Micron’s bull case rests on a belief the market formed during a shortage and hasn’t updated. The HBM scarcity narrative — supply constraints, hyperscaler desperation, Micron as the American alternative to SK Hynix — was true in 2024. Whether it holds at $378 and a 42.4% capex-to-revenue ratio is a question the consensus hasn’t answered.
Start with the price. $378. For that to be defensible, work backward: Micron needs to sustain revenue somewhere near its $37.4 billion run rate per its 2025 filings, keep operating margins in the mid-to-high twenties, and avoid the thing that has destroyed memory investors in every cycle — the moment when supply catches demand and surpasses it. The operating income jump from $1.3 billion in 2024 to $9.8 billion in 2025 is large. But it happened on the back of HBM pricing power, which exists only as long as supply is constrained.
The market has decided that constraint is structural. That’s the belief to examine.
Where the Shortage Story Came From
HBM — high-bandwidth memory, the stacked DRAM architecture that sits on AI accelerators — had scarcity in 2023 and early 2024. SK Hynix was the only vendor with mature HBM3E yields. Samsung was struggling. Micron was qualifying. Nvidia’s allocation decisions were driving anxiety among hyperscalers. That was the environment in which the market priced Micron as a beneficiary not of the cycle but of a structure. The logic: AI demand is infinite, HBM supply is complex, therefore margins hold.
That logic originated in a time-limited moment. Micron’s HBM yields have matured. Samsung has stabilized its process. SK Hynix is running at utilization and expanding. The moat that made 2024’s scarcity narrative credible is narrowing — not gone, but narrowing — and the market hasn’t adjusted its habit of treating Micron as if the 2024 supply picture still applies. The capex number makes this uncomfortable. $15.9 billion in capital expenditure in 2025. That’s not maintenance spend — that’s an expansion bet, the kind you make when you believe demand justifies capacity. But every dollar Micron spends on HBM capacity is a dollar Samsung and SK Hynix are also spending. Three companies building for the same demand signal. Memory cycle historians will recognize this pattern.
What $9.8 Billion Requires
The $9.8 billion operating income figure is the number that justifies the price. Unpack it: this represents margin expansion from roughly 5% in 2024 to roughly 26% in 2025 — a recovery so sharp it looks like a different company. Micron has cycled through operating margins ranging from negative to around 30% at peak. The 2025 number is near peak. Not above it, not unsustainable for another year, but not a baseline. If HBM pricing softens 15% on competition — and HBM spot markets are not transparent, so this would show up late — operating income doesn’t fall to $8 billion. It could fall to $5 or $6 billion. The revenue line holds. The margin line doesn’t. That’s a different company at $378.
R&D-to-revenue dropped to 10.2% in 2025 from 13.7% in 2024 per Micron’s filings. The read is that scale is doing the work — same dollars, bigger denominator. That’s probably right. But it also means Micron isn’t accelerating differentiation spend at the same pace it’s accelerating production. When a memory cycle turns, differentiation is what keeps you alive.
The Nasdaq at 22,018, down from a 52-week high of 24,020, creates a pressure the bull case tends to bracket away. Multiple compression in a risk-off tape hits high-capex names hardest. Micron’s valuation is not a function of its earnings trajectory alone — it’s a function of what multiple the market will apply to cyclical-but-maybe-structural earnings in a tape that’s been selling off for three months. The bull case on fundamentals can be correct and the stock can still underperform if the multiple compresses.
If Samsung accelerates HBM3E qualification faster than its roadmap suggests, and if SK Hynix expands capacity beyond hyperscaler contracts, the oversupply moment could arrive in 2026 rather than 2027. Add any softening of AI infrastructure spend — a hyperscaler capex pause, a model efficiency breakthrough that reduces memory-per-GPU requirements, a broader enterprise budget freeze — and the margin story unravels before the capacity story plays out. The bear case doesn’t require a collapse. It requires two or three things that are in motion to move faster than the bull case assumes.
The earlier framing was built around supply constraints as the durable thesis — that still holds for now, but the runway looks shorter than it did. The constraint is present. The question is whether it lasts long enough for the capex cycle to pay off before competitors close the gap.
The bull case goes like this: AI infrastructure spending is not discretionary — it is survival for every hyperscaler. Micron is one of three companies that can produce HBM at scale. The complexity of HBM3E and HBM4 manufacturing provides a longer lead time before commodity dynamics assert themselves than DRAM cycles did. And the 42.4% capex ratio is what you spend when you’re trying to lock in supply agreements with customers who need allocation certainty. You don’t spend $15.9 billion without having conversations with your customers first. The assumption most likely to break: that HBM’s manufacturing complexity provides a lasting shield against the commodity pricing dynamics that have governed every memory cycle before it.
The earnings recovery is here. The demand signal is here. The assumption that HBM will commoditize on the same timeline as DRAM may be wrong. But the market is pricing Micron as if the scarcity of 2024 is a permanent feature of the landscape, and it formed that view during an episode that was always going to be temporary. The 2024 supply constraint is reflected in the stock. The risk of 2026 oversupply is not.
Wall Street spent two years telling you memory was a trash business, then spent one year telling you it was the only business — and somehow both times you were supposed to pay full price.