AI needs more than powerful processors. Every model also depends on fast memory, working storage, and massive data capacity to operate efficiently. As AI infrastructure expands, demand is growing across four layers: high-bandwidth memory (HBM), conventional DRAM, NAND flash, and traditional hard drives. Yet memory stocks have experienced sharp pullbacks despite strong industry demand, raising an important question: is the sell-off creating an opportunity or signaling that the current cycle is nearing its peak? 📉

Micron and SK hynix are benefiting from tight HBM and DRAM supply, while SanDisk faces a different outlook as flash production expands. Western Digital and Seagate could benefit from continued hard-drive demand, and Rambus offers exposure to the technology supporting server memory. Each business has different growth drivers, valuations, and risks, making it important to look beyond recent stock performance.

🔍 Compare six memory and storage stocks, explore why their supply-demand outlooks differ, and examine three ETFs: DRAM, EWY, and SOXX.

We also explain how customer commitments, new manufacturing capacity, broker fees, and overlapping ETF holdings could affect your investment decisions.

Imagine setting aside $500 a month for $ADMA ( ▼ 0.24% ) for five years, buying a fixed dollar amount on a schedule, a method called dollar-cost averaging. In our example the buy happens on the first trading day on or after the 9th of each month, from Nov 2021 through Oct 2026 (the first buy is on Tuesday, Nov 9, 2021, and the last on Friday, Oct 9, 2026): 60 buys, $30,000 in total.

Caution: past pace rarely continues. ADMA's Nov 4, 2026 release for the quarter ended Sep 30, 2026 showed a gross margin of 71% (up from 56% a year earlier), revenue of $141.1 million (up 5% from a year earlier), net income of $49.9 million and adjusted EBITDA of $77.3 million, with ASCENIV demand still growing, but the stock fell from its $20.38 closing high on Dec 11, 2025 to a $7.60 close on Jun 2, 2026 before recovering to $12.73 on Nov 6, 2026 (it rose 26.0% on Nov 5 and slipped 3.6% on Nov 6). Prices can pull back a long way even when the business keeps progressing. The stock has swung widely, so patience and position size matter as much as the pick itself.

AI investing is often associated with Nvidia, advanced processors, and the companies building massive data centers. But there is another part of the AI infrastructure boom that deserves just as much attention: memory and storage.

Every AI system needs somewhere to process information, keep data readily available, and store what it learns. As AI models become larger and more demanding, the need for faster memory and greater storage capacity continues to grow. That creates opportunities across several parts of the semiconductor industry, not just among chip designers.

Yet memory stocks have already experienced dramatic gains and sharp pullbacks. Micron, for example, has delivered extraordinary gains this year, while several memory-related stocks have fallen 20% to nearly 50% from their June highs. That decline has raised an important question: has the AI memory opportunity peaked, or is the market temporarily overlooking demand that remains stronger than supply?

The answer depends on understanding that memory is not one single business. High-bandwidth memory, conventional DRAM, flash storage, and hard drives serve different purposes, face different supply conditions, and may offer very different investment opportunities.

Think of an AI processor as a chef working in a busy kitchen. The processor can work extremely quickly, but it still needs information delivered at the right speed. Memory and storage provide that information, with each layer serving a different function.

The first layer is high-bandwidth memory (HBM), which sits close to AI processors and allows them to access enormous amounts of data quickly. It is the most specialized and expensive type of memory, making it particularly important for advanced AI accelerators.

The second layer is conventional DRAM, the working memory found in servers, computers, and smartphones. It is less specialized than HBM but remains essential because AI workloads require substantial amounts of fast-access memory.

The third layer is NAND flash storage, commonly used in solid-state drives (SSDs). Unlike DRAM, flash storage retains information even when power is turned off. It provides the capacity needed to store applications, files, and the growing volumes of information generated by AI systems.

The fourth layer is traditional hard drives. They are slower than SSDs but offer large amounts of storage at a lower cost per gigabyte. As cloud platforms and AI companies accumulate more data, hard drives remain an important part of the infrastructure.

This distinction matters because the strongest investment opportunity may not be in the fastest-growing technology. It may be in the segment where demand is increasing faster than manufacturers can expand supply.

The first two layers are particularly attractive because they sit closest to the computing workload. HBM helps advanced AI processors operate efficiently, while DRAM supplies the working memory required by servers.

Three companies dominate HBM production: SK hynix, Samsung, and Micron. SK hynix is the largest supplier, accounting for approximately half of the global market. Its position makes its financial results an important indicator of demand across the memory industry.

Despite concerns about competition from China's CXMT and potential helium supply disruptions affecting Korean manufacturers, SK hynix delivered a record quarter. The company generated approximately 62 trillion Korean won in operating profit, demonstrating how strong pricing and demand can translate into exceptional profitability.

Its customers are reportedly requesting more than twice the HBM volume it can currently supply. The company has also secured long-term agreements covering more than half of its expected production for 2027 and 2028, while management expects tight supply conditions to continue through 2030.

That does not mean the risks have disappeared. CXMT is expanding its memory operations, and additional manufacturing capacity from SK hynix, Micron, and Samsung could eventually reduce the shortage. However, new semiconductor factories take years to build and bring into production. Industry estimates point to limited additional capacity in 2027, followed by more substantial additions in 2028.

Conventional DRAM faces a similar challenge. HBM production requires substantially more manufacturing capacity than ordinary DRAM, meaning manufacturers must balance their production priorities. As more factory resources go toward HBM, the supply of conventional DRAM can remain constrained.

This is important because the two segments do not necessarily reach a supply-demand balance at the same time. Current expectations are that conventional DRAM could remain tight until around 2028, while HBM shortages may last longer.

For investors, SK hynix provides a useful signal, but Micron offers a more direct way to participate in both HBM and conventional DRAM through a US-listed company.

Micron $MU ( ▼ 0.86% ) stands out because it manufactures both HBM and conventional DRAM. Its latest quarterly sales were nearly five times higher than the year-earlier period, while operating profitability reached approximately 81 cents for every dollar of revenue.

Another important development is the company's customer commitments. Micron reportedly holds approximately $12.7 billion in customer cash tied to agreements that cannot easily be canceled. This represents a meaningful change from previous memory cycles, when customers could reduce orders quickly whenever demand weakened.

Long-term agreements and upfront payments can make future demand more predictable. They do not eliminate the possibility of a downturn, but they may reduce the sudden order cancellations that historically damaged memory manufacturers.

Micron's valuation also attracts attention. Its forward PEG ratio is among the lowest of the companies discussed, suggesting that its expected earnings growth is not fully reflected in its current valuation. However, low valuation multiples are not proof that a stock is cheap. Earnings estimates can change quickly when memory prices, supply, or demand shift.

The biggest question is whether the current cycle can avoid the severe reversals that have repeatedly affected the memory industry. Micron's customer commitments provide a stronger foundation than in some previous cycles, but investors still need to account for the possibility that new factory capacity eventually pushes prices lower.

Not every company benefiting from memory expansion actually manufactures memory chips.

Rambus $RMBS ( ▲ 0.66% ) supplies semiconductor components that help manage communication between server processors and memory. Its technology is used in a significant portion of new server memory modules, making it an indirect participant in the growing demand for memory infrastructure.

Unlike Micron or SK hynix, Rambus does not benefit directly from rising DRAM prices. Its business is more closely tied to the number of memory-related components and systems being deployed.

That distinction can matter when memory prices eventually cool. Manufacturers may experience lower revenue and margins as supply catches up, while companies supplying essential supporting components may face a different set of business pressures.

Rambus is estimated to have a more moderate three-year potential return than several memory manufacturers, but its earnings estimates are considered more consistent because its business is less dependent on memory pricing.

For investors who want exposure to the expansion of server infrastructure without relying entirely on higher memory prices, Rambus represents a different part of the opportunity.

The third layer, NAND flash storage, is where the outlook becomes more complicated.

Flash demand continues to benefit from AI data centers, but supply is expected to grow by approximately 25% annually, faster than demand growth of slightly less than 20%. If those estimates hold, the market could begin moving toward better supply availability in the second half of 2027.

That could put pressure on flash prices even while AI-related demand remains healthy.

SanDisk $SNDK ( ▼ 2.0% ) is particularly important because it focuses on flash storage. Its latest results were stronger than Wall Street expected, with earnings of approximately $49.50 per share compared with expectations of around $46. Data center sales also increased by approximately 60% in three months and now account for more than 40% of total revenue.

The company also provided a stronger earnings outlook for the following quarter. However, its consumer business, including memory cards and retail storage drives, declined approximately 29% sequentially. That weakness was relatively small compared with its overall sales mix, but it still highlights the difference between enterprise AI demand and traditional consumer storage.

SanDisk's shares have experienced a substantial pullback from their recent highs, although they remain significantly higher for the year. The company's longer-term sales growth expectations also suggest that management recognizes the possibility of a more moderate flash market beginning in 2028.

This makes SanDisk an interesting but less straightforward opportunity than HBM and DRAM manufacturers. Strong current results do not automatically mean that today's growth rate can continue for several years.

Investors should also be aware of its announced 8-for-1 stock split scheduled for November 20. A split increases the number of shares while proportionally reducing the price per share; it does not change the total value of an investor's position.

The fourth layer may be the most surprising part of the memory opportunity.

While flash storage could see supply catch up with demand, traditional hard drives face a different situation. Manufacturers are increasing storage capacity by fitting more data onto each drive rather than building entirely new factories. Supply growth is estimated at approximately 25% annually, while demand is growing faster.

That imbalance could support pricing power through 2028.

Two American companies are central to this market: Western Digital $WDC ( ▲ 0.26% ) and Seagate Technology $STX ( ▼ 0.94% ).

Western Digital has benefited from selling more storage capacity at higher prices. Shipments increased approximately 22% year over year, while the price charged per terabyte rose around 28%. Together, those improvements pushed sales up by more than 50%.

Seagate has taken a different approach by securing customer commitments well into the future. Much of its data center hard-drive production is reportedly committed through 2028, with certain cloud agreements extending through 2030. Its latest results exceeded expectations, and its next-quarter guidance was also stronger than anticipated.

These developments suggest that hard-drive manufacturers may retain pricing power even as other memory segments begin to normalize.

However, valuation matters. Seagate trades at approximately 18 times expected earnings, compared with roughly five times for Micron. That difference reflects different expectations, business models, and perceived risks. It also means investors should not assume that strong demand automatically makes every storage stock equally attractive.

These three-year scenarios illustrate how differently these companies could perform if current earnings expectations and valuation multiples hold. These are hypothetical estimates, not guaranteed returns, and they assume the memory shortage remains supportive.

These figures should be treated as valuation scenarios rather than price targets. They assume current market multiples remain in place and that expected earnings materialize. A return projection can change dramatically if earnings estimates fall or investors become less willing to pay the same valuation.

The broader lesson is that the most attractive stock is not necessarily the one with the highest projected return. The business model, supply cycle, customer commitments, and valuation all influence how much risk is attached to that potential reward.

Buying individual memory stocks requires investors to evaluate each company's manufacturing position, customer relationships, and exposure to changing prices. ETFs can simplify that process by spreading exposure across multiple companies, although fund structure and concentration still matter.

Roundhill Memory ETF $DRAM ( ▼ 1.29% ) is the most directly focused option here. It invests in the memory industry and provides exposure to major manufacturers, including Micron, Samsung, and SK hynix.

Approximately three-quarters of its exposure is concentrated in these three companies, making it a relatively direct way to participate in the memory cycle. It also offers US investors exposure to Samsung, which does not trade as an ordinary US-listed stock.

The fund charges an annual expense ratio of approximately 0.65%. However, investors should look beyond the headline holdings. Much of its Micron exposure comes through contracts with major banks rather than directly owned Micron shares. That means part of the investment depends on financial counterparties, not just the performance of the underlying company.

There is also a trading-cost issue to investigate. Fidelity reportedly charges an additional transaction fee of up to 5% on DRAM purchases, capped at $100 per trade. A $1,000 purchase could therefore incur an extra $50. Schwab may introduce a similar charge, while Vanguard, Robinhood, and E*TRADE reportedly do not impose the same fee.

Because broker fees and fund policies can change, check the current transaction charges before placing an order. A fund's expense ratio is only one part of the total cost of ownership.

iShares MSCI South Korea ETF $EWY ( ▼ 0.82% ) offers a broader way to gain exposure to South Korean companies, including SK hynix and Samsung.

The two memory manufacturers are each estimated to represent close to a quarter of the fund. Unlike DRAM's bank-linked exposure, EWY provides exposure through actual shares of the South Korean companies.

Its annual expense ratio is approximately 0.59%.

The trade-off is that EWY is not a pure memory ETF. It provides exposure to the broader South Korean equity market, so its performance can also be influenced by companies and industries outside semiconductors.

EWY and DRAM also overlap substantially because both rely heavily on Samsung and SK hynix. Holding both does not necessarily provide as much diversification as investors might expect.

iShares Semiconductor ETF $SOXX ( ▲ 1.19% ) offers a broader semiconductor approach, with memory manufacturers representing a smaller portion of the portfolio.

Approximately 11% of SOXX is estimated to be allocated to memory-related companies. Its annual expense ratio is approximately 0.33%, making it less expensive than the two more concentrated options discussed above.

SOXX may be more suitable for investors who want memory exposure alongside a wider range of semiconductor businesses rather than making a concentrated bet on the memory cycle.

There is also substantial overlap between SOXX and the VanEck Semiconductor ETF (SMH), estimated at approximately 69%. Holding both may create unnecessary duplication rather than meaningful diversification.

The right choice depends on how much memory exposure is already present in a portfolio. DRAM offers a concentrated memory strategy, EWY provides broader South Korean exposure, and SOXX spreads the investment across the semiconductor industry.

Memory stocks are cyclical, and the industry has a long history of moving from shortages to oversupply. That history explains why investors remain cautious even after record earnings and strong AI-related demand.

The current cycle has some important differences. Customers are signing longer-term agreements, committing to future production, and in certain cases providing cash upfront. These arrangements may give manufacturers more visibility than they had during previous downturns.

At the same time, the supply response is already underway. New factories from Micron, Samsung, and SK hynix, alongside China's CXMT expansion, could eventually change the balance. Flash storage may face that adjustment sooner than HBM or conventional DRAM, while hard drives could remain constrained for longer.

For a busy investor, the goal does not have to be predicting which memory stock will hit its lowest price. A more practical approach is to understand where demand is strongest, which companies have secured customer commitments, and which parts of the industry may face oversupply first.

ETFs can make that process easier, but they are not automatically safer. A concentrated fund can still fall sharply when the memory cycle turns, and overlapping funds can leave a portfolio more exposed to the same companies than intended.

The same lesson applies to ADMA: its roughly 836% five-year gain came with a close about 48.1% below its April 2025 high, so knowing what drives the business matters more than the latest price move.

Ready to Revolutionize Your Wealth?

Here's what's waiting for you:

As the world becomes increasingly digital, this question will be on the minds of millions seeking new income streams in 2026.

The answer is—Absolutely!

Thank you for taking the time to read today’s email! Your support is what allows me to send out this newsletter for free every day. 

 What do you think of the new format? Please provide your feedback in the poll below, and if you find the newsletter valuable, feel free to share it with other investors!

Disclaimer: This newsletter is for informational purposes only and should not be considered financial advice. Please consult with a financial advisor before making any investment decisions.