I bought Micron at $668. Here’s why I’m not buying at $881.

How a stampede starts

Wall Street analysts capitulate in a predictable way: not all at once, but in a sequence that makes each individual decision look reasonable. Nobody wants to be the first to break from consensus. But nobody wants to be the last one defending the old view either.

Mizuho moved first, on May 6, bumping their PT from $545 to $740. A respectable raise. Nothing dramatic.

Then on May 13, Bank of America raised their PT from $500 to $950. That’s a 90% increase in a single note. DA Davidson initiated coverage the same day at $1,000. TD Cowen raised to $660. Three banks moving in one session. The dam had a crack.

On May 19, the crack became a flood. Citigroup went $425 to $840. HSBC went $750 to $1,100 — landing exactly on my number. Mizuho raised again, to $800. Melius Research the next day matched HSBC at $1,100.

Then on May 26, UBS broke the format. $535 to $1,625. A 204% increase. The largest single-note PT raise I’ve tracked. And just to make the point, UBS’s implied market cap of $1.8 trillion would make Micron the seventh-largest company in America — ahead of Tesla, ahead of Meta, ahead of Berkshire Hathaway.

Twelve hours after the UBS note, Micron crossed a trillion dollars in market cap for the first time in company history. It didn’t stay there long. But that’s not the point. The point is that twenty days earlier, the consensus thought it would never get there.

What actually happened

It’s tempting to say I was right. I’ll resist the temptation, because what actually happened is something more interesting and more useful to understand.

The variant view wasn’t a prediction. It was an observation. Wall Street had a math problem. Earnings estimates had moved aggressively. Price targets had not. Sooner or later, one of those two things had to give. Either the analysts would have to come back down on EPS (justifying their low PTs), or they’d have to raise PTs to match the EPS they had already raised. The thesis was simply that, given the data flowing in — DRAM contract prices up 90-95% in a single quarter, hyperscaler capex guides of $745-775 billion, MU printing 75% gross margin in real time — the EPS direction wasn’t going to reverse. So the PTs would have to move.

This is not a forecast that requires being smart. It requires being patient and refusing to rationalize. The math wins eventually. It just doesn’t tell you when.

Chart 2 — Why the EPS estimates were moving

Q1 2026 DRAM contract prices +90-95% QoQ — the largest quarterly increase ever recorded.

When prices move like that, earnings models update. Eventually price targets do too. Q2 2026 is projected another +58-63% on top. Source: TrendForce contract pricing reports.

Why the cycle didn’t end (yet)

Every previous memory cycle ended because supply caught up to demand. Sometimes demand fell off a cliff — 2018 (smartphones), 2022 (gaming and crypto). Sometimes supply rushed in too fast — 2008 (Korean overinvestment). The textbook says cycles end. The textbook is right.

What’s different this time isn’t that the cycle won’t end. It’s when. AI demand is so large and so concentrated that the cycle peak is taking longer to play out. Hyperscalers committed to memory at a scale that’s still being absorbed. Microsoft alone said $25 billion of its $190 billion in 2026 capex was just paying up for components — that’s $25 billion sitting on Micron’s pricing power, in one line of one earnings call.

Chart 3 — The structural shift

Memory went from 8% of hyperscaler capex two years ago to ~30% today.

That’s not cyclical; that’s the supercycle’s economic engine. Hyperscalers told us memory was in shortage by paying up for it. Source: company disclosures, author estimates.

So the bears were technically right (cycles end) but practically wrong (the timing was off by several quarters). The bulls were technically wrong (cycles end) but practically right (you make most of the money on the way up). The trade was between the two views: long the durability of the peak, short the permanence of it.

UBS thinks this time is different. I don’t.

The new bullish argument — the one UBS used to justify $1,625 — is that AI has “permanently reshaped memory market fundamentals.” That phrase deserves to be examined.

If you believe it, memory becomes a structural growth business. Recurring AI infrastructure demand. Long-term agreements that smooth pricing. A tri-opoly of Samsung, SK Hynix, and Micron with capital discipline. Apply a 16× multiple instead of 11×. Apply growth-stock economics. The PT goes from $1,100 to $1,800.

If you don’t believe it, memory is still memory. The cycle still ends. The only question is what triggers the end, and when.

I’d give UBS partial credit. The 5-year long-term agreements Micron signed earlier this year are real and remove some of the spot-market volatility from peak pricing. The HBM/DDR5 wafer-area math is genuinely structural — every Gb of HBM displaces three Gb of conventional DRAM, which tightens the entire DRAM market mechanically as AI accelerator volumes grow. CXMT, the Chinese hopeful, is at least two generations behind and under export-control pressure. These things matter.

What I don’t give UBS credit for is the supply ramp. Samsung committed $73 billion to 2026 semiconductor capex — the largest single-year semi investment in history, up 128% year-over-year. SK Hynix’s $12.85 billion advanced packaging plant goes live in Q1 2028. Micron’s CHIPS-funded Idaho fab ramps through fiscal 2029. TSMC’s CoWoS capacity, the binding constraint on HBM packaging, expands fivefold from 2024 to end-2027.

Chart 4 — The supply response

Samsung’s $73B 2026 capex is the largest single-year semi investment in history.

It isn’t capex. It’s a declaration that Samsung intends to recover market share. That capacity has to come online somewhere. Memory cycles end when supply meets demand. Source: company disclosures.

None of that capacity hits the market in 2026. Some of it hits in 2027. Most of it hits in 2028. And the consensus path for hyperscaler capex growth — currently +51% in 2026 — decelerates to +13% in 2027 and +5% in 2028. If supply grows 30% per year in 2027-2028 while demand growth slows to 5%, you don’t need to forecast a recession to see what happens. You just need to know how subtraction works.

Why I’m not raising my price target

I get this question a lot now. UBS at $1,625. DA Davidson up to $1,500 the same week. HSBC and Melius at $1,100, matching me. Stock through $920 by Friday, looking like it could blow through $1,000 in a few weeks. Why not move my number up?

Three reasons.

First, methodology is methodology. My $1,100 came from consensus FY27 EPS times the median multiple Micron has historically traded at during recovery years. The math drives the number, not the other way around. If I raise the multiple to chase the street, I’m back-fitting — picking the answer first and then finding a justification. That works once, by luck. It doesn’t work as a process.

Second, UBS’s $1,625 implicitly assumes the cycle doesn’t end. That’s a stronger claim than I’ve been willing to make. I’ve explicitly modeled this trade as a cycle peak in fiscal 2027 with an exit by fiscal 2028 because the supply math is real and quantifiable. If I promote UBS’s view to my base case, I’m abandoning the framework that worked. The framework worked because it was honest about what it didn’t know.

Third, if I keep $1,100 and the stock blows through it, my pre-committed trim ladder catches the upside automatically. I don’t need a higher headline number. I need to execute the discipline I already wrote down.

The trim ladder, in case you forgot

Here’s what I committed to in early May, when this thesis was new:

Notice what this framework does. It commits me to selling into strength before I know whether UBS is right. If UBS is right and the stock runs to $1,800, I still capture most of that move on the way up — but I trim along the way so I don’t ride the full cycle exit on the way down. If UBS is wrong and the stock peaks at $1,150 before rolling over, I’ve already taken two of three trims by then.

The discipline doesn’t require predicting the top. It requires admitting in advance that you can’t predict the top.

What could still go wrong

The bear case has been delayed, not disproven. Some things I’m still watching.

Chart 5 — Where the supply ramp arrives

Through 2026, demand outruns supply. 2027 they cross. 2028 supply leads.

That’s the cycle exit window — and the reason we sell into strength, not on the way down. Source: TrendForce, Counterpoint, company guidance, author projections.

The 2028 supply ramp is the strongest of the bear arguments and the one I’ve already discussed. Samsung’s $73 billion arrives in late 2027. SK Hynix’s new packaging plant arrives Q1 2028. Micron’s Idaho fab arrives in fiscal 2029. If hyperscaler capex decelerates the way consensus expects, the balance flips by late 2028.

Geopolitical escalation is the tail risk that nobody can model. Taiwan disruption catastrophically affects TSMC CoWoS, which catastrophically affects HBM, which catastrophically affects Micron’s economics. I have no way to handicap this beyond knowing it’s not zero.

Politically, the AI windfall tax narrative is still floating around. South Korea officially denied it in May, but the seed is planted. Watch for European or Japanese echoes. Post-November 2026 mid-term elections, the framing becomes politically palatable in the US too, especially if a few hyperscaler CEOs say something tone-deaf about their cash positions.

Workload architecture shifts — in-package SRAM, optical interconnects, processing-in-memory — could compress the HBM bandwidth premium over time. None of these are near-term. All of them are long-term.

And then there’s Micron itself. The Idaho fab ramp is the largest greenfield project in company history. CHIPS-backed capex of roughly $50 billion over twenty years. Project delays, yield issues, or capital allocation missteps would compress the FY27-FY29 earnings path that this entire trade depends on. Companies of Micron’s size and complexity have execution risk that doesn’t show up in spreadsheets until it’s already shown up.

None of these have fired. All of them could.

What I keep coming back to

Memory has been a graveyard for confident analysts. The 2018 cycle had everyone certain of a structural rerate. By 2023 the same names were printing negative gross margins. Five-year-ago Brandon would have looked at the UBS note this week and felt vindicated. Two-years-ago Brandon would have used it as justification to load up at $900. Today Brandon notices the pattern and reaches for the trim ladder.

The trade isn’t about being right. It’s about having a framework that survives both being wrong AND being right too early. The framework that worked in May at $668 is the same framework that says trim at $900 in late May. Refusing to update the framework just because the market caught up — that’s the discipline.

I was a buyer at $668. I was a holder at $881. I was a trimmer at $923 last Friday. I’ll be a trimmer again at $1,100, and an exiter at $1,500. None of this is about courage or conviction. It’s just arithmetic with a calendar attached.

UBS may end up being right that this cycle is different. If they are, I’ll miss some upside. That’s the cost of refusing to chase. I’ll take it.

Position summary