SanDisk just built a valuation floor. SK Hynix just poured the ceiling

Yesterday the memory question was whether SanDisk is a contracted utility or a cyclical in disguise. Today both sides of that question got a fresh, hard number, and they point in opposite directions.

SanDisk answered the durability question with cash, not words. It guided Q1 FY2027 revenue to $10.3B-$10.8B with non-GAAP EPS of $44-$46, and, more tellingly, expanded its buyback authorization by $15.5B. A buyback that size is a management statement about the floor: it says the company believes post-earnings dips are a buying opportunity for its own balance sheet, and it puts real money behind the "utility, not cyclical" claim. The market's discrimination was visible in the tape too. SanDisk fell about 3% on guidance that merely missed elevated buy-side hopes, while legacy peer Western Digital fell nearly 8%. The pure-play NAND story is now trading as a distinct, higher-quality thing than the HDD-legacy story it separated from.

SK Hynix answered the supply question by committing $38B to build more of it. Its Q2 was enormous, revenue of 79.32T won up 257% year over year with operating profit of 60.5T won against 9.2T won a year earlier, yet it still fell about 10% on the exchange for missing a 64T-won consensus. That is the same "beat and bleed" reflex hitting Asia. The part that matters two years out is the capex: KRW 54T, about $38B, roughly two-thirds toward the Yongin Y2 DRAM plant that starts construction in July 2027 and completes June 2029, plus the Cheongju M17 NAND fab. Demand is currently locked, industry billings ran up 134% year over year in June and volumes are pinned by multi-year long-term agreements with around ten hyperscale customers. But locked demand and a $38B greenfield supply wave landing in 2027-2029 are the exact ingredients of a classic memory over-build.

So the trade splits on a timeline, not a direction. Through the current agreement window, the take-or-pay contracts and prepayments do the work, and the buyback marks the floor. Into 2027-2029, the new fabs test whether that contracted demand can absorb capacity that was greenlit at the top of the cycle. The bull and bear are not really disagreeing about whether memory is in demand now, they are disagreeing about whether the contracts outlast the fabs.

Why it matters (allocator lens): this is the difference between a business you can hold and a moment you have to time. A $15.5B buyback is a floor mechanism you can point to; a $38B fab is a supply risk you have to schedule. The memory complex is offering utility-like cash flows for the next several quarters and a cyclical-oversupply question for the years after, in the same names, at the same time. You cannot own the first without inheriting the second.

So-what for ABC. Alpha sleeve, and size it as the high-volatility exposure it is. The variable to track is not the stock price, it is contract quality and the fab calendar: the share of revenue under prepaid, locked-volume agreements, and how much of the $38B (and the rest of the industry's greenfield capex) actually lands on schedule versus slips. The buyback tells you management sees a floor; the fab timeline tells you where the ceiling is being built. Let disclosures about contract renewals and construction milestones move your conviction, not the target ranges the sell-side keeps re-rating. Neutral, because the contracted present and the over-built future are both real and simply have different due dates.

Takeaway: SanDisk marked the floor with a buyback and SK Hynix poured the ceiling with a fab. Own the contracted window if you want it, but put the 2027-2029 supply wave on your calendar the day you buy.

Analytics & education, not advice. DYOR.