Micron just reported a quarter that would have sounded insane eighteen months ago. Revenue of 41 billion dollars, up roughly four and a half times from a year earlier. Gross margin of nearly 85 percent. Twenty five dollars of earnings per share against one dollar sixty eight a year ago. Then it guided next quarter to 50 billion in revenue and 86 percent margins, and announced multi-year contracts designed to lock customers in. The stock dropped 13 percent the day before the report, then rose 15 percent after. Like nothing happened.

But Micron is just the one that reported this week. It's the smallest of the three makers and the only one that reports on this calendar, so it's the canary, not the story. The same forces, the same extreme margins, the same questions apply to all three, Samsung and SK Hynix included, and to the whole memory trade I'm actually long through the DRAM basket. SK Hynix posted operating margins north of 70 percent last quarter, higher than Nvidia. This isn't one company having a freak result. It's an entire oligopoly printing peak numbers at the same time.

So this piece isn't about one stock. It's about whether the entire memory re-rating is real, or whether I'm looking at the top of the most violent cycle in this industry's history and telling myself a story. I'm long. I hold DRAM and Micron, I've added on weakness, and my conviction is intact. Which is exactly why I owe it to myself to make the other side's case as well as they would make it. A thesis you can't argue against isn't conviction, it's a blind spot. Here is the strongest case that I'm wrong, and I'm not going to soften it.

Three companies are simultaneously earning software margins on what has, for forty years, been one of the most cyclical commodities on earth. That is not a sign of health. It's a sign of an extreme.

Memory has always been brutally cyclical, and the deepest pattern in the entire industry is that margins this high do not last, because they create the conditions for their own collapse. Sky-high margins pull in capacity, invite competition, and pull demand forward. Every previous time memory margins spiked toward levels anywhere near this, across all the makers at once, they reverted, and they reverted hard, often into outright losses within a couple of years. The whole sector booms together and busts together, because they all sell the same thing into the same demand.

The bull, me, wants to say this time the margins hold because demand is structural. But notice that's exactly what's always said at the top. The honest bear points out that I'm assuming a duration I cannot actually know. I tell myself these margins last two or three years. The bear says I have no idea, they could last two or three quarters, and the fact that I've quietly handed myself a comfortable multi-year runway is itself the warning. Three makers at peak margin simultaneously isn't proof the cycle is dead. Historically it's what the top looks like.

This is the argument that should worry any honest bull the most, because it doesn't even require the business to stumble.

The makers' earnings can keep climbing while the stocks fall. The market may simply refuse to pay a high multiple for peak cyclical earnings, because it's looking past them to the down cycle it believes is coming. This is the classic cyclical trap. These stocks look cheapest, on price to earnings, at the exact top, because the earnings are peaking. Memory at single-digit forward earnings is not cheap if those earnings are about to halve. So the whole basket can be printing record profits and still de-rate, together, because the market is pricing the cycle, not the quarter. Multiple compression doesn't need a bad business. It just needs the market to decide the good times are temporary. And the market has decided exactly that about memory many times before, correctly.

My whole bull case rests on one idea. That AI, and agentic AI specifically, makes memory demand structural in a way it has never been before, so the old cycle no longer applies because the world genuinely changed.

I believe that. But honesty requires admitting it's precisely what bulls believe at every cycle peak. The internet bulls were right that the internet changed everything and still lost their shirts on the price. The thing that's genuinely new, agentic AI, can be real and transformative and the stocks can still be priced for more than even that reality delivers. "History has never seen anything like this" is true. It's also exactly what you hear at the top. Both can be true at once. The newness of AI does not automatically justify the price, and letting the two blur together is the most seductive mistake available to me right now. The whole memory complex is being re-rated on the belief that this time the cycle doesn't end. That belief might be right. It has also been wrong every single time before.

The people who look at a sector that ran 100 percent, dropped 20 percent in three days, and bounced right back, and feel cautious, are not fools. They've lived through cycles I may not have. They're pattern matching to real history. A sector that swings 13 percent in a day is telling you how crowded and fragile the positioning has become. I respect that. Dismissing it as inexperience would be arrogance, and arrogance is how people give back years of gains in weeks.

Here's the honest part, because I won't pose as cautious while doing the opposite.

I've read this bear case. I think it's real. And I'm still long the memory makers, and still adding on weakness, because my conviction is that the structural demand carries this through at least 2027, and the multi-year customer agreements the makers are now signing give the near term more visibility than memory has ever had. No company has a clear view past five to ten years, not Google, not Amazon, not Meta. I'm not betting on forever. I'm betting on the next stretch, and I think the odds favor it.

But writing this down matters, because it means I know exactly what I'm watching for. The day spot prices start to free fall, not drift, free fall, the cycle is turning. The day the people I work with stop telling me about the new things they're building with AI, demand is cooling at the ground level where I can see it before the data does. And the day hyperscaler capex guidance turns down, the engine is losing power. Those are my exits. Not a scary headline, not a 20 percent drawdown, not a bear on the internet. Those three specific things.

If none of them happen, I hold and I add. If they start happening, I've already written down what they mean, so I won't be able to explain them away in the moment. That's the entire point of making the bear case now, while the numbers are euphoric and I least want to hear it. The euphoria is exactly when you write down what would prove you wrong, so that future me, the one who's emotional and fully invested, has a sober note from past me waiting.

I'm bullish on memory. I think the bears are early. But I could be the one who's wrong, and I've now made their case as honestly as I can, so I never get to pretend I didn't see it coming.

I hold a long position in DRAM ETF (ticker: DRAM) and Micron, and I continue to add on weakness. Not investment advice. Do your own research.