SK Hynix Supplies More Than Half the World's HBM Memory. Its Stock Costs Under 4 Times Next Year's Earnings.

SK Hynix reported Q2 revenue of 79.3 trillion won (~$55B), up 257% YoY, with operating profit up 557% to 60.5 trillion won (~$42B) and an operating margin expanding to 76%. The company held a 58% share of the HBM high-bandwidth memory market (Counterpoint) and began mass shipments of HBM4 in Q2, supported by long-term agreements with ~10 customers and a Nvidia/SK Group partnership announced at >$500B. Despite this momentum, the stock trades at a forward P/E below 4—implying earnings peak risk and potential sharp downside if profitability declines as supply ramps.

Analysis

The market is still valuing this like a late-cycle memory spike, but the more important change is that HBM is behaving less like a spot commodity and more like reserved capacity inside the AI supply chain. That shifts bargaining power toward the producer with the best yield and qualification pipeline, which should keep margins elevated longer than a typical DRAM upcycle. The second-order winner is the AI infrastructure stack that can secure supply early; the loser is anyone forced to buy HBM at market prices or delay system builds.

The near-term risk is not demand collapse; it is supply normalization. If Samsung closes the yield gap or if SK Hynix’s next-generation ramps stumble, the market will quickly re-rate the earnings base because the current multiple is only cheap if the profit stream extends beyond next year. Over 1-3 months, the key catalyst is whether customer agreements translate into visible volume commitments and whether pricing data continues to firm; over 6-18 months, the debate is whether HBM stays structurally scarce or becomes another margin-dilutive component once capacity catches up.

The contrarian take is that consensus is underappreciating how much AI memory content per accelerator still has room to rise, which can keep unit demand strong even if GPU growth slows. But the flip side is that low multiples in this part of semis are often the market’s warning that peak earnings are being capitalized too aggressively. If forward estimates start drifting down before the next annual budgeting cycle, the stock can de-rate faster than the fundamentals deteriorate, so the risk is not a bad quarter but a sequence of modest estimate cuts.

Key Decisions for Investors

• Buy SKHY on weakness rather than chasing the post-rally move; best entry is a 5-10% pullback. Hold 3-6 months for estimate revisions to catch up. Thesis breaks if forward EPS consensus for 2027 starts falling or Samsung materially narrows the HBM share gap.
• Pair trade: long SKHY / short SSNLF for a 3-6 month relative-value expression of HBM share leadership. This works if HBM remains supply-constrained and qualification matters more than raw capacity. Cover the short if SSNLF shows meaningful HBM yield or customer-win progress.
• Use NVDA as a secondary beneficiary, not the primary one. A small long can work if supply assurances reduce build risk across the AI stack, but tighten risk if NVDA commentary indicates memory costs are pressuring unit economics or data-center shipments.
• Set a watch item on HBM pricing and capex announcements from both SKHY and SSNLF over the next earnings cycle; if spot pricing flattens while capacity rises, the trade should be de-risked immediately.
• If options/liquidity are available, consider a limited-risk SKHY call spread 3-6 months out only on pullbacks; the setup is asymmetric, but a straight equity chase has poor reward if estimates start rolling over.