rather written a check to Johnson & Johnson , which has the best cancer franchise. I know that Walmart CEO Doug McMillon said that “AI is going to change literally every job,” according to a story in The Wall Street Journal . That’s a strong statement, and we know that Walmart is a savvy user of data centers, especially those connected with Azure to compete with Amazon and everyone else. However, we can’t see it. As a customer of Walmart, I haven’t seen any difference at all. Have you? Sure, the article acknowledged that Walmart is writing code that might supersede other code, perhaps from outside vendors, but that’s hardly shocking unless you are a company like Salesforce , which keeps being rumored to be hurt by these internal code writers. Surely, there must be more to AI than companies saving on their Salesforce bills. Maybe Altman is right. It’s all on the come. We keep hearing, for instance, that Vera Rubin — the next iteration from Nvidia, a big leap from the current Grace Blackwell platform — will challenge us with reasoning powers that alleviate a lot of mistakes and can take the place of anyone in the entry level to five years of experience at auditing, accounting, and law firms. I want to believe it. But then again: On Friday, I needed a calculation about how well Costco’s stock has done versus the S & P 500 over 20 years, and I got a reasonable answer from one of the systems, but a different sum from another — somewhat close and very convincing — and I asked myself, “How in heck does anyone use these chatbots with any confidence?” They are wrong so often when it comes to stock performance that I would prefer working with high-performing high schoolers to get these numbers. I fall back on something that seems to get no traction but makes plenty of sense to me: If we think the buildout and AI are both busts, we have to believe that a sucker is born every minute and the sucker is Meta CEO Mark Zuckerberg. Or Tesla CEO Elon Musk. Or Google’s Sundar Pichai. Or Microsoft CEO Satya Nadella. What fools these mortals be! Really? A work in progress For a moment, let’s agree with Altman that AI is a work in progress. Let’s accept that OpenAI’s 700 million weekly users, four times the amount of last year, aren’t going there under duress. There is no AI gun to the head. To me, the use cases make a ton of sense. Nadella needs it to meet corporate demand, both for Azure and for the rapidly adopted Copilot. Chances are, you don’t use it because it isn’t ingrained in your thinking or your machine. But Microsoft isn’t lying about the use. Meta’s Zuckerberg uses it to power the best ad machine in the world, and for powering his smart glasses. OpenAI can’t meet the demand fast enough for teams using it for research. Google needs it to keep people going to Google and, by extension, Gemini, its AI assistant. Tesla needs it for autonomous driving and humanoids, arguably the two biggest markets in the world. Amazon has to have it, maybe even more of it than it has, if it wants to stay No. 1 in the online retail business and No. 1 in cloud computing through its AWS division — and has a hook-up with AI researcher Anthropic to do so. AI search engine Perplexity is an orphan with research traction. And Apple is the potential Big Kahuna that might negotiate a deal with any one of these AI companies where, as with Google, one of them pays Apple, say $50 billion a year, to be embedded in its operating system. That’s a small price considering Apple’s 1.5 billion iPhone users. After the Google antitrust decision, which found nothing wrong with the $20 billion rumored payment to Apple, why shouldn’t Apple be the biggest AI winner? Ah, you say, that’s all small potatoes compared to what you thought AI would bring you. And worse: The physical buildout and the need for more power, a secondary issue, is overwhelming the first in costs by leaps and bounds. Why does it look less like a bubble and more like an industrial revolution to me? It may be because I played a bit part in the 2000 dot-com burst, and I know how and why everything went kerflooey the way it did. The analogy does not hold up under scrutiny. Two bubbles Before we start deconstructing what happened back then, you have to understand that there was not one but two bubbles in the years leading up to 2001, when we knew the destruction of capital was obvious and deadly. The Nasdaq peaked in March 2000, but didn’t fully bottom out until late 2002. Let’s challenge each one. The first bubble was in connectivity. When I started thestreet.com in 1995, we were all dial-up. Then we managed to snag T-1 lines, paying thousands of dollars to the phone company to be sure we had enough to send an e-newspaper out to everyone. T-1s seem to go at lightning speed. At that moment, I was working with Starwave, a company owned by Paul Allen, late of Microsoft. I never met him. I did spend a lot of time with his excellent team. Others must have too because we got investments from The New York Times to run a joint newsroom, which at the time meant something. The Times was trying to learn; at least we knew enough from our first-mover status to teach. We were strictly a banner ad shop. But the Starwave guys came in and explained that if we got high-speed bandwidth, we could easily run commercials that looked like, well, commercials, and we could start making real money. I loved this aspect, as banner ads kept losing their value, and kept thinking we couldn’t get to video fast enough. We didn’t. Why? Because laying the fiber was no easy task and costs billions and billions of dollars. It wasn’t as big as the AI buildout, but it was costly and had to be done with debt. The companies involved had little to no pedigree. Or money. They just borrowed and borrowed and borrowed some more. They, meaning outfits like Worldcom and Global Crossing — Google or ChatGPT them — used Nortel and Lucent. There were lots of ancillary companies like Copper Mountain, which produced DSL equipment or internet working products like Wellfleet or Synoptics, which merged to form Bay Networks and gave you terrific switching. Cisco was the “backbone” of the internet, and its boxes were everywhere. These companies were so busy that you couldn’t get them on the phone. You were forever hopeful that you could get the boxes you needed, and as soon as you saw them, you bought the stocks of everything inside them. The internet build-out had a lot of fathers … and a lot of children. Of course, we know it didn’t end well. The cracks started right at the top. On Jan. 6, 2000, Lucent announced a gigantic shortfall, blaming bottlenecks and supply problems. Things were being put up so fast that many believed them. But in October of that year, Nortel, a much better operator, also announced that business had grown very weak. The orders were drying up. That was the one-two punch. The helium didn’t go out of the bubble all at once. But that hardware side was the most capital-intensive and used the most junk bonds and borrowed the most from vendors who eagerly advanced money, so it was a nightmare. Almost every company, save Cisco, that was in networking either blew up then and disappeared or hung on but kept losing money and then vanished. The plant and equipment bubble should have been spotted because of all the money that was borrowed. I got lucky. I knew Lucent well and point-blank didn’t believe it. I was short this side of the equation the whole year simply because the company couldn’t be trusted. I also had the advantage of watching thestreet.com’s stock go to $2 after opening at $63 less than a year before. That woke me up to the bogus nature of the operators using the web and producing ads and subscriptions. What’s the analogue here? Pretty simple. People want to compare CoreWeave , the AI cloud computing company, to these myriad buildout companies from the dot-com bubble. It’s a hardware and software infrastructure company, and it uses billions of dollars in debt to build and maintain the data centers. I would say that, judging by the orders it keeps getting — a surprise one almost weekly —it is doing what a half-dozen companies were doing in 2000. Given the lead CoreWeave has and the expertise that all the CEOs in this business, especially Nvidia’s Jensen Huang, trust, it makes sense to do what the company is doing. CoreWeave wants to be the only builder and operator in town, but it has to build out first, and for that, it needs debt — just like the players in 2000. But, you argue, the chips CoreWeave uses will be hopelessly out of date in five years, and it will be left holding the debt bag. But CoreWeave will tell you there are many use cases beyond five years, and the payback over five years will be enough anyway, especially because it will own the AI interstate highway and all of the rest stops on it. The bubble-ists think that the Coreweave people are hopeless dreamers at best and shysters at worst. The doubters must think there will be a domino effect when they go down. I believe the Corewave guys, led by tireless CEO Michael Intrator, a former bitcoin miner — there’s another credential the bears hate. They are really great at what they do, which is to manage these very complex and easily broken data centers. And I continue to believe that when the chips they use in various data centers are no longer powerful enough and are replaced by others in the Nvidia line, there will be a market for them, and they won’t be stuck with them. The obsolete won’t be obsoleted. For the pessimists, Coreweave is Lucent or Nortel. It’s the one that first says it can’t pay its bills. Now I want you to think about this: The not-much-talked-about hardware side blew a gigantic hole in the stock market in 2000 because the buildout’s capital intensity was only matched by a lack of clients willing to pay for it. Back then, we had “vendor financing.” The analogy would be if Nvidia were offering Meta money to take Vera Rubin and Meta took delivery of thousands of platforms, and there were no customers for the product. Now, unfortunately, until the recent Oracle announcement about its almost $500 billion in firm orders, mostly from OpenAI, we seemed to be on a decent, cash and carry basis, no debt for the majors. But this time we blew a hole in the thesis. Especially when Nvidia agreed to pay $100 billion for equity in OpenAI, over time. That sounded like vendor financing to everyone but Nvidia and its devotees, including me. I think Nvidia is making a bet that the stock market will reward OpenAI with a huge valuation and Nvidia will be even richer than it already is. We have to accept that the stock market will give OpenAI that kind of valuation. Why not? It is loved. A trillion seems a trifle, another thing we have to get used to. Nvidia’s Jensen Huang has been an unbelievably shrewd investor in many projects and companies, including Corewave, where he made the IPO happen with a gigantic investment that has tripled in value. The bears would say that he should just buy back stock, like Apple. The most cynical of pessimists, I believe, think that Corewave CEO Intrator is Bernie Ebbers, the crook behind Worldcom, and Altman is the embodiment of AOL, the king of the web, at least at its inception. The real indictment is that all of these hyperscalers can’t possibly make money, so they are wasting it, and when the bill comes due, it will be too great to bear. Or a waste of time. Or only one winner, like Google, and everyone else is Bing. The latter is the sub rosa nightmare come true. Now, let’s talk about the part of the bubble that’s remembered in a false narrative by clueless or motivated bears. I was intimately involved with this part of the equation and remember it well, because thestreet.com was financed by it. The public market drove the buildout. Here’s the background. In 1996, thestreet.com and a couple of other companies burst on the scene and had instant viewers, something I detail in “How to Make Money in Any Market,” my new book that goes on sale Tuesday. Our readers, who usually got to us via AOL, discovered outfits like E-Trade and started buying anything that even sounded like a dot-com stock. If anyone remembers the craziness of the MarketWatch IPO, that’s what really got the dot-com movement going and turned a lot of bankers’ heads. It was Jan. 15, 1999. We had a couple of IPOs that were tangentially involved with the web. Marketwatch was the first really in-your-face deal. It was priced at $17. It closed at $97.50. Retail investors drove it. They didn’t know how to buy. They put in market orders. They were bunched by a couple of companies, and the openings were controlled by the companies, especially a company called Knight Trading Group. They regularly fleeced the public with these market orders and were rumored to be short all of the high openings. With the public willing to finance anything, company after company came public, and the venture capitalists were making fortunes often on the deals and then on the follow-ons. Ultimately, there were about 330 companies that went public this way. They almost all blew up. There was a huge lawsuit against all the brokerage firms that got the deals. Suffice it to say, almost everyone had a hand in it. (thestreet.com survived, but never amounted to what I wanted it to be, as is self-evident.) The new class of investors was legion. They got it all financed. And they lost everything. They never came back. Who survived? Amazon, Yahoo, and Google, although the latter didn’t go public until August 2004. Let’s think of this: So many joke companies went public. Almost all failed. Billions of dollars lost. A generation wiped out. And a couple of winners that are winners today. What’s different this time Now we only have about a half-dozen companies that are trying to make it. To do so, you need a buildout that’s much bigger than it was back then, but so are the stakes. You want to be the next Google or Amazon, don’t you? The difference this time is that there are only a few who can do the buildout, namely Oracle and Coreweave, as well as Nvidia and other chipmakers that help out in the computation and communications process. Then there are the power companies, and the turbine builders—led by GE Vernova , a bunch of builders, including Jacobs , and lots of electric connectors, companies like Eaton and Emerson , and a couple of others. There are plenty of specs here, including anything nuclear, which has generated some spectacular runs. You should ring the register on these companies because most of them don’t have enough money to become what they want to be. So now let’s step back: In the dot-com era, a bunch of infrastructure and telco companies spent billions they didn’t have and built a fiber network that wasn’t needed. Not enough customers on either side. Now we have a few companies building out the infrastructure, which is expensive, but if the money isn’t spent by all the hyperscalers, then whoever doesn’t spend it will be left behind. Given the demand, can you afford not to? Coreweave is trying to dominate, but it will have to use some debt. Only Oracle is set to challenge Coreweave in terms of building and running data centers. It will need money, but it claims OpenAI will pay. I can’t dispute that because OpenAI will have a trillion-dollar IPO and can take that money and pay anyone for anything. Now, what makes this a bubble? It’s all self-funded, except Oracle’s bills and those of Coreweave. Oracle can get the money. Coreweave, with the help of Core Scientific , could be the only independent that comes out. The hyperscalers need Coreweave and Oracle, and the electric companies and the power they create just to protect their flanks. There’s too much business to be had. To me, it’s not a bubble. It’s not even a gold rush. It’s a belief that Jensen Huang’s innovations and all of its accoutrements will help these companies, which are already super rich, to keep their dominance. Bubble? How about a necessity? (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.
<
