Micron at $1 Trillion: The Memory Maker NVIDIA Can't Run Without

From a Dentist's Basement to the Trillion-Dollar Club

On October 5, 1978, four engineers - among them twin brothers Ward and Joe Parkinson - started a chip-design shop in the basement of a dentist's office at the corner of Cole and Northview in Boise, Idaho. While Silicon Valley chased processors, the part of the computer that thinks, they bet on the part nobody wanted: memory, the part that remembers.

On Tuesday, May 26, 2026 - the day after Memorial Day - that company became the first pure-play memory maker in history to cross a $1 trillion market cap. The stock rose +19% in a single session. A year ago, Micron was worth $108 billion. The trip from the dentist's basement took 48 years. The last nine-fold leg took twelve months.

We used Barebone AI to pull apart the two documents actually driving this - Micron's fiscal Q2 2026 earnings call from March 18 and the UBS note that lit the fuse on May 26 - and then checked the tape, the filings, and the insider record behind them.

Here's the short version. The CEO told Wall Street he can only fill 50% to two-thirds of some key customers' orders. Gross margin is guided to ~81% - twenty points above the best quarter of the last boom. UBS tripled its price target to a Street-high $1,625. And every reported C-suite insider transaction for the past year has been a sale.

All four of those things are true at once. That's the whole Micron debate.

What Micron Actually Sells

Think of an NVIDIA GPU as the brain of AI. A brain without blood is dead. Memory is the blood - it feeds data to the processor fast enough for it to think. Every flagship NVIDIA data-center chip ships with stacks of memory bonded right next to the processor, because a GPU that has to wait for data is just expensive sand.

That stacked product is HBM - high-bandwidth memory: dies of DRAM, the standard working-memory chip in every computer, piled on top of each other like floors of a skyscraper and wired together vertically. It is brutally hard to manufacture, and exactly three companies on Earth can do it at scale:

Company	Listed	HBM share 2025	HBM share 2026 (proj.)
SK Hynix	Korea (KRX)	59%	50%
Samsung	Korea (KRX)	20%	28%
Micron	US (Nasdaq)	21%	22%

Micron is the only one of the three you can buy on a US exchange - which matters more than it should, because when American capital wants the AI-memory trade, Micron is the only native ticker. Beyond HBM, the same company sells DDR5 (standard server memory), low-power memory modules for AI servers, graphics memory, and data-center SSDs. If it stores a bit inside an AI data center, Micron sells a version of it.

One more piece of context the chart can't show: in late 2022, this stock traded at roughly $49, and management was telling investors the memory industry faced its worst downturn since the financial crisis. On May 28, 2026, it closed at $923.52 - about 19x off that low. Remember the 2022 quote. It's coming back at the end.

Anatomy of a Melt-Up

The trillion-dollar milestone wasn't a slow climb. It was a detonation.

Micron opened April at $367.85. By the May 28 close it was $923.52 - +151% in two months, including +79% in May alone. The final push came from a single research note: on May 26, UBS analyst Timothy Arcuri raised his price target from $535 to $1,625, a 204% hike and the new Street high, implying a market value around $1.8 trillion.

His argument wasn't really about demand. It was about the kind of company Micron is becoming. Memory contracts used to be quarterly or one-year volume deals, repriced whenever the market turned. The new long-term agreements run three to five years, with fixed volume commitments and partially fixed pricing. Arcuri's conclusion:

"No reason why MU should trade a whole lot differently than NVDA in terms of P/E."

That sentence is the entire bull case in eleven words. Memory companies have always traded at single-digit multiples of next year's earnings, because next year's earnings were a coin flip - SK Hynix trades around 6x forward earnings even now, Samsung roughly the same. UBS is arguing the contracts have changed the coin: apply a 15x multiple to Micron's new earnings power and you get $1,625. The market spent the last week of May deciding how much of that re-rating to believe.

"We Can Only Fill Half Your Order"

Strip out the price action and look at what the business actually reported on March 18. The numbers barely look legal for a memory company:

Metric	FQ2 2026 (reported Mar 18, 2026)
Revenue	$23.9B, +196% YoY
Non-GAAP EPS	$12.20 vs $8.79 expected
Gross margin	75% non-GAAP (74.4% GAAP, vs 36.8% a year earlier)
DRAM pricing	up mid-60s percent in one quarter
FQ3 revenue guide	$33.5B ± $0.75B
FQ3 gross margin guide	~81% non-GAAP
FY2026 capex	raised above $25B, from $13.8B in FY2025

A quarter of growth like that usually means a one-off. This one came with a confession instead. CEO Sanjay Mehrotra, on the call:

"Some of our key customers, we are able to fulfill only 50% to two-thirds of their demand in the medium term."

Read that again. The customers are the largest AI builders on the planet, and the message is: we cannot make enough, and we won't be able to for a while. TrendForce's read of the same quarter noted that as fabs pivot capacity toward HBM, prices for standard memory keep rising too - the shortage in one product tightens all the others, and warnings across the complex now stretch the tightness into 2027.

When demand outruns supply this badly, prices don't just rise - they get contracted. The same call disclosed something the memory industry had never seen: Micron's first five-year strategic customer agreement, with robust volume and supply commitments, in an industry where deals historically reset every twelve months. More are in negotiation. And the customers aren't just buying whatever's available: Mehrotra noted that "a clear majority of our customers rank Micron Technology number one in quality."

This is what UBS means by a different kind of company. The old Micron sold a commodity at spot prices. The new one is selling multi-year capacity, pre-committed, to customers who can't get enough of it.

A Bet on One Chip vs a Bet on Memory Itself

Here's the comparison most investors skip. SK Hynix is the HBM king - roughly 59% share in 2025, and UBS expects it to keep the largest slice of HBM4, the next generation now ramping for NVIDIA's Rubin platform. Its record profits are, to an unusual degree, one product's story. SK Hynix's own 2026 outlook calls this an "HBM-led" supercycle.

Micron holds about a fifth of HBM - management says its HBM share reached parity with its overall DRAM share during calendar 2025 - but its AI exposure is deliberately wider:

• HBM3E and HBM4, both shipping in calendar 2026
• Low-power server memory, where a new 256GB module puts 2TB of capacity beside a single CPU - quadruple the content of a year ago
• DDR5 for conventional servers, repricing upward with the shortage
• Data-center SSDs, where revenue more than doubled sequentially to a record

So the cleaner way to frame the pair: SK Hynix is concentrated leadership in the single hottest product in semiconductors. Micron is a diversified claim on every type of memory an AI data center consumes - with the only US listing among the three companies that matter. Neither is the "safe" version. They're different bets on the same shortage.

The Industry That Always Blows Itself Up

Now the section that earns this article its keep - because Wall Street has heard "memory is different now" before, and it has never once been true.

Memory is the most violently cyclical large business in technology. High prices trigger capacity expansion; capacity arrives in a wave; prices collapse; repeat. Here is Micron's own gross margin - the share of each revenue dollar left after manufacturing costs - through two full cycles and into this one:

In fiscal 2018, gross margin hit 59% (the peak quarter touched 61%), and the consensus was that industry consolidation had finally tamed the cycle. By fiscal 2020 margins had been cut nearly in half, to 30.6%. The next boom topped out at 45% in fiscal 2022 - and then fiscal 2023 delivered something the bulls said couldn't happen: -9.1%. Negative gross margin. Micron was selling chips for less than it cost to make them, and management called it the worst downturn since the financial crisis. That was the $49 stock.

Three years later it's guiding 81%.

The bull answer is that the five-year contracts exist precisely to break this loop - locked volumes and partially fixed pricing mean the next downturn gets absorbed by the contract book instead of the income statement. Maybe. But notice what else is in the same earnings call: capex raised above $25 billion this year, with construction spending guided to step up another $10 billion-plus next year - and Samsung, whose HBM share is projected to jump from 20% to 28% in 2026, is expanding too. Every memory bust in history was funded by exactly this kind of boom-time capex. The supply wave is being poured right now; the only question is whether demand keeps outrunning it. (Software is the other tail: a single Google compression paper knocked the whole memory complex down 3 - 6% for two sessions in March.)

And then there's the insider tape. Over the past year, every reported C-suite transaction in Micron stock has been a sale - not a single open-market purchase. Most of it is mechanical: Mehrotra's September 2025 sale of 7,500 shares ran through a 10b5-1 plan adopted back in August 2024, the kind of pre-scheduled selling executives use to diversify. He still holds over a million shares directly and in trust. But two things can be true: the selling is routine, and nobody in the C-suite looked at this stock all year and decided to buy more. The wry footnote - those September shares went at about $158. The stock has since risen almost sixfold. Even the people who know Micron best priced this supercycle wrong, in the cautious direction.

What This Means

At $923.52, Micron is no longer a bet on whether the AI memory shortage is real. The shortage is documented, contracted, and priced. The bet is now about duration: whether five-year agreements genuinely killed the most reliable boom-bust cycle in technology, or merely scheduled the bust for later.

The framework we'd watch, in order:

1. The FQ3 report in late June. A $33.5B revenue guide and 81% margin guide leave zero room for "slightly worse than expected."
2. HBM4 execution. Micron is shipping it this year into NVIDIA's next platform; yield stumbles hand share straight back to SK Hynix.
3. Samsung's ramp. The projected 20% to 28% share jump in 2026 is the first real supply response. Watch whether HBM pricing holds as it lands.
4. The contract book. Every new multi-year strategic agreement extends the floor under earnings; if announcements stop, the re-rating thesis stalls.
5. Standard DRAM spot prices. They rose mid-60s percent in a single quarter. The first sustained decline is the cycle clearing its throat.

The dentist's-basement company has survived every memory winter since 1978 - that's precisely why it's the last American DRAM maker standing. The question UBS just put a $1,625 number on is whether winter itself has been cancelled. History says no. The contracts say maybe. The margin chart says you should at least know what the last two "this time is different" episodes looked like before deciding.

---

Data: Barebone | Sources: Micron FQ2 2026 earnings call (March 18, 2026), Micron SEC filings (FY2018 - FY2025), UBS research note of May 26, 2026 as reported by Bloomberg and Reuters, CNBC, TrendForce, BoiseDev | Prices as of May 28, 2026 close