The Hidden Growth in Semiconductor Stocks

The majority of investors dismissed Western Digital’s announcement in February of last year that it was dividing into two distinct businesses as corporate housekeeping, which produces a brief flurry of analyst notes before being forgotten. In a world that seemed to be moving away from hard drives, Western Digital was a hard drive manufacturer.

The flash memory company being spun off, SanDisk, was a storage brand that most people were familiar with from the tiny cards found in their digital cameras. There didn’t seem to be any clear momentum in either half of the split. SanDisk had increased by almost 640 percent by November. One of the best trades of the year in any industry, anywhere, was being held by the investors who had closely monitored what was actually going on inside memory markets. Although not as well, the investors who had been keeping an eye on Nvidia were also doing well.

It’s important to comprehend why semiconductors continue to exhibit this pattern. Most retail investors never consider the supply chain behind Nvidia’s well-known AI infrastructure build-out, which includes massive GPU orders, data center construction, and hyperscaler capital expenditure commitments in the trillions. It takes more than just processing power to run inference at scale and train large language models. Large amounts of fast memory must be positioned next to those processors in order to feed them data fast enough to prevent them from having to wait.

One of the most supply-constrained parts in the entire industry is high-bandwidth memory, the specialized DRAM that stacks directly next to AI accelerators. While the financial media concentrated on chip designers with more effective marketing, the companies that manufacture it and the machines that make it have been quietly compounding.

Perhaps the best illustration of this dynamic in action is Micron Technology. In a single fiscal quarter, the company’s high-bandwidth memory revenue reached $2 billion, and data center operations now make up 56% of total sales. This represents a significant change from Micron’s appearance two years ago, when memory was still experiencing the oversupply hangover that periodically destroys margins throughout the industry.

This time, the structural difference is that AI demand came before manufacturers had significantly increased capacity, so instead of collapsing as is typically the case, prices held and margins grew. Over the course of 2025, Micron’s stock increased by more than 177 percent as that dynamic became unavoidable. The easy money in identifying the setup has already been made, but the run may still have legs.

Perhaps the more paradoxical tale is Seagate. Hard disk drives were expected to be in terminal decline, replaced by solid-state storage and placed in the same historical category as floppy disks and CD-ROMs. In 2025, the company reported gains of more than 200 percent, which is unusual for a dying technology. Solid-state drives are far too costly for cold storage at cloud scale, and AI creates massive amounts of historical data that must be inexpensively stored for extended periods of time, including training datasets, model checkpoints, logs, and archives.

When AI workloads began producing data at rates that exceeded previous infrastructure assumptions, Seagate’s HAMR technology, which allows drives to reach 44 terabytes with a roadmap that extends well beyond that, put the company in a position to precisely capture the tiered storage demand that hyperscalers discovered they needed. Revenue from mass capacity increased by 40% annually. The market eventually changed its valuation to reflect the fact that the story about hard drives dying turned out to be true for consumer electronics but just false for enterprise cloud storage.

Then there is ASML, a Dutch company whose lithography machines are the sole means of producing sophisticated chips. ASML is not a chip manufacturer. It produces the machinery that uses intense ultraviolet light to print circuit patterns onto silicon wafers.

This process is so technically challenging that ASML essentially has no significant competition at the frontier level, which is an uncommon position for any industrial company to hold. The company entered 2026 with a backlog of $38.8 billion and record quarterly bookings of €13.2 billion, despite a 29 percent year-over-year increase in fourth-quarter net sales. Memory systems, specifically EUV machines used by clients like Micron to produce the high-bandwidth memory required by AI infrastructure, accounted for about 56% of those reservations.

The CEO of ASML called the current state of memory demand a “perfect storm,” pointing out that despite growing raw capacity requirements, customers are still adopting DRAM nodes that require more EUV layers. The business anticipates net sales of €34 billion to €39 billion in 2026. The underlying order book provides a fairly clear picture of the direction of demand, even though the stock is trading at an elevated multiple and the job cuts that were announced concurrently with the earnings report added some noise.

Taiwan Semiconductor Manufacturing occupies a position in the AI supply chain that is both the most significant and the least talked about, setting it apart from all of these tales. TSMC is the physical plant behind the intellectual property, where the designs become tangible items that fit inside servers, laptops, and data centers.

It manufactures chips for Nvidia, AMD, Apple, Broadcom, and pretty much every other company that designs cutting-edge semiconductors. Some investors have been cautious due to geopolitical risk surrounding Taiwan, which is understandable and not totally irrational.

However, the more one considers TSMC’s structural position—producing cutting-edge chips for the entire industry regardless of which designer wins any given product cycle—the more it begins to resemble the type of company that gains from the development of AI infrastructure regardless of which company’s chips end up dominating. It’s the traditional pick-and-shovel reasoning: regardless of who finds gold, the shovel manufacturer gains from the gold rush when everyone needs shovels.

Observing all of this, it seems as though the semiconductor industry has entered a phase where the supply chain is just as fascinating as the flagship products, and perhaps even more lucrative. For many years, memory was viewed as a commodity to be manufactured and traded through cycles rather than being owned through structural change. Demand for AI has caused a structural shift big enough to overwhelm the cycle, at least temporarily.

Whether capacity additions eventually surpass demand growth—a variable that no one can predict with certainty—will determine whether that continues through 2026 and beyond. Manufacturers have been slower to grow than during previous boom times, either as a result of institutional learning or the preparation for a supply shortage that keeps prices higher than anticipated. Manufacturers have learned discipline from painful past cycles. Most likely a combination of the two.

It appears more obvious that the investors who dug a little deeper than Nvidia—into memory, fabrication, and lithography equipment—discovered something genuine. The stocks’ returns attest to this.