The Memory Cycle Has a Clock. Here's Where the Hands Are Pointing.

“GUYS GET ON THE MEMORY HYPE TRAIN BEFORE WE TAKE OFF! ONE LAST CHANCE!!”

Sounds familiar?

Because that’s exactly what’s happening in memory stocks right now. And many of my students have been asking me this, exceptionally even more the past week.

That feeling where everything looks incredible, the numbers are breaking records, the stocks are surging, and a little voice in the back of your head starts saying: “What if this time really is different?”

And there you have it.. FOMO.

I’ve felt that feeling before. Usually right before a painful lesson.

So instead of thinking “should I buy?” which won’t give you any answers, I started asking a different question instead: where are we actually in this cycle?

I’ve been doing deep research into it, and the more I dug, the more I felt something familiar creeping in.

Here’s what I found. And trust me, you’ll want to read till the end.

SK Hynix just reported Q1 2026 results that were genuinely jaw-dropping.

Revenue of ₩52.6 trillion. Operating profit of ₩37.6 trillion at a 72% margin. That single quarter generated more operating profit than the entire year of 2024.

In what is normally their weakest quarter of the year.

Micron isn’t far behind. Revenue nearly tripled year on year to $23.9 billion. Gross margin hit a company record of 75%. Their CEO said they can only fill 55 to 60% of core customer demand.

The AI memory supercycle is real. The question isn’t whether the business is good. The question is what’s already priced in, and what happens next.

Rather than listening to what analysts are saying or letting Reddit “to-the-moon” thinking guide you to blind faith, the proof is in the data. The kind of data that tells you where we actually are in a cycle, not where people feel like they are.

So here are six specific indicators across inventory levels, pricing trends, supply capacity, capital expenditure behaviour, revenue momentum, and the supply pipeline timeline that I found, each one telling you a different part of the story.

Together, they paint a picture that I think every investor in this space needs to understand before FOMO-ing into anything.

As of January 2026, Samsung’s DRAM inventory had fallen to just 6 weeks of supply, roughly half the typical 10 to 12 weeks. SK Hynix is even tighter, with DRAM inventory at 2 to 3 weeks and NAND at 3 to 4 weeks. Both are at historically low levels.

The situation became so acute that by late January 2026, all three major suppliers, Micron, SK Hynix, and Samsung, began requiring customers to disclose their end customers and actual order volumes. They were screening for hoarding. When suppliers start verifying whether your demand is real before they will even sell to you, that is not a market close to peak. That is a market in the middle of a supply crisis.

Enterprise contract prices rose 90 to 95% in Q1 2026, with another 58 to 63% projected for Q2. Samsung and SK Hynix are now signing 3 to 5 year agreements with Microsoft, Google, and Amazon, requiring advance payments just to secure allocation.

But in March 2026, something shifted. Consumer spot prices fell for the first time in nearly a year. DDR5 retail kits dropped 20 to 27% in Chinese channel markets. The trigger was Google’s TurboQuant announcement, a memory compression algorithm that spooked stockpilers.

The split is now visible. Enterprise locked in and still surging. Consumer spot showing its first crack. In every prior memory cycle, consumer softens first. Enterprise follows later.

IDC published a formal report in February 2026 calling this not just a cyclical shortage but a “permanent, strategic reallocation of the world’s silicon wafer capacity.” Every wafer going to an HBM stack for an Nvidia GPU is a wafer denied to a smartphone or laptop. The entire production priority of the industry has inverted in two years.

Micron can fill only 55 to 60% of core demand. Samsung can fill only 70%. Together these three companies control over 90% of global DRAM. When every supplier in the oligopoly is simultaneously capacity-constrained, there is no release valve. Demand is growing 35% this year. Supply only 23%. That gap cannot close in 2026.

This one surprised me the most, because it’s where this cycle genuinely looks different from history.

In the 2017 to 2018 memory boom, Samsung raised capex by over 50% year on year. That flood of new capacity is exactly what caused the brutal 2019 bust. The industry essentially crashed itself by overbuilding in a period of euphoria.

This cycle? SK Hynix is up 17%. Samsung up 11%. Both are focused on process upgrades and advanced packaging, not raw capacity expansion. Even more telling: SK Hynix quietly cut its HBM-specific capex in 2025 because internal modelling flagged potential oversupply risk in 2027. That is not the behaviour of a management team drunk on boom profits. That is capital discipline. And it is one of the main reasons this cycle has lasted longer and stayed more rational than the prior ones.

SK Hynix posted 198% revenue growth year on year this quarter. Micron nearly tripled revenue. These are extraordinary numbers by any measure.

But here is the uncomfortable mathematical reality that most retail investors miss.

Those numbers are now the comparison base. Even if both businesses stay at exactly current levels of revenue and profit, the year on year growth rate mathematically trends toward zero by early 2027. And the stock market does not price absolute levels. It prices the rate of change.

The moment revenue growth decelerates from 198% to even 80% (which it must), momentum investors begin to exit. Even if the business is still excellent. Even if earnings are still at record highs. The stock can correct 30 to 40% not because anything broke, but because the rate of growth slowed.

This is the second derivative trap. Understanding it is what separates investors who hold through corrections from investors who panic at exactly the wrong moment.

SK Hynix’s new mega-fabs in Korea and the US Yongin cluster are scheduled to begin production in 2027 at the earliest. Micron’s new Japan plant comes online in late 2028. Samsung’s expanded lines are similarly timed.

The capex decisions being made right now, all those billions of dollars being committed across three suppliers, will translate into actual wafers in 2027 to 2028. Deloitte has already flagged the overcapacity risk in their semiconductor outlook. Not for 2026, which remains tight. But for the 2027 to 2028 window.

This is not a reason to panic. The cycle still has runway. But it is a reason to understand that the clock started ticking a while ago. Every quarter that passes without a supply correction is one quarter closer to when the new supply arrives.

Taken together, those six indicators tell me we are in mid expansion, roughly 50-60%. Not the early innings, but also not crash territory. Somewhere in the middle, with a business cycle that still has 12 to 18 months of strong runway, and a stock price cycle that may have less because the market is always forward looking.

Enjoyed this? Subscribe to get posts like this straight to your inbox every week.

The semiconductor AI trade isn’t one story. It’s three waves, each at a different stage.

Wave 1: 2023 to 2024. The GPU trade. This was Nvidia, TSMC and AMD. If you were in, you made extraordinary returns. Nvidia is up 800%, TSMC up 300%, AMD up 400%. If you missed this wave, chasing it now at these valuations, in my opinion is a high risk entry point.

Wave 2: 2025 to 2026. The memory and infrastructure trade. Micron, SK Hynix, Marvell, Broadcom. This is the wave that’s live right now. Some names have already run hard. But the rotation into this layer only really started in late 2025. It’s mid-innings, not end game.

Wave 3: 2026 to 2028. The application and edge wave. AI software, edge compute, automotive chips, industrial AI. This also includes software companies, and it be a make-or-break them depending on how well they adopt AI in their businesses.

Most retail investors I speak to are trying to get into Wave 1 right now, when Wave 1 has been over for a year. Or they’re adding aggressively to Wave 2 at all-time highs, when Wave 2 is already well priced. The patient money is starting to look at Wave 3 while everyone else is still crowding into Wave 2.

When you see Intel up 131% in a month, Micron up >100% year to date, the SOXX index having its best month in history, it’s very hard not to feel like you’re standing outside the party looking in through the glass. I’ve been there, and you probably have too.

But I want to offer you the same reframe I had to give myself after weeks of this research.

The “this time is different” feeling is one of the most tricky feeling to deal with. And don’t get me wrong - it’s not because the AI narrative & underlying tech is baseless. But because when everyone can see the opportunity clearly, most of the easy money has already been made.

Cars transformed civilisation. If you bought Ford stock at the 1920 peak, you waited 40 years to break even. The technology won. The investors who bought at peak euphoria didn’t.

I’m not saying we’re at that level. The memory supercycle has real runway. But the question is not “is this a good sector?” - The question is “what price am I paying for it, and what margin of safety do I have if I’m wrong about the timing?”

If you already own these names, you are already on the boat. You don’t need to jump. Sit tight and let the thesis play out.

If you’re considering adding at all-time highs purely because everything is going up, then you got to ask yourself if you are just jumping on the boat because “the grass is greener on the other side”.

Three signals will tell me whether the supercycle is extending or quietly topping.

Hyperscaler capex. The moment Microsoft, Amazon, or Google says “we’re trimming AI spend”, it’s a signal that perhaps the cycle is turning. Until then, the demand floor is intact.

HBM pricing commentary. Specifically the gap between enterprise contract prices and consumer spot prices. If spot starts bleeding into enterprise territory, that is the first signal of supply catching demand.

Samsung’s HBM recovery. If Samsung fixes their yield issues and takes back meaningful HBM market share from SK Hynix in 2026, the pricing power dynamic shifts. Their quarterly updates are worth watching closely.

None of those have broken yet. Which is why the thesis is still intact.

But the clock is ticking. And knowing that changes how you position, how much you size, and how you manage your own plan regardless if a correction or continued rally happens.

Enjoyed this? Subscribe to get posts like this straight to your inbox every week.

And always remember -

Patience builds wealth,
Bjorn

If you want to go deeper on this, I break down the market every week on YouTube.

For quick takes when news breaks, follow me on TikTok.

For real-time updates and analysis as things unfold, join my Telegram community.