Big tech companies are spending hundreds of billions of dollars on chips and equipment to outfit their growing data centers to continue pushing the potential of generative AI. Total infrastructure spending from the top 10 AI tech companies could climb from $435 billion last year to over $1 trillion by 2028, according to estimates from Dell'Oro. In other words, there's still a lot of room for AI chip stocks to keep growing.
Perhaps the most important component of any data center focused on AI training and inference is the GPU cluster. Bigger clusters of more powerful GPUs are capable of training bigger models faster. Nvidia has established itself as the leading GPU maker, and its revenue and profits have skyrocketed as hyperscalers snatch up its chips as fast as it can produce them.
But GPU makers aren't the only chipmakers benefiting from the rapid growth in AI spending. There are a lot of different components that go into data centers. As demand continues to grow over the next few years, one chipmaker's stock could climb up to 85% from its price as of this writing through the next 12 months, according to one Wall Street analyst.
Here's what investors need to know.
Image source: Getty Images.
Don't forget about this component
One of the most expensive parts of building a GPU, which is more than just a piece of silicon these days, is the high-bandwidth memory, or HBM. The most advanced GPUs process tons of data every second. But in order to process that data, the unit must have a way to access the data quickly. Traditional memory components quickly became a bottleneck as the processing power of GPUs increased. As such, demand for HBM chips has exploded alongside the growing demand for the most advanced GPUs.
There are only a few chipmakers producing HBM chips. The leader in the market is SK Hynix, which established a strong relationship with Nvidia. But another memory chip maker, Micron Technology (MU 3.75%), is poised to take share from the market leader.
Micron was late to HBM, but now it's ramping up development on its latest generation HBM3E 12H product, which it expects to be its biggest source of HBM shipments in the current quarter. It earned a big design win with Advanced Micro Devices' latest GPU, the MI355X. Analysts expect AMD's latest chip to compete for more share of data center spending against Nvidia, which bodes well for Micron as well. Overall, Micron expects its share of the HBM market to grow to the same level as its total DRAM (general memory chips) market share, about 25%, at some point in the second half of this year. That's pretty rapid progress after starting from a near standstill a few years ago.
NASDAQ: MU
Key Data Points
Importantly, Micron's next-generation HBM4 progress is going well, too, with performance 60% higher than its HBM3E chips with 20% less power consumption. Management says it's delivered samples to customers, and it expects to ramp up volume production next year.
The early results are evident in Micron's financials. HBM revenue grew 50% sequentially in its most recent quarter. As a result, total revenue from DRAM sales (of which HBM is a part) climbed 51% year over year. Management expects the ramp in HBM3E to push its gross margin higher this quarter, reaching 42%, up from 39% in its most recent quarter.
While HBM is the driving force behind Micron's recent results, it's not the only factor pushing demand for Micron's chips. Device makers are also in need of more traditional memory chips from Micron as it's also an important component for on-device AI capabilities. Everything from PCs and smartphones to automobiles with advanced computing capabilities (like self-driving) need more powerful memory chips. On-device AI capabilities may drive significant smartphone upgrades over the next few years, representing another potential catalyst for Micron.
The 80% upside in the stock
Micron is making strong progress in the HBM market, and that led Rosenblatt Securities to slap a $200 price target on the stock following its third-quarter earnings report in June. The key factor behind the analyst's Street-high price target, which is an 80% jump from today's price, is the capacity constraints on the market.
SK Hynix notably said it had already sold out its entire capacity for 2025 by the end of the first quarter and it expected to finalize its 2026 volume within the first half of the year.
"With DRAM wafer capacity expansion over 18 months away, we see this cycle driving Micron's income model to all-time highs," Rosenblatt's Kevin Cassidy wrote in an investor note. He find's Micron's current valuation, less than 14 times forward earnings, as extremely attractive, especially given the strength of its balance sheet and potential earnings leverage.
Before investors run out and buy the stock, though, there's an important risk to consider with Micron. It's an extremely cyclical stock.
Since it manufactures its chips itself, it has significant capital expenditures for equipment and capacity. A downturn in demand would not only hurt volume but pricing as well, because most of its products aren't very differentiated from SK Hynix or other competitors. That would weigh heavily on earnings. Micron saw a significant shock to its earnings in 2023, as inventory levels rose and demand from China vanished.
That said, the strong expected growth in AI spending could provide a huge boost to Micron's revenue and profit margins over the next few years, as long as it's able to maintain leading-edge technology with HBM. That could mean an extended earnings cycle with very strong earnings growth for the next few years. At some point, however, Micron will see a big drop in demand and earnings will severely suffer. But the long-term trends favor growing demand for memory both in data centers and consumer devices and automobiles.
As such, Micron is a great value at today's price, especially for investors looking for a way to invest in the growing spending of the hyperscalers without paying up for expensive GPU chipmakers like Nvidia and AMD.
