Micron CEO Sanjay Mehrotra did plenty to quiet the bears. That’s our conclusion from Jim Cramer’s extended interview with Mehrotra, which aired Thursday night on “Mad Money.” Mehrotra’s pioneering career in the volatility-prone memory industry spans more than four decades, beginning at Intel in the early days of the personal computing era, before going on to co-found SanDisk in the late 1980s. Now, Mehrotra and Micron are at the center of a pivotal time for the industry, thanks to a dramatic supply-and-demand imbalance brought on by the AI boom. “Memory has become a key enabler of AI,” Mehrotra told Jim on Thursday. “Memory is the intelligence behind artificial intelligence.” Micron and its main rivals, Korea’s SK Hynix and Samsung, have been rewarded with soaring profits and stock prices, despite a recent pullback for the group. All three companies are investing billions to build more capacity to meet the demand — Jim and Mehrotra spoke on a sprawling construction site in Boise, Idaho, where Micron is building two fabrication plants — but there’s a multiyear lag between shovels in the ground and chips in customers’ hands. It is a memory cycle for the ages. But how long will it last? Indeed, the single greatest question on Wall Street is regarding the durability of the cycle. It’s why Micron’s stock trades at less than 7 times forward earnings. The bears believe history is on their side, arguing the current boom will be followed by a bust of epic proportions. They will also tell you that the four most dangerous words in the English language are “this time is different,” even if this stretch of unprecedented demand is leading to deeper relationships between the memory producers and their customers. The counterargument, of course, is that if you can actually figure out when it really is different, you stand to make a boatload of money. Time will tell who is right, but after hearing Mehrotra speak on the dynamics of the current cycle, this is the moment to bet on things actually being different. We certainly have no regrets about starting a position in Micron earlier this month. What it comes down to — as is almost always the case — is supply and demand. Jim’s interview covered a lot of ground, and listening to the whole thing would be a worthwhile endeavor, offering important insight into not only Micron, but the broader AI buildout. The full interview is available at the end of this story. Now, here are our three important takeaways for long-term investors as they look to filter out the bear calls trying to scare everyone out of the AI trade. Cycle durability The AI data center buildout is currently the main driver of memory demand. But Mehrotra explained that there’s more to the story; demand is coming from everywhere. He’s got a point. While large AI models are trained in the data center and most of the day-to-day model usage known as inference, also occurs in data centers, we are seeing efforts to push the processing power to “the edge.” That means onto the devices consumers are actually interacting with. As these devices are tasked with more AI-heavy tasks and other advanced applications, they too need more memory. Here’s what Mehrotra told Jim: “You know, today there is no AI without memory. AI systems need more memory. They need higher performance memory. They need lower power memory. So, the value of memory, that equation has totally changed. Our customers, to drive their own growth, they need more compute. They need more memory. So, memory really has become a key enabler. … This is not only in data centers. Even in your phone, you know, to have richer experiences in AI-enabled phone, you need more memory content. Your PCs, your self-driving cars, all of them now need more and more memory content. So, the need for memory is ever greater, and now in AI, it’s not just the devices and the servers. Now you have agentic AI, agentic. It’s all the agents, all the context windows that are getting larger. That’s all. They fill with more memory as well.” Consider that the original iPhone came with 128 megabytes (MB) of dynamic random-access memory, known as DRAM. Today’s models come with 8 to 12 gigabytes (GB). There are 1,024 megabytes per gigabyte. Similar dynamics can be seen throughout the consumer device space. The leading AI chips from Nvidia, AMD and Google use a specialized form of DRAM called high-bandwidth memory (HBM). In simplistic terms, HBM is a bunch of individual DRAM wafers stacked on top of each other, and Micron, SK Hynix and Samsung are shifting more of their capacity toward HBM. That’s one reason why there’s such an acute memory shortage. AI chips, smartphones, laptops and other devices are all fighting over the same basic DRAM real estate. Each new iteration of technology needing more memory isn’t a new idea. But it’s being supercharged by these devices gaining AI capabilities and demand pouring in from relatively new sources. Vehicles are one such example, as the automotive industry moves away from analog internal combustion engines toward more digital, electrified offerings. The move to increased levels of autonomy is also a compounding factor driving demand. In high-stakes applications like autonomous vehicles, latency is quite literally the difference between life and death. So, you need to ensure all the inference needed to get from point A to B can be done on the vehicle itself (at the edge), as there can be no tolerance for connectivity issues or latency. That means more and more memory needed as the cars become increasingly autonomous. At another point in Thursday’s interview, Mehrotra mentioned robotics as another source of demand further down the road. We’ve likely not even started to scratch the surface on what demand there looks like, considering robots are the ultimate edge computing device. Like with self-driving cars, robots will also need to be able absorb and process massive amounts of multi-modal inputs, all while having access to relevant contextual information to make the right decision. When it comes to AI, the common theme here is context. After all, the correct answer to a question or the proper solution to a task is going to depend on the context. As tasks become more complex, that context expands. Memory is what allows it to expand. That’s why it is the key enabling technology to more advanced AI offerings. That brings us to the supply side. You can have all the demand in the world for a product, but the amount of available supply is going to determine the sale price. That is the big concern when evaluating the durability of this cycle. Historically, the moment supply starts to exceed demand — even one chip too many — pricing power has cratered. So, the question investors keep asking is: When will supply catch up? According to Mehrotra, that won’t be any time soon, even with Micron building the two Boise fabs and another complex in Clay, New York, near Syracuse. The company held a groundbreaking ceremony in Clay in January of this year. Micron plans to invest $250 billion in the U.S. through 2035, with up to $6.2 billion coming from money received from the Biden-era CHIPS Act. “Memory has secular demand with AI, across data center, across consumer markets, and, of course, industrial automation and, in future, robotics,” Mehrotra said. “By end of the decade, each of these robots — there will be, over time, hundreds of millions of these robots — they all require tons of memory. So, memory demand is tremendous. Supply is constrained. We see 2027 even tighter with respect to 2026. What we are building here [at the first Boise fab], first wafers out in 2027, production ramping in 2028 timeframe. Clay, New York, production ramping, by 2029 to 2030 timeframe. The second Boise Fab will start ramping up production in late ’28 timeframe.” It also takes Micron competitors Samsung and SK Hynix multiple years to bring new fabs online. The thing to understand, is that this isn’t like past cycles, where you maybe just need to install a few more machines and pump more chips out of the same factories. What we need now is brand-new factories, and a lot of them. As a result, the risk of an oversupply in the chip space is mitigated simply by how long it takes to bring these factories online. Between the combination of tight supply, long lead times to stand up factories, along with the growth and change in demand profile we’re seeing for memory, this time may well be different. Even if you argue that it’s still cyclical — on a long enough timeline, all things are — we would argue that if you hold to the timing of past cycles, you will significantly underestimate the sustainability of the current boom. Technological advancements are driving unprecedented demand while structural limitations on capacity (not management planning around pricing power) are preventing supply from being anywhere close to meeting demand. Of course, you might say this is all one CEO talking his book — and as investors and risk managers, we certainly appreciate the skepticism. However, Micron’s customers clearly seemed to be aligned with Mehrotra’s view. Strategic customer agreements One way in which Micron has changed things up to better protect itself from over-investment this time around is with strategic customer agreements (SCAs). These are multiyear, take-or-pay agreements that Micron has inked with customers to ensure the orders being placed now are still there by the time the chips are ready — no matter the customers’ own demand profile. When Micron reported fiscal 2026 third-quarter results in late June, the management team said it had inked SCAs with 16 customers, including “four very large customers and three medium-sized customers.” That represented a major increase from one SCA on the books at the time of Micron’s March earnings call. Mehrotra told Jim on Thursday that since June, “we have signed more strategic customer agreements,” though we didn’t get a specific number. That alone is a sign that customers don’t see any relief coming anytime soon from the supply side, so they’re willing to lock in prices — or a pricing framework— now for years into the future to ensure they get the supply they need, based on the demand they’re currently seeing. Is it possible the customers themselves are overestimating demand? Sure. However, the growing number of contracts does increase our confidence that isn’t the case because it means that orders are coming from more places. That implies that more companies are seeing what Mehrotra is describing, not less. Put another way, consensus among those buying memory for their own products is growing that the boom is real and will last for years. Shareholder returns While the supply-and-demand picture certainly bodes well for sales and earnings, it’s no secret that we love a good buyback program. That’s why we were thrilled to hear Mehrotra commit to repurchasing shares in the future — just as we’re seeing headlines around massive capital return programs for SK Hynix and Samsung. “We are growing the business. We are investing in the business. We are investing in advanced research, R & D roadmap for our products, investing, as you see here, in building out the memory supply base. So, we will, of course, invest first in growing the business. We had $25 billion of [operating cash flows] last quarter. Free cash flow this quarter will even be greater. Of course, excess cash, we will return to shareholders. We are best positioned ever to grow the business, as well as provide return to our shareholders at larger levels than before. And yes, we are committed to doing that.” Micron’s hands have recently been tied on this, having agreed to a two-year ban on large-scale buybacks in order to secure CHIPS Act funding. However, that prohibition is set to expire in early December. As a result, we expect to see a lot the cash Micron is pulling in now returned to shareholders in 2027. Micron’s free cash flow was about $17.5 billion in its fiscal third quarter reported in June. It is expected to be about $50 billion in fiscal year 2026. If the company were to follow in the footsteps of peer SanDisk, which recently committed to return 100% of excess cash to shareholders over a three-year period, Micron could buy back roughly 10% of its market cap in fiscal year 2027, based on its estimated $124 billion of free cash flow. (Jim Cramer’s Charitable Trust is long MU and INTC. 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.
Source link
