Nine thousand people working long hours, six days a week, to meet a deadline for full production of memory chips by the first quarter of 2027. The workers know they can each make $200,000, maybe more. They seem united in their mission; no Potemkin village going up in Boise, Idaho. Just two state-of-the-art fabs for the number three producer of dynamic random access memory (DRAM), Micron.
All for one reason: Demand for DRAM is so strong that this proud American company — which stuck with memory-chip manufacturing even as peers, including Intel, closed U.S. factories — needs to get more chips into data centers as quickly as possible. Sanjay Mehrotra, Micron's CEO since 2017, does not want his company to be the bottleneck in the great AI build-out in the U.S. and around the world. He cares passionately about how well Micron is doing, how many thousands of patents this innovative company amasses, and how specialized and powerful his semiconductors are, especially high-bandwidth devices.
Mehrotra hears the argument from influential voices, including Tesla and SpaceX CEO Elon Musk, that the balance of power in the data center buildout is shifting toward DRAM makers. Maybe he reads the postings. But one look at the commitment you see in Boise, the kind of job you might think could only be done by China's fabled Eighth Route Army, tells you he doesn't want that kind of chokehold power. Sanjay, as everyone knows him, would rather have it stem from better, more powerful chips and the proprietary intellectual property they contain. He wants to take share from Samsung and the all-powerful SK Hynix, rather than wield the de facto power that comes from being a bottleneck.
He wants to beat the Koreans, not hold Musk or any other hungry data center or neo-hyperscaler CEO hostage to tight memory supply. He believes Micron can do that even as its rivals make their chips in the obviously lower-cost geographies. Sure, the U.S. is a higher-cost producer. But Mehrotra says technology can make up the difference. Two massive fabs are taking shape in Boise, and their importance can't be overstated. On top of that, Micron is building two fabs in Clay, New York, near Syracuse, this time hiring both union and non-union employees, as in Boise. They could be tougher. Will the unions allow employees who want to work six days a week to do so? Will it matter that it's a CHIPS and Science Act project? How does that cut? Micron's putting up $100 billion here. Will the $6 billion that the government is putting up matter in this ten-year project? Mehrotra is hopeful.
No matter what, I saw a true marvel when I visited Boise last week. There was a spirit at our luncheon that I have not seen in any of our site visits. They get the high stakes. I think they will deliver. Their cause seemed so meaningful to so many that I felt like a mental Lilliputian when I asked when he would start buying back stock in companies like Sandisk, Western Digital, and Seagate. Technically, he can't until December, when the CHIPS Act allows him to begin. However, he could say he was going to if he felt it was the right thing to do. He didn't seem to mind the recalcitrance of these other players even as they are the true bottlenecks. They have chosen to return shareholders' money rather than use it to build more plants. Perhaps they are mindful that, in the past, doing so left them with a surfeit of product and a valley of earnings, like the shadow of death. Mehrotra doesn't mind. He's on a mission to grow, and he is certainly growing, as befits the company's trillion-dollar status.
The stakes are incredibly high for the U.S. to maintain its lead in data centers. It seems to be getting harder and harder in this election year. The governors of Texas and Pennsylvania, one solidly red state, the other a key battleground, have taken steps to slow things down. If this turns into something like the Biden administration's restrictions on new LNG export projects, these states risk losing future investment. Virginia, already home to some 250 data centers, might be willing to take more. But the opposition to data centers — projects backed by companies that will fold if communities make development difficult — now seems to know few bounds. There's a loose coalition of data center companies, but to me it's too loose. What's needed is a common code of conduct that puts cooperation ahead of naked competition if the U.S. is going to stay ahead. A president who says he would welcome data centers if he were a local politician, perhaps demanding that developers help pay for the infrastructure they require, doesn't do the job. That's laissez-faire, and laissez-faire just isn't working.
Which brings me to last week. We saw what can happen when the data center thesis gets bogged down. We bought some GE Vernova because we thought it was down enough. That seemed like a decent bet at the beginning of the week. By the end, I wish we had sold it. Broadcom is helping to arrange debt financing to accelerate the AI buildout, something the market didn't mind but I sure did. I want Broadcom to use its capital to buy back stock as it falls. It still had $10 billion of authorized share repurchases as of May 3. I would have preferred a reload. But if that money instead led to more AI infrastructure being built, that may matter more, especially as many investors seem to be questioning the data center thesis. Corning and Qnity Electronics were standout losers, the latter with a personnel change well chronicled by my colleague Jeff Marks in his Homestretch. It was bad.
Oh, and don't get me started on my favorite stock, Intel, the one that can't stop going down, perhaps because of a looming overhang. The government's stake in the chipmaker becomes eligible for sale on Aug. 27, and it's unclear how that block may be sold. A tight syndication would do the trick and let the stock rally. I had thought it would be a clearing event that would restore interest in the stock. The revulsion for the data center makes it a tougher sell. I believe the data center issue will be resolved by the election. Those with a strong anti-data-center slate of victors will most likely spell the end of data center growth in their states. If it is too many, then I will have been resoundingly mistaken and will have to pay the price for a lack of vision. I didn't see it coming. I knew there was a backlash, but I presumed there were enough states and enough locales that it wouldn't matter.
In a moment of despair, I called my Mexican contacts for my wife's agave spirits business and asked whether Querétaro, the home of giant plants that make parts for data centers, couldn't welcome the data centers themselves. They were looking into it, but the inquiry smacks of desperation, given the state of play regarding the stability of the country needed to protect the projects. I find myself reluctant to do what I typically would do: buy down in a pyramid-style bet that the selling will overshoot the event. However, now we may just have to watch and wait. Let the drumbeat take its toll. And yes, it has crossed my mind that it is not too late to cut our gains. Selling more Corning or more Broadcom makes sense. I don't know what to make of Qnity. The hyperscalers themselves no longer trade with each other. That move above $600 by Meta on a mere musing by CEO Mark Zuckerberg about using compute for a web service says to me you've got to stay long. That resilience of Eaton, to be sure, says not everything in the data center will suffer equally. There's trepidation for certain. With the political drumbeat so loud, I sense more pain. I don't want to cut and run from an unstoppable story. I don't think progress can be stopped. But if the majority of shareholders think otherwise, I know the pain will be the worst we have suffered this year.
Micron's demand gives me confidence in the AI buildout and makes me want to steel myself. We most likely will. However, I know it's no longer a given; the demand isn't the problem, but the powerful political backlash that could prevent projects from getting built. In some ways, it is the data center companies' fault. Had they gotten together and given out palatable game plans, it would have made a great deal of difference. Right now, no one is canceling an order that I can tell to GE Vernova and its derivatives. That's heartening. I just hope the desire to have data centers in some states obliterates Friday's pain. I had thought I would be celebrating Micron breaching $1,250, but instead saw it fall less than 1% to $963. I wish I could say that was the worst we were hit. Instead, when it came to the data centers, it was among the softer of punishments. We need to see if some politicians speak out in favor of AI this week. If they don't, we will look intransigent. I don't like being intransigent. You know I have said that I like some more than others. You know we have ample cash, but I don't want to buy to offset these stocks. It's a new development, and it's not gameable. Let's see what happens. No matter what, though, Micron will be the one we buy more of. Intel? Only if the government sells it. Everything else? Too risky. We have to see if mob rule takes over.
