Over the past couple of months, I have warned investors that the extraordinary rally in memory stocks was becoming overheated. That caution proved well timed, as Micron Technology ((MU Quick QuoteMU - Free Report) ),Sandisk Corporation ((SNDK Quick QuoteSNDK - Free Report) ),SK Hynix ((SKHY Quick QuoteSKHY - Free Report) ) and other beneficiaries of the memory boom corrected sharply from their early-summer highs.

Importantly, my concern at the time was less about the underlying fundamentals and more about positioning. The trade had simply gotten too crowded. Sentiment was euphoric, investors were heavily concentrated in the same names, and that excess eventually came to a head with the liquidation of the highly leveraged Situational Awareness AI fund, which helped trigger a sharp unwind across the group.

Initially, I thought that correction might also mark the beginning of the end for the broader memory cycle. These are highly cyclical businesses, and after some of the leading stocks had appreciated many times over in barely a year, it seemed reasonable to expect that much of the upside had already been captured.

I have since reassessed that view.

The correction appears to have done exactly what was needed: shake out excessive positioning without meaningfully damaging the fundamental story. Memory remains extremely supply constrained, earnings expectations continue to rise and recent developments across the AI industry suggest demand could remain elevated considerably longer than I initially expected.

Now the stocks are beginning to confirm that view. After spending the last several weeks consolidating, the leaders are turning higher again. The Roundhill Memory ETF (DRAM), which is heavily exposed to Micron, SK Hynix, Sandisk and the broader memory complex, recently pushed to fresh one-month highs.

Let's take a closer look at what these companies contribute to the AI boom, why the fundamental setup remains so strong and why I think memory stocks may be entering their next leg higher.

Image Source: Zacks Investment Research

What Micron, SK Hynix and Sandisk Contribute to the AI Boom

Although these companies are often grouped together as memory stocks, they occupy different parts of the AI infrastructure stack.

Micron and SK Hynix are two of the world's dominant suppliers of DRAM and, most importantly for the AI buildout, high-bandwidth memory. HBM sits alongside GPUs and other AI accelerators, rapidly feeding those processors the enormous amounts of data required for training and inference.

As AI systems become larger and more capable, memory bandwidth has increasingly become a bottleneck alongside compute itself. That has created extraordinary demand for HBM while also tightening the supply of conventional DRAM, since manufacturers have increasingly dedicated production capacity toward higher-value AI memory.

Those constraints are showing up clearly in Micron's results. Fiscal third-quarter revenue surged to $41.46 billion from $9.30 billion a year ago, while Cloud Memory revenue climbed to $13.77 billion. Micron also guided for roughly $50 billion in fourth-quarter revenue, showing just how strong AI-driven memory demand remains.

Sandisk plays a slightly different role. The company is primarily exposed to NAND flash and enterprise solid-state drives, which provide the high-speed storage needed throughout the AI data cycle.

That opportunity is expanding rapidly as inference becomes a larger piece of AI infrastructure. Larger context windows, retrieval-augmented generation, agentic applications and enormous volumes of generated data all require more storage and faster access to it. Sandisk's enterprise SSD revenue more than doubled sequentially last quarter as demand from data centers accelerated.

So while the product mixes differ, all three companies sit directly in the path of rising AI infrastructure spending.

OpenAI's Astra Raises the Bar for Memory Demand

The latest catalyst came with OpenAI's release of GPT-6 Astra.

Astra represents another substantial increase in the scale and capability of frontier AI models, and according to Axios was developed using OpenAI's largest training run yet, involving more than 100,000 GPUs at its Stargate facility in Texas.

The investment implication is straightforward: better AI models continue to require extraordinary amounts of compute, and all of that compute needs memory.

More importantly, demand is increasingly moving beyond training. As AI systems shift toward more compute-intensive reasoning, agentic workflows and large-scale inference, the amount of memory and storage required to operate them continues to grow.

The connection to the memory manufacturers is also increasingly direct. Samsung and SK Hynix have already signed agreements aimed at supplying advanced memory for OpenAI's Stargate infrastructure, while OpenAI said this week that it is deepening its work with Samsung on next-generation chips.

That helps explain why the release of Astra immediately reignited interest in the memory trade.

The bigger point, though, is that we continue to get evidence that the AI scaling cycle is not slowing down. If anything, the compute, memory and storage requirements of the leading models continue to move higher.

Memory Stock Earnings Estimates Keep Rising

That demand is flowing directly into earnings expectations.

Micron and SK Hynix currently carry Zacks Rank #2 (Buy) ratings, while Sandisk holds a Zacks Rank #3 (Hold). The particularly important signal is that earnings expectations have continued to improve even after the enormous gains these stocks have already generated.

Valuations also remain surprisingly restrained. Micron and Sandisk recently traded at roughly 6-7 times forward earnings, enormous discounts to the broader technology sector despite exceptionally strong earnings growth forecasts.

There is an important caveat here. Memory stocks often look cheapest near the top of the cycle because earnings temporarily explode, making the denominator in the P/E ratio unusually large. A low multiple alone is therefore not enough to make these stocks attractive.

But that is also what has changed my view.

If the cycle were beginning to roll over, I would expect earnings estimates to start falling as analysts anticipated weaker pricing and additional supply. Instead, expectations remain elevated, supply remains constrained and the industry continues to report extraordinary pricing power.

More importantly, this is not a normal cycle. The scale of AI infrastructure spending increasingly looks like a supercycle, and demand may still have room to expand beyond the already high estimates as training, inference and storage requirements continue to grow.

In other words, the stocks may look like they are pricing peak earnings, but the earnings estimates themselves have yet to show much evidence of peaking.

Memory Stocks Break Out From a Base

The technical picture is now beginning to line up with the fundamentals.

Following the sharp early-summer correction, Micron, Sandisk and SK Hynix spent several weeks digesting their enormous prior gains. That sideways action allowed momentum to cool, shook out some of the excessive positioning and established a much healthier base.

Now the group is beginning to turn higher again. Below, we can see that Micron stock has broken out from this bullish pennant, while SK Hynix stock made new record highs just today.

Image Source: TradingView

The initial rally was driven by an extraordinary fundamental cycle but eventually became too crowded. This time, the stocks are attempting to break higher after positioning has reset while earnings expectations remain exceptionally strong.

Should Investors Buy Shares in MU, SNDK and SKHY?

Memory remains cyclical, and that risk should not be ignored. Eventually, higher prices will encourage additional supply, demand will normalize and the current earnings growth rates will slow.

I simply no longer believe we are there yet.

Supply remains tight, AI infrastructure spending continues to accelerate, estimates remain strong and the technical picture is turning higher again after a meaningful correction.

Among the group, Micron and SK Hynix get a slight edge with their Zacks Rank #2 (Buy) ratings, while Sandisk's #3 (Hold) suggests a somewhat more neutral earnings revision picture. But all three remain well positioned if the broader memory cycle continues.

A couple of months ago, the risk was that investors had gotten too far ahead of the fundamentals. After a sharp correction and several weeks of consolidation, that excess has largely been worked off.

With the fundamentals still intact, I think the next leg higher in memory stocks may now be getting underway.