Micron Is Down From Its High. The Free Cash Flow Doesn't Fit the Old Fear.

Micron's stock has pulled back roughly 25% from its early-year high of $1,255, and the reason investors cite is the same one they've cited for three decades: memory chips are cyclical, cycles end, and when they do, they end badly. Michael Burry made the bet explicit with a short position near $1,052, betting the classic boom-bust pattern would overwhelm whatever AI demand narrative is in vogue now. The stock is down around 35% from the high of $1,255.

The stock pain is real. But the business numbers behind it don't match the pattern that's supposed to follow.

Over the trailing twelve months, Micron generated $26.2 billion in free cash flow. Not guidance. Not analyst projections. Reported free cash flow. On top of that, operating cash flow reached $51.4 billion, gross margins ran at 72.6% — a level historically reserved for software companies, not commodity chip manufacturers — and the company holds $25 billion in cash against $33.4 billion in total debt, leaving net debt of minus $20.3 billion.

These are not the numbers of a business halfway to a bust. They are the numbers of a business at the peak of what looks like the most profitable cycle in memory history.

The question for anyone watching this stock now is not whether the numbers are big. They are. The question is whether this cycle behaves like the old ones — and whether the $977 price today reflects a rational discount for an eventual downturn or an overreaction to a pattern that may no longer apply.

What Changed the Memory Game

For thirty years, the memory chip industry has followed a brutal rhythm: prices rise, three manufacturers overbuild, supply floods in, prices crash, margins go negative, and stocks fall 50 to 80 percent. In 2018, Micron stock dropped 56% from peak to trough; in 2022, it fell roughly 50% — with fab construction taking 2–3 years at costs of $15–20 billion per facility.

Three things are different this time.

First, High Bandwidth Memory — the specialized memory that AI chips like NVIDIA's GPUs need to function — is sold out. Not soft-sold. Sold out. SK Hynix, Samsung, and Micron have all committed their premium AI memory capacity through 2026 and into 2027. Premium AI memory inventory sold out through 2026, with the supply shortage stretching into 2027. Samsung's executive vice president warned that incremental supply is unlikely to increase through 2028 and constraints expected to become even more severe in 2027. HBM requires roughly four times the wafer capacity of standard DRAM per gigabyte, so expanding HBM production actually reduces conventional memory supply.

Second, hyperscalers are spending more than $700 billion this year on AI infrastructure, keeping Micron's HBM production sold out through 2026. Memory is the constraint, not the afterthought. Multi-year supply contracts are replacing the volatile spot market that used to create boom-bust whipsaws, giving Micron revenue visibility and pricing power it never had in prior cycles.

Third, management discipline has shifted. The three oligopolists controlling roughly 95% of global DRAM production are prioritizing profitability over market share. Micron secured over $6 billion in CHIPS Act funding for fabs in Idaho and New York, tying expansion to timelines through 2027–2028.

None of this guarantees the cycle won't turn. But it does mean that if it does, the mechanics are different. Industry analysis predicts any downturn would be shallower: a 15 to 25% revenue decline rather than 30 to 40%, with margins compressing to the 35 to 40% range rather than collapsing to negative. Historic bust scenarios called for 30–40% revenue declines and margins going negative, but the current structural shift is predicted to produce shallower 15–25% revenue drops with margins holding at 35–40%. AI inference demand creates a floor that PC upcycles and smartphone booms never did.

The Earnings Trajectory

Micron has been beating consensus estimates consistently, and the gap between expected and actual has been widening, not narrowing. In fiscal Q2 2026 — the most recent full quarter — EPS came in at $12.20 versus the consensus forecast of $9.19. The quarter before, $4.78 versus $3.96. Revenue reached $23.86 billion, up from $8.05 billion in Q2 2025, with gross margin hitting 74.4% GAAP.

The forward P/E of roughly 155x looks absurd — and it does, at first glance. But forward multiples on cyclical stocks reflect what the market expects earnings to look like at the trough, not the peak. A forward P/E of 155x on a $1.1 trillion market cap implies the consensus expects next-year earnings to be dramatically lower than the current run rate. That is the old story baked into the multiple.

The trailing P/E tells a different part of the story: 21.9x on TTM earnings. The trailing free-cash-flow multiple sits at 21.5x on $26.2 billion of reported FCF. These are not cheap multiples — but they are real multiples, attached to real cash generation, not a multiple that only makes sense if you assume earnings collapse.

What Would Prove the Fear Right

The bear argument is not wrong by default. Memory is cyclical, and cycles have existed for a reason. Here is what would validate the downturn thesis:

Hyperscalers cut AI capital expenditure plans. If Amazon, Microsoft, and Google collectively reduce their AI buildout, demand for HBM falls with it. Meta has already signaled that its data center builds may not extend beyond current plans. Meta signaled that data center builds may not extend beyond current plans. If that signal turns into an industry-wide pullback, the sold-out HBM narrative evaporates and the cycle follows its historical path.

A second risk is technical disruption. If AI architecture evolves in a way that requires dramatically less memory per inference — a breakthrough in memory efficiency, computing localization, or a new chip design that sidesteps the memory wall — the demand assumption unravels from the supply side.

The third risk is execution. Samsung resolving HBM4 yield issues and SK Hynix listing on the Nasdaq as a U.S.-traded alternative to Micron adds competitive pressure. Samsung staged a comeback in early 2026 by resolving yield issues on 12-layer HBM4 stacks. SK Hynix's Nasdaq listing coincided with Micron's sharp three-week decline. If all three manufacturers simultaneously ramp capacity in 2028, the oversupply dynamic returns.

Any one of these conditions breaking the demand-supply balance would make the forward P/E look prescient, not absurd.

Where the Free Cash Flow Points

But those are the conditions that need to break, not the conditions that currently exist. Right now, the evidence runs the other way.

Micron's balance sheet shows the company at net negative debt — $25 billion in cash against $33.4 billion of total obligations. The quick ratio is 298.5%. Return on invested capital sits at 58.6%. These are not defensive positions taken in preparation for a downturn. They are the financial characteristics of a company operating at peak profitability with a balance sheet that can absorb a rough patch without diluting shareholders or cutting the dividend — which the board just raised 30% to $0.15 per share. The Board approved a 30% increase in the quarterly dividend to $0.15 per share.

The free cash flow bridge is the hard proof here. $26.2 billion of TTM FCF, up 1,291% year-over-year, means the business is converting revenue into cash at a rate that makes even a significant downturn harder to price in. If next year's revenue falls 25% from the current peak and margins compress to 35%, the company still generates material cash. It doesn't go negative. It doesn't need to refinance. The downside is real, but it is not the 2018 or 2022 scenario.

What the Market Is Pricing

The aggregate analyst signal from AInvest still labels Micron a Buy, with a composite rating of 8 out of 10 and a fundamental rating of 4. But the forward multiple tells us the broader market has already folded a deep earnings decline into the price. The stock pulled back from $1,255 to roughly $977 — and that pullback itself may be the market's way of pricing the old cycle fear into a stock whose operating numbers haven't yet given it a reason to.

This is the tension that defines the setup right now. The market is still pricing the old risk profile — the 50-to-80% crash that ends every memory cycle — while the operating setup looks like the most protected boom in the industry's history. The sold-out supply through 2027, the $25 billion cash cushion, and the FCF machine running at $26 billion annually create a floor that the old cycles never had.

The market bar is low. The forward multiple implies earnings should collapse. If they don't — if revenue stays near current levels and margins hold above the 35% floor — the multiple compresses from 155x down toward something closer to the trailing 22x, and the stock has work to do on the upside.

Conversely, if hyperscalers pull back on capex or the supply constraint breaks earlier than 2028, the pullback deepens. The proof point to watch is not the stock price — it's the free cash flow in each coming quarter. If FCF stays above $15 billion on a run-rate basis and HBM capacity remains fully committed into 2027, the old story stays stale. If FCF falls below that threshold and contract renewals slip, the cycle has found its crack.

Either way, the cash flow will tell you before the headline does.