HAUNTED BY THE CYCLE

(MU), (NVDA)

There’s a special place in semiconductor hell reserved for the phrase “this time it’s different.” 

I’ve heard it too many times. They said it in 2010 before NAND fell off a cliff. They said it in 2018 before DRAM imploded. 

And they’re saying it again now, as Micron (MU) rides high on AI exuberance and margin ecstasy. 

Spoiler alert: ghosts do not stay buried. They come back, usually right after a CapEx binge.

Micron’s current momentum is hard to argue with. Gross margins are pushing the mid-40s, pricing power is intact, and AI infrastructure spending continues to throw logs on the demand bonfire. 

But when you zoom out, this still looks and smells like a classic memory upcycle. 

Investors piling in now on the belief that we’ve entered a structurally higher valuation regime may find themselves back in the familiar position of explaining why Micron always looks cheapest at the top.

The earnings power looks real. But it’s also inflated. Micron is not just selling more memory, it’s doing so in a market where demand has temporarily outrun supply. That’s not a durable advantage. 

CapEx data shows Micron and SK Hynix spending heavily to expand capacity. Samsung, despite stumbling with its HBM3E qualification for Nvidia (NVDA), is preparing to re-enter the battlefield with HBM4. 

That mistake with HBM3E gave Micron a window to gain share. 

But Samsung’s missteps do not repeat for long. Their entire playbook is based on volume, vertical integration, and the ability to crush margins if they want to.

When memory prices are rising, investors extrapolate. They project fat margins into the future and treat CapEx as validation. 

But the reality is simple. 

High prices incentivize oversupply. And when those new fabs come online, pricing power evaporates quickly. 

The DDR5 market is particularly vulnerable. Pricing strength there has carried earnings this cycle. The moment those prices flatten or tick down, the cycle is already in decline.

Samsung’s HBM4 strategy matters because it could front-load the next glut. Their decision to use 1c DRAM on a more advanced process could allow them to leapfrog competitors on power efficiency and speed. That’s not just a technical detail. 

If Samsung gets traction with HBM4, they will not wait to raise prices. They will undercut to grab share, even if it means bleeding margins in the short term. That is a scenario Micron is not priced for.

There’s also the question of low-power DRAM, which is migrating from smartphones into AI data centers. That shift is gaining traction faster than expected. 

Samsung’s SOCAMM2 and SK Hynix’s LPDDR6 are both aiming to position this segment as a core component of future AI architectures. 

Micron is active here, but may not be first to market. Whoever captures early enterprise adoption could shape this category and claim the early profits. 

That creates a second risk: not just pricing pressure, but competitive loss in a new memory form factor that investors are not yet pricing into their models.

Micron bulls point to long-term demand trends for AI, cloud, and edge compute. Those are real. 

But the history of memory tells us that long-term demand does not protect against short-term oversupply. It never has. And based on current investment and capacity plans, supply will catch up in 2027. 

Experts already forecast a slowdown in EPS growth after August 2026 and an outright decline into 2028. If those estimates hold, Micron is no longer a growth play by the time new fabs are up and running.

Micron is still a reasonable buy for 2026. There’s momentum, pricing support, and a tailwind from AI-driven demand. 

But don’t treat is as a secular rerating. It’s a cycle. And like all cycles, it turns. 

The trap for investors is not that Micron is suddenly weak. It’s that the story remains strong just long enough to lull them into staying too late. That’s when the compression hits.

Semiconductor memory has a long memory. We should, too. Because the only thing worse than missing the top is thinking you are early when you are already late. 

And if you listen closely, you can already hear the faint footsteps of the ghosts of gluts past.