TheChek News Flow

The Takeaway

When Micron Technology reports its fiscal first-quarter earnings on Wednesday afternoon, investors will be listening for the memory chip maker’s outlook in the next few quarters. Micron is one of three companies, along with Samsung and SK Hynix, that make high-bandwidth memory, a key component of AI chips. Surging demand for AI chips has lifted memory chip revenues for all three companies in the past year or so.

That in turn has lifted all three stocks by more than 100% this year—Micron alone is up 165%. Given the seemingly limitless appetite for AI chips, demand for memory should continue to outpace supply for the next couple of years, which suggests that even after their recent run-up, these secondary beneficiaries of the AI boom should have more room to rise. The question now is how much more room the stocks have to run.

“AI spending is going to be strong next year, and memory is a clear area, kind of like storage drives, where you don’t have a ton of competitors and you don’t have a lot of capacity coming on,” said Gus Zinn, a Nomura portfolio manager who oversees the firm’s $6.7 billion science and technology fund.

High-bandwidth memory runs alongside a graphics processing unit—the AI chip, typically made by Nvidia—and helps pipe data to that processor faster and more efficiently. In addition to selling these memory components, Micron, Samsung and SK Hynix offer other data memory products that are useful in training and running AI models as well as for handling traditional cloud computing workloads and for use in smartphones and personal computers. The massive demand for HBM in AI data centers in particular has led to a shortage, allowing all three companies to raise their prices.

Doubling Density

The market for memory chips, including HBM, is highly concentrated: Micron, Samsung and SK Hynix together make up 94% of it, according to the latest figures from Counterpoint Research. And demand from AI chipmakers for memory capacity, sometimes referred to as density, has been steadily increasing as they churn out new generations of processors.

“Every generation of accelerators requires a lot more HBM. The density of HBM [per chip] has been pretty much doubling from generation to generation” if you look across the latest AI processors from Nvidia, AMD and Google, said UBS equity analyst Tim Arcuri.

HBM companies’ manufacturing capacity is nowhere near adequate to meet today’s demand. That’s likely one reason why OpenAI struck deals with Samsung and SK Hynix this fall to secure HBM chips for data centers associated with its Stargate project. OpenAI had already struck a deal to use Nvidia technology, which theoretically should include the HBM. But dealing directly with the memory chip makers could give the company more flexibility in the long term, making it easier to pair the memory with its own custom AI chips.

SK Hynix’s parent company in October said OpenAI’s demand alone will require the entire sector to double its capacity. Making enough memory chips to meet OpenAI’s needs could take anywhere from one to five years, said Kevin Krewell, a retired computing and deep tech analyst who spent more than a decade at Tirias Research.

And even if the pace of training new AI models slows down, demand for HBM chips could remain strong.

“Even if you’re not putting high-bandwidth memory or high-performance memory in the GPUs or the [other specialized AI chips], you still need the [central processing units] that are running the software, and those also require a significant amount of memory, as these large models take a lot of memory,” Krewell said.

The chipmakers’ pricing power has already lifted their profit margins. In the September quarter, Micron’s gross margin jumped to 46%, up 9 percentage points from a year earlier. A similar trend has played out at SK Hynix, which like Micron mainly focuses on selling memory and storage products.

Micron and SK Hynix have both been growing their top line by more than 30% in each of the last several quarters. Wall Street analysts polled by S&P Global Market Intelligence expect that pace of growth to accelerate for both companies next year. They project that Micron, for one, will grow its revenue by 58% in its fiscal 2026, which ends in August. In fiscal 2027, the analysts predict that Micron’s growth will slow to 18% but its gross margin will jump higher again that year to 57%, a level it hasn’t reached since 2018.

Samsung’s gross profit margin has also risen with the latest boom in demand for its memory chips, but less than the margins for the other two. It is overall growing much more slowly. That’s because at Samsung, chips make up less than half of sales, with consumer electronics such as smartphones and TVs contributing the rest. But even for Samsung, the AI memory shortage is a notable catalyst: The company’s semiconductor device solutions division, which includes its memory chip business and a chip manufacturing foundry, contributed 58% of its operating profit in the September quarter, up from 42% a year ago.

To be sure, the companies’ shares aren’t cheap after their rally this year. Each is trading slightly above its respective average multiple of forward earnings before interest, taxes, depreciation and amortization over the last decade. Micron currently trades at seven times next year’s Ebitda, and the other two trade at around four times. But their valuations today look more attractive than in mid-2023, when Micron and SK Hynix traded at around 15 times and had negative gross margins as a result of overbuilding capacity to meet surging pandemic demand. Arcuri of UBS is betting that memory chips’ usefulness to AI will mean their makers can maintain better pricing power than ever before.

“The whole thing about Micron is it has been uber-cyclical, so you pay a super-low multiple off of a high earnings number. That’s changing, because it’s becoming less cyclical,” he said. (Arcuri has a “buy” rating for Micron, but he doesn’t cover the other two firms, which are based in South Korea.)

Long-term contracts should insulate the memory chip makers from any near-term slowdown in AI chip demand, Arcuri added. “Nvidia, AMD and Google are all procuring HBM at least a year, in some cases two years out, and locking in volume and pricing.”

Nomura’s Zinn is holding on to a position in Micron, which he acquired more than a year ago. He also added a position in Samsung to his portfolio a few months ago. Samsung, he argued, has a long-term advantage in its foundry business, which makes it one of just a few companies in the world besides Taiwan Semiconductor Manufacturing Co. that can produce large numbers of semiconductors. Samsung this year restored its dual-CEO structure and is starting to get “back on track” after falling behind its peers technologically, Zinn noted.

Still, “the whole storage market is in a giant bull market,” he said. Trying to pick one winner among the stocks that make data storage and memory chips doesn’t make sense, he said, because “they’re all great, and they’re all kind of the same trade.”

Expectations are high ahead of Micron’s upcoming earnings report on Wednesday. Micron has projected 44% year-over-year revenue growth, but the consensus among analysts is that revenue will grow 48% to $12.88 billion, according to S&P Global Market Intelligence.

Besides cooling AI demand, the other long-term risk for the memory makers is that down the line they could end up adding so much excess capacity that prices have to come down. For now, that scenario looks to be a ways off, so it’s not too late to get in on the memory trade, Zinn said.

So when would Zinn consider selling his shares in Micron, which he’s owned for over a year?

“I don’t like to think about it in terms of stock price, but more in terms of time,” he said. “I think the spending and AI in memory and all these areas that we’ve been talking about is going to be strong next year, so I feel pretty good owning it for the next six months.”