SK Hynix at 5x earnings: the market is pricing a memory peak, not a bargain

SK Hynix reported revenue of 79.3 trillion won, up 257% year over year, and the stock fell roughly 10% ^[1][12]. The results, dated July 28-29 and described as Q2 2026, landed below a consensus closer to 84 trillion won, about 5.6% short, and concern over a slowdown in HBM4 shipments added to the selling, according to a Cryptobriefing summary of Tom's Hardware reporting ^{[2][11][18][13][16]}.

After the decline the shares traded at approximately 5 times next year's expected earnings ^[24]. That is not a valuation of a growth franchise. Put a pedestrian 10x mid-cycle multiple on a cyclical semiconductor business and a 5x forward multiple implies the market believes roughly half of next year's expected earnings will not survive ^[25]. Memory investors have run this exercise before: you do not get paid a growth multiple for peak earnings, because the peak print is the collateral, not the income.

The sequential figure is the tell. Revenue rose 51% from the prior quarter ^[3], a rate that compounds to about 420% annualised if repeated for four quarters ^[19]. Nobody underwrites that, and the multiple says nobody is trying to.

The reported numbers also deserve a note of caution. Operating profit of 60.5 trillion won on 79.3 trillion won of revenue implies a 76% operating margin ^[10][17]. Working backwards from the stated growth rates gives year-ago revenue of about 22.2 trillion won and year-ago operating profit of about 9.2 trillion won, or a margin near 41% [d5a][d5b][d5c]. A move from 41% to 76% in a single year in a memory business would be extraordinary, and this account is a secondary summary rather than the filing ^[16]. Anyone sizing a position off these figures should reconcile them to the primary disclosure first.

What is not in dispute is the capital response. SK Hynix has outlined capex above $31 billion for 2026 and is expanding fab capacity in South Korea against what it treats as durable AI memory demand ^[4]. Capacity commissioned at peak pricing is the standard mechanism by which memory cycles end. Samsung and Micron competition in high-bandwidth memory is intensifying ^[7], the company's share of global HBM revenue is around 58% ^[14], and Nvidia is the dominant buyer with SK Hynix the leading supplier, which concentrates the roadmap risk in one customer ^[6]. US-China trade tension sits on top of that ^[8].

The sell side has not moved to the tape's view. Consensus ratings lean Buy or Strong Buy with average targets implying 40% to 90% upside, and one KRW-based consensus pointed to about 92% ^[5]. Note what that implies: on unchanged earnings, a 92% move takes the stock to roughly 9.6 times forward earnings ^[23], still an unremarkable multiple. The argument between price and target is therefore not about the multiple. It is about whether the earnings denominator is real for more than a few quarters. The bull case is that AI infrastructure spending is not decelerating and HBM demand still outstrips supply ^[9]. The company crossed a $1 trillion market cap earlier in 2026, so this is no longer a small-cap re-rating story ^[15].

Watch three things. Whether HBM4 shipment cadence and pricing firm up in the next quarterly disclosure ^[13]. Whether the $31 billion capex plan is confirmed, raised, or quietly trimmed, which is management's own vote on cycle duration ^[4]. And whether the gap between 5x and the sell-side targets closes by price going up or estimates coming down ^[24][5].