Don't Forget: 5 Memory Stocks for AI's Next 4 Waves

Your brain has over 85 billion neurons and an estimated 100 trillion connections.

That's a lot.

And it runs on about 20 watts.

That's a single dim bulb... or one bright LED.

Much like people.

Now, consider what the brightest people in Silicon Valley are building...

Meta's Prometheus AI supercluster is being designed for ONE GIGAWATT of power capacity.

Its first phase is already running, and Meta is working toward roughly a million interconnected GPUs.

One gigawatt is enough electrical power for 50 MILLION human brains operating at 20 watts apiece.

As DD has expertly shown you in his essay Nuclear Just Powered an NVIDIA Chip... this is why energy is a critical AI bottleneck.

As AI writes more software... runs businesses... researches drugs... makes scientific discoveries... helps you fix a broken light switch and more...

Every task places new demands on a piece of hardware that isn't nearly as glamorous as Nvidia's GPUs.

Don't forget it.

MEMORY.

Gartner forecasts global memory-chip revenue will rise from $220 billion in 2025 to $837 billion in 2026...

And surpass $1 TRILLION in 2027.

So, is this a once-in-a-lifetime opportunity for the 5 memory stocks below...

Or the setup for a spectacular semiconductor bust?

That depends on how many waves of AI come together at the same time...

Four Waves. One Bottleneck.

Micron recently presented a slide that displays the memory opportunity in one chart.

Many investors don't realize its implications.

Each new wave of AI increases memory demand.

So, let's review each of these four waves...

Wave One: Training AI Models

First, we have to teach AI models.

That means feeding enormous amounts of data into powerful computers so they can learn patterns, recognize images, understand language, and solve problems.

Those computers need memory to hold the information they're working with.

Lots of it.

Nvidia's H100 AI processor came with 80GB of high-speed memory. Its newer B300 has 288GB.

That's 3.6 times as much memory in just one generation.

And the bigger these AI models become, the more memory they may need.

But the memory needed for teaching AI isn't the only use case.

There's a second wave...

Wave Two: AI Reasoning

Now consider what happens when you ask ChatGPT or Claude a difficult question.

Instead of instantly producing an answer, newer AI models work through your prompt, step by step, before responding.

Each step takes additional computing power.

And more work means more information flowing between the processors and their memory.

This is called inference.

And the harder the questions, the greater the computing demands.

Micron says the amount of memory per AI server has doubled in just three years.

But answering questions is only the beginning.

What happens when AI stops simply answering questions and starts completing tasks end-to-end for us?

That's wave three.

Wave Three: AI Agents

Imagine asking a future version of ChatGPT or Claude to plan a golf vacation for you and your buddies.

Not just recommend courses and hotels.

Actually compare flights, check everyone's calendars, confirm with them, book tee times, plan the entire trip...

And even remind your buddy he should bring a few extra boxes of Kirklands.

Several AI agents could work together, each handling different tasks.

And each needs memory to track what's been done... and what comes next.

Micron reports that the amount of information AI can keep track of (its context) has been growing roughly 30-fold annually.

And the computing demands are exploding.

Nomura projects AI data-center power requirements will quadruple to 40 gigawatts by 2028.

So, putting AI to work will require more computing power than teaching it.

Nvidia CEO Jensen Huang saw this coming years ago.

He talked about it at Nvidia's GTC conference in April 2016.

Had you invested $10,000 in Nvidia the next trading day, you'd have about $2 million today.

That's the power of recognizing where technology is headed... finding the right stocks... and letting it play out.

And the next wave of tech takes AI beyond the data center.

Into our homes, cars, and machines.

Wave four.

Wave Four: The Brains Leave the Data Center

Cars, drones, robot dogs, factory machines and humanoid robots must recognize the world, and sometimes make decisions without waiting for an internet connection.

This is referred to as edge AI.

Micron's presentation says a Level 4+ robotaxi may carry MORE THAN TEN TIMES the memory (DRAM) found in a current Level 2+ vehicle.

Plus, the next generation of AI devices is under secret development around the world.

OpenAI has teamed up with Jony Ive (the Apple designer) to develop new AI hardware.

The product and its memory suppliers remain unannounced...

But imagine a small edge device with a camera, microphone and Wi-Fi connection.

No one knows what it looks like, but that hasn't stopped people from guessing...

It could be your assistant, teacher, guide and second pair of eyes... and an agent that actually gets things done.

It doesn't have to replace your iPhone.

A compact edge model can handle immediate tasks. The cloud model can tackle the complex questions.

A hypothetical 100 million NEW AI companions with 24GB each would require 2.4 billion gigabytes of DRAM.

That's roughly a quarter of the memory installed in all new smartphones worldwide in an entire year.

(And that only assumes adoption equivalent to less than 10% of annual global smartphone sales.)

Imagine what happens as AI companions become as common as smartphones.

An entirely new category of memory demand...

On top of data centers, cars, drones, robots, and the smartphones we already own.

Could the iPhone Be Unseated?

Don't forget.

Technology has a direction.

Capabilities once confined to data centers are reaching desktops.

Better chips and memory will push more of them into smaller devices.

The remaining obstacles are cost, heat, battery life, and bandwidth, not new laws of physics.

Perfect that, and you have a monopoly.

Could the iPhone monopoly eventually be unseated in the coming years?

Apple spent years perfecting its iPhone empire while Siri (once a glimpse of the future) fell behind.

I have a hard time imagining Steve Jobs being content with what Apple has become...

Because they're no longer on the cutting edge.

Apple's new AI Siri just shipped in beta. Is it the assistant he dreamed of?

We'll see.

No matter how it plays out...

I'm excited to profit from which AI device wins.

For now, we can watch the memory stocks already riding this next computing revolution.

Lots of Demand. Very Few Suppliers.

Again...

All those AI data centers, robots, and pocket-sized edge device companions need memory.

But they need different kinds...

HBM is extremely fast (HB = high bandwidth) memory for powerful AI processors.

LPDDR is energy-efficient memory (LP = low power) for smaller, edge battery-powered devices.

NAND flash stores information even when the power is off.

And only a handful of companies manufacture the most advanced chips.

Samsung, SK hynix, and Micron controlled approximately 87% of global DRAM revenue in the second quarter of 2026.

Three companies. Most of the market.

More demand and limited supply mean higher prices...

Until new factories catch up.

That's the opportunity, and the danger, behind our five memory stocks.

To be clear, these five are on my watch list, not official Moonshot Minute recommendations. Do your own research before you buy any of them.

Stock #1: Micron Technology (MU)

One company: Every wave of AI.

Micron makes HBM, LPDDR and NAND.

In fiscal 2026, revenue reached $133.2 billion, up from $37.4 billion a year earlier.

Gross margin hit 80.7%!

That's the power of a supply-and-demand squeeze sending memory prices soaring.

Micron is sampling LPDDR6 for physical-AI applications and spent $27.4 billion on capital expenditures during the fiscal year.

The Risk: Today's margins may be cyclical rather than permanent.

What I'm Watching for You: Whether advanced products and multiyear customer agreements make its earnings more durable than in previous memory booms.

Stock #2: SK hynix (000660.KS)

The HBM Specialist.

SK hynix held about 50% of HBM revenue in the second quarter, according to Counterpoint...

And it just reported a 328% (yes, 328%) rise in revenue year over year thanks to being first in HBM memory.

But it's not only a data-center story.

Its new LPDDR6 technology promises 33% faster data processing and more than 20% better power efficiency than its LPDDR5X comparison.

Battery-powered AI needs efficiency.

The Risk: Samsung and Micron are gaining ground in HBM, and even the leader can suffer when prices reverse.

What I'm Watching for You: The next generation of HBM, and whether edge memory becomes a meaningful second growth engine.

How to Buy: SK hynix (SKHY) trades directly on the Nasdaq through American Depositary Receipts. Investors can also purchase Korean shares (000660.KS) through Interactive Brokers.

Stock #3: Samsung Electronics (005930.KS)

The Korean giant that can make almost everything.

Samsung does it all. HBM. Conventional DRAM. LPDDR. NAND. And finished devices to use them.

Samsung reported record quarterly memory revenue and operating profit in the second quarter of 2026, helped by AI demand and stronger prices.

Samsung can supply both big and small AI systems.

The Risk: It's not a pure memory stock. Phones, displays and chip foundries complicate the bet.

What I'm Watching for You: HBM execution and how well Samsung defends profitability when supply catches up.

How to Buy: You can buy Korean shares of Samsung through Interactive Brokers (005930.KS).

Note: You can also own SK hynix and Samsung through the iShares MSCI South Korea ETF (EWY). The ETF holds about 23% Samsung and 23% SK hynix... but the rest is spread across other South Korean stocks.

Flash Storage Is the Second Half of the Story

Agents save their work, robots need maps and cameras record video.

That information doesn't all belong in expensive working memory.

Much of it belongs in NAND flash memory.

Stock #4: Sandisk (SNDK)

The flash storage bet.

Sandisk reported fiscal 2026 revenue of $20.25 billion, up 175%, with data-center revenue rising 437%.

But don't mistake that for comparable growth in the number of bits sold.

The company said roughly two-thirds of its fourth-quarter sequential revenue increase came from PRICE increases.

The Risk: NAND can become oversupplied even while HBM remains scarce.

What I'm Watching for You: Enterprise-storage demand after pricing normalizes.

Stock #5: Kioxia Holdings (285A.T)

The Japanese manufacturing partner.

Kioxia and Sandisk jointly develop and manufacture NAND in Japan.

They extended key agreements through 2034 and began producing tenth-generation 3D flash at Kitakami this summer.

These aren't two independent hedges against a NAND downturn.

The Risk: Both suffer if too much flash capacity arrives.

What I'm Watching for You: Cost per bit, advanced storage products and the speed of new factory ramps.

How to Buy: Kioxia trades on the Tokyo Stock Exchange under 285A, and American investors can access Japanese equities through Interactive Brokers.

How Long Will the Shortage Last?

Nomura expects memory shortages to persist through at least 2028.

Memory has historically been cyclical.

High prices encourage new factories.

Micron, Samsung and SK hynix are investing billions in new production capacity.

Meanwhile, China's CXMT is expanding aggressively, with plans that could double its DRAM manufacturing capacity.

New factories will begin adding capacity in 2027, with substantially more arriving in 2028 and beyond.

But advanced AI memory is difficult to manufacture.

Factories can take years to build, equip, and perfect.

So, I have my eye on a bigger short-term risk factor.

The code behind AI is getting more efficient...

And better algorithms can do more work with existing memory and electricity.

That's a risk to memory demand per task.

But it's also what will make AI smaller, cheaper, and available everywhere.

So, that's why I'm bullish no matter how the tension between memory cost and better algorithms plays out.

Because I believe demand from new AI applications and devices will grow faster than manufacturers can add supply.

Again, technology has a direction.

AI is getting bigger (and smaller) at the same time.

Massive data centers. Self-driving cars. Robots.

Intelligent edge companions small enough to fit in the pockets of 8.3 billion people at home and at work.

Technology wants to be everywhere.

And everywhere it goes, it needs memory.

Manufacturing this memory is limited to just a handful of companies.

That's why these five stocks belong on your research list.

Micron, SK hynix, and Samsung are the dominant memory manufacturers.

Sandisk and Kioxia are major players in flash storage.

All positioned along one of technology's most important growth paths.

And next issue, we'll explore the coming AI companion device disruption.

What happens when a little device with eyes becomes a mentor to billions of people?

That's the subject of a twenty-thousand-plus-year-old apprenticeship...

Always be prospering,

Harry Seldon
Moonshot Minute

🔓 Premium Content Begins Here 🔒