Forget the bubble talk — why you need to own AI stocks in 2026
I am tired of the lies, exaggerations, and dispersions that are constantly told in the assault on the stocks of companies tied to generative artificial intelligence and accelerated computing: They will dominate in 2026, just as they have since ChatGPT came on the scene four years ago. I want to straighten out some of these misconceptions — many of them deliberate — and share my view on the tech setup for the new year. First, we have to cut it out with the “bubble” talk. Can we admit, for a moment, that it’s been harmful? It left many investors out of a significant share of the performance, and that matters. Performance avoidance is performance lost. But those market watchers and analysts who screamed of a bubble are unchecked in their touting. If they were to put their money where their mouths were, they would be struggling to pay the rent. Second, age has so few benefits. One upside, however, is the experience of different markets. I started a moderately successful internet business, with thestreet.com, and ran a hedge fund at the same time, one that was up 36% in 2000 from being short the market. And can I say that as much as I like reading about those who did well before — “Big Short” investor Michael Burry, for instance — my advantage is that I lived through it, invested in it, and made big money off both the inflating of and the deflating of the bubble. It was all circumstantial, but it was relevant. Most of the negative stuff I read is irrelevant. Third, as I spelled out in “How to Make Money in Any Market, ” there are winners and losers in a boom-bust scenario, so stop thinking everything’s a loser and start identifying companies that can win in either environment. They exist; they will make you fortunes. Now, why am I not as worried as others that 2026 will be a no-invest-zone for tech, considering that all tech is, in some way, shape, or form, about AI? I have to start with a luxury: I read the Eye on the Market research that’s written by the finest mind on Wall Street: Michael Cembalest, chairman of market and investment strategy at J.P. Morgan Asset & Wealth Management, who also happens to be a great friend of mine. (You may have seen pictures of our recent trip to Trinidad to fish for tarpon in my X handle.) Michael has thought more about the bubble concept — he called in The Blob back in early fall — than anyone else. He is not indiscriminate like so many. He understands that the issues about AI have much to do with adoption and infrastructure than with the overall “craze” that so many others seem to cling to and live by. He questions how much business people really use it away from tech, where it does help write software. He questions the capital allocations big companies are making. Most importantly, though, he questions whether there is enough electric power to make it all come together. Power gating. That’s what might matter. I go a step further: I now believe, after extensive time analyzing GE Vernova , including interviewing CEO Scott Strazik, that there isn’t enough power generation to even keep a bubble growing, let alone pop it. We don’t have enough electricity to allow hyperscalers to grow as they want. To put it another way: there isn’t enough air to inflate a bubble, so how can it be popped? The companies that can adapt to this world and accept the constraints of limited power can do well. Hence, the admission that we can’t live without owning Alphabet , which has the most going away from its AI business to handle a slower, power-constrained build-out and still thrive. (We initiated a position last week.) It is the best-positioned hyperscaler, with the best, most accurate site, Gemini 3, and the best business model, thanks to the ingenious way it dovetails with Google. Gemini3 will soon cement its dominance by cutting a deal with Apple to be its sole source of AI content. That’s 1.5 billion users. What will the others, including OpenAI, have? OpenAI seems to have gotten enough seed money from SoftBank , Microsoft , Nvidia , and others that it can come to market with a gigantic IPO that will be prohibitive to the rest of the S & P 500 — it will be admitted automatically I presume — because the money to buy the stock will come out, heavily, from index funds that will have to buy it. Ready-made cash that they seem somewhat clueless about. The good news here is that OpenAI will be subject to the same power gating as everyone else. There are only three companies that make the giant turbines that turn the abundant fuel of natural gas to electricity: Siemens , Mitsubishi, and GE Vernova, and they are all booked up through 2030. If they could add more turbines quickly, I would be more worried. If the hyperscalers were to accept more dirty fossil fuels, like “clean” coal, I would be concerned. If someone were to come up with a less energy-intensive way to produce compute, I would be very nervous. They haven’t and they won’t. To these ends, I remain unenamored by nuclear, which is a sideshow, a down portion of the card in an evening of fights, one you shouldn’t factor in, even as those who are less clued in remain buyers. Nuclear power is part of last year’s Magical Year of Investing, which includes speculative stocks in quantum, nuclear, data center extensions, and others that soared in 2025 but fell back to Earth by year’s end. When things are bad, my friend Michael from J.P. Morgan likes to talk about what can go right. When things are good, he tends to talk about what can go wrong. Given that things went well, Michael spends a lot of time talking about Taiwan’s vulnerability to China. Consider Taiwan Semi ‘s leadership role in AI. Duly noted. Again, though, you know my hardline on the People’s Republic of China. (Michael does agree with Nvidia CEO Jensen Huang’s theory that we may have poked the Chinese bear into having its own competitive AI ecosystem by prohibiting Nvidia from selling chips to the Chinese.) I want to shade things differently. To do that, you have to go back to Jensen’s core principles, which seem to be totally forgotten and need to be refreshed, so you know why Nvidia is the world’s biggest company. The centrality of Nvidia remains a mystery to all but those who have actually studied the company. My identifying Nvidia at $2 a share and almost force-feeding it to Club members — many of them now Nvidia millionaires — gives me a clearer head than most about the chipmaker’s fundamentals. So let’s go there. First of all, Nvidia is a hardware and software company. The latter, which can’t be duplicated, is what makes it impossible to speak of Advanced Micro Devices in the same breath. It makes it impossible to compare Alphabet’s chips to Nvidia’s, either. Or Amazon’s chips, for that matter. There is no real benchmarking here: they can’t do what Nvidia’s chips can when loaded with Nvidia software. Nvidia is not just an AI company. People keep either misunderstanding or forgetting that there are not one but two revolutions going on here, the second being accelerated computing. The reason Jensen is so sure of the trillions of dollars in total addressable market (TAM) when it comes to Nvidia chips has much more to do with the speed of his GPU semiconductor form factor versus the CPU form factor of AMD/Intel and how the difference is so fundamental, so palpable that literally all of the old compute must be replaced. You scrap the entire computer system and replace it with Nvidia, and you have a powerful reason why you have a $4 trillion company — without AI . Hence, my “own it, don’t trade it” stance for 2026. If the PRC takes over Taiwan by force, my thesis is a loser. A new caveat for 2026? Now, it’s the second part of Nvidia’s greatness that truly calls the bears’ negative thesis into question. What nobody ever seems to recognize is that it’s Jensen’s world and we all play in it. The new chip, the Vera Rubin, is the first Nvidia chip that will actually allow users to reason with their computers. Rubin is a reasoning chip. Remember that Jensen believes that right now, only 90% of what the chatbots produce is accurate, which isn’t enough to take seriously in many scenarios. He accepts that until hallucinations can be curtailed dramatically, it won’t be nearly as useful as it needs to be to be commercially successful. It makes too many mistakes. But that will change, and it will change faster with Vera Rubin than it does now. Yes, his iterations are that meaningful. He has a clear roadmap for the future that includes the Richard Feynman, named after the great 20th-century physicist. What you need to remember is that Jensen has a unique way of thinking. He tries — and succeeds — by looking out 20 years and then working backward to figure out how to bring us there. You think it’s not true? In the late 90s, he shared his current vision with Intel, and the legendary Andy Grove laughed him out of the Intel building. One of my late friends, who was on the board of Intel, actually heard the outline of Jensen’s presentation and couldn’t believe how Andy railed against Jensen even as Jensen proved to be totally right and Andy totally wrong. Twilight of the Idols! If we just think about Nvidia as static, it probably does appear to be the greatest short story ever told. But Nvidia isn’t static, and Jensen’s busy adjusting to a power-constrained world. (So are Corning and Broadcom, which are working to come up with a more fiber-based copper-free chip that produces less heat and needs less electricity.) The new Nvidia chips will make a “quantum” leap over the current iterations, not obsoleting the old ones — although many will say they will make the depreciation schedules more obvious and shorter — but making AI much more successful. It will be better in a host of questionable verticals: legal, health care, sales (agentics), and manufacturing. We will not need to be as skeptical. It can make AI a necessity where it is, right now: a possibility with no clear ROI in most industries. Static thinking, however, dominates the bears’ rationalizing. Their reasoning, if it is that, goes away if Nvidia makes chips that can allow robots to adjust and pivot to most takes. Their bearishness leaves them behind if we can use Nvidia chips to learn faster, and not just compile or sift, exclude, and include, but reach conclusions quickly and better than we can now. Think of it like this. There is a work problem. We can game it out, figure it out, argue over it, and produce a result that, over many weeks’ time and many millions of dollars of manhours, is almost surely inferior to the problem-solving of a Vera Rubin-led effort. In any industry. Is that the stuff that you want to get short? Don’t you want hyperscalers, plus Eaton , GE Vernova, and Broadcom in that world? I would be frightened not to own them. I like Nike . I think Capital One is awesome. It could be Boeing’s year to have an explosion of cash flow, the only real metric that mattered and made the sellers in the $180s to be revealed as first-class bozos. You have to think that Eli Lilly ‘s GLP-1 pill will be extraordinary. Who doesn’t want to be in Goldman Sachs in an era of financial liberation? Yes, Brian Niccol will turn Starbucks in 2026, and a weak dollar and new leadership will vault Procter & Gamble to the top of the consumer packaged goods sector. But you leave this world of accelerated computing and generative AI when power gating constrains the winners from spending more quickly? You abandon the winners? The Amazons and the not-yet-public Googles of the year 2000? Go ahead? Make the shorts’ day. So I say, stay skeptical about valuations, but stay skeptical of their critics, too. That’s because they are looking at static valuations — and higher price-to-earnings ratios on those static valuations — and are doomed to critical-reasoning failure. Understand that you have to maintain the same leap-of-faith analysis that I did about Jensen’s moves when his stock was at two bucks. Yes, accept the centrality of Nvidia’s role in the market. And get ready for a steadier, less capital-intensive level of “reasoning” that will make the current business-to-consumer (B2C) chatbot businesses more integral to the business world. Only Anthropic is business-to-business (B2B) and will be successful no matter what, given that status. There’s a reason why I identify these tech growth stocks as part of “How to Make Money in Any Market”: self-improvement. They get better and better by nature and DNA. They are, therefore, not frightening or fictional. They are factually better able to handle the future than almost all other companies. They worked in 2025. They will work again in 2026. (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.
<
