Memory Is Not a Commodity: Why MU and SK Hynix Are Priced for a Recession That Isn't Coming

The market just priced in a memory-cycle peak. The physical data says the opposite: HBM is sold out through 2027, conventional DRAM is in outright shortage, and the two leading memory stocks trade at 4x to 5x forward earnings while growing revenue 200% to 300% year over year.

Memory stocks just had their worst three-day rout since 2020, then ripped higher on Thursday in the biggest single-session reversal the sector has seen in years. Micron fell 18% across Monday through Wednesday. SK Hynix dropped 21%. On Thursday, both names surged 17% to 18% as a cascade of data confirmed what the selloff was questioning: the memory supercycle is not breaking. It is accelerating.

The washout was a forced deleveraging event, not a demand signal. Leopold Aschenbrenner's $24 billion fund Situational Awareness unwound all public stock positions after steep losses, and prime brokers at Bank of America, Goldman Sachs, and JPMorgan managed the margin calls. The selling was mechanical. The bounce, when it came, was violent, and it was backed by real numbers.

The Numbers That Matter

MU (Micron, $954B market cap, $871.23):

• FQ3'26 (ended May 28): $41.5B revenue, +345.7% YoY. Gross margin 84.6%. Operating margin 80.4%. Free cash flow $17.6B in a single quarter.
• Net cash: $19.6B (cash $26.0B, debt $6.4B). The balance sheet is fortress-grade.
• Trailing P/E: 19.1x. Forward P/E: 5.5x. PEG: 0.11.
• 42 analysts: consensus strong buy, mean PT $1,507 (104% upside from current levels). Range: $361 to $2,200.

SKHY (SK Hynix, $1.2T market cap, $148.10):

• Q1'26: ₩52.6T revenue (~$36.4B), +198% YoY. Operating margin 71.5%. Free cash flow ₩18.5T.
• Net cash: ₩32.5T (~$22.4B).
• Trailing P/E: 23.1x. Forward P/E: 4.2x. PEG: 0.58.
• 3 analysts: consensus buy, mean PT $282 (122% upside from current levels).

A 5.5x forward multiple on a company that grew revenue 345% year over year with 80%+ operating margins is not a growth premium. It is a recession price. A 4.2x forward multiple on the world's leading HBM supplier is a valuation that assumes the cycle is already over.

The Goldman Sachs Note: Supply Shortage, Not a Glut

Goldman Sachs dropped a comprehensive memory outlook this morning that lays out the bull case in detail:

• Conventional DRAM pricing is expected to see solid double-digit sequential growth in both Q3 and Q4 2026, driven by an ongoing supply shortage.
• HBM pricing could double next year. Goldman expects SK Hynix HBM pricing to rise 87% year over year, well above the Bloomberg sell-side consensus of 52%.
• Long-term agreements now cover over half of server DRAM, backed by sizable prepayments, take-or-pay clauses, and cancellation penalties. Coverage ratios are expected to rise.
• Chinese competitors pose a limited near-term threat due to gaps in yields, technology, and reliability.
• Capacity additions are outpacing historical levels, but bit growth is below historical averages because of the high HBM trade ratio. HBM consumes roughly three times the wafers per bit of output.

This is the structural shift that Andreas Steno was talking about when he tweeted today: "If I hear 'memory is a commodity' one more time, I am jumping off a bridge. People haven't even investigated the topic."

He is right. The old memory cycle was quarterly spot pricing, boom-bust capacity adds, and no customer lock-in. The new memory cycle is multi-year LTAs with take-or-pay clauses, contracted HBM supply sold out through 2027, and a wafer trade ratio that makes it physically impossible to flood the market with conventional DRAM while also ramping HBM. Memory is behaving like contracted infrastructure, not a commodity.

The SK Hynix Call: The Pass-Through Gap Is an Accounting Problem

SK Hynix reported Q2 earnings this week. Revenue came in at ₩79T, operating profit ₩61T, and operating margin 76.3%. The stock sold off anyway because the headline numbers missed consensus, but the miss was an accounting artifact, not a demand problem.

HBM prices are negotiated annually and locked for the calendar year. The prices SK Hynix ships HBM at today were set in late 2025. Meanwhile, conventional DRAM is repriced quarterly and surged 90% to 95% in Q1 and another 58% to 63% in Q2. HBM, the company's most advanced product, is currently selling below commodity DDR5 on a per-gigabyte basis. SK Hynix actually preferred keeping DDR5 lines running because of the margins.

When HBM reprices for 2027, that gap closes. Management confirmed on the call that 2027 HBM price negotiations are going smoothly on robust demand. They also confirmed HBM4 mass production is on track for the second half of 2026, with yields near HBM3E levels, and that volume production will ramp through 2027.

The market sold the miss. It should have bought the pass-through setup.

The Samsung Signal

Samsung Electronics reported Q2 2026 results with operating profit of ₩89.5T, above the ₩88.1T expected, and DRAM and NAND sales at record levels. According to IBTimes, Samsung warned the memory chip shortage could persist through 2028. That is not a cyclical peak. That is a multi-year supply deficit.

The $500 Billion NVIDIA Partnership

In late July, NVIDIA and SK Group announced a $500 billion-plus strategic partnership spanning AI factory construction and next-generation memory supply. SK Telecom will build a 2-gigawatt NVIDIA Vera Rubin AI factory powered by SK Hynix HBM4. NVIDIA and SK Hynix will co-develop and optimize future HBM generations together.

This is not a commodity transaction. It is a strategic partnership between the world's most valuable semiconductor company and the world's leading HBM supplier. It locks in demand for years.

The Valuation Floor

The most important number in memory right now is 3x. That is the forward P/E that pure cyclical shipping companies bottomed at in 2021 after their freight-rate supercycle peaked and collapsed. Container shipping has no AI exposure, no technological moat, and no structural growth story. It still found a floor at 3x forward earnings.

SK Hynix is already trading at roughly 4x 2027 forward earnings. MU is at 5.5x. If the market insists on valuing memory as a commodity, the valuation floor is not far below current levels. But memory is not a commodity. The LTA structure, the HBM wafer trade ratio, the NVIDIA partnership, and the Samsung supply-shortage guidance through 2028 all point to a business that is structurally more stable than any previous memory cycle.

The forced deleveraging from Aschenbrenner's fund created a price dislocation. The selloff was a leverage event, not a demand break. The physical data, from SK Hynix's 76% margins to Samsung's record profits, says the cycle is intact and accelerating.

What to Watch

• HBM 2027 price negotiations: The annual reset is the single most important catalyst. If HBM pricing closes the gap with conventional DRAM, the revenue uplift is material. Goldman's 87% YoY HBM pricing forecast is the bull case. The outcome decides whether 2027 consensus revenue needs to be revised up or down.
• MU FQ4 earnings: Micron reports in late September. The pass-through ratio — whether conventional DRAM price surges flow into reported revenue — will be the same test SK Hynix just faced.
• Hyperscaler capex: The NVDA revenue-to-capex correlation runs at roughly a three-quarter lead. If hyperscaler capex holds, memory demand holds.
• The short gamma setup: Both MU and SKHY are in short-gamma territory. MU's gamma flip is at $883, and SKHY's is at $149. Below those levels, dealer hedging amplifies moves in both directions. Above them, the regime flips to stabilizing. The bounce from Wednesday's low to Thursday's high was magnified by this dynamic. A reversal would be amplified too.

More research at bargo.ai/research.