Spot Memory Prices Are Running 4 Times Contract Prices. That Is Not What a Cycle Peak Looks Like.

HBM3E pricing signals continued tight supply: a 36GB HBM3E chip reportedly sells for about $2,100 on the spot market versus ~500,000–700,000 won (~$370–$510) under long-term contracts—a 4–5x premium. Korea DRAM exports show volume down ~13% from May to July while shipment value rose ~19% (unit price +37%). SK Hynix’s Q2 results reflect the upcycle with revenue +257% YoY to 79.3T won and operating profit +61% QoQ to a record 60.5T won (76% operating margin), though HBM4 ramp risk from Samsung and expected price/margin volatility remain key overhangs.

Analysis

This is still a pricing-power story, not a supply-normalization story. The spot/contract gap says end buyers are paying up for immediate allocation, which usually only happens when capacity is already committed and the near-term incremental bit of supply is effectively zero. That supports another quarter or two of strong gross margin for the tightest HBM vendors, but it also means the market may be underestimating how long the shortage can persist once customer qualification and packaging bottlenecks become the real constraint rather than wafer output.

The cleaner second-order winner is not the whole memory complex but the names with the strongest mix into AI-grade HBM and the best allocation discipline. SKHY should keep taking share of industry profit pools, but Samsung’s rising HBM share matters because it can cap SKHY’s multiple expansion even if the memory upcycle stays intact. NVDA is more nuanced: tighter HBM supply can support pricing, but if memory remains scarce it can also slow GPU unit ramp and push out revenue recognition, which is a delivery bottleneck rather than a demand problem.

Contrarian risk: the market may be too early calling a peak because the usual rollover signals are absent; the first real warning would be spot prices collapsing toward contract, not just moderating from extreme levels. Over 1-3 months, the main reversal catalyst is an announced capex wave or evidence that AI demand is digesting inventory faster than new HBM capacity can be qualified. Over 6-18 months, semicap names are the eventual beneficiaries if the industry responds rationally; if that capex starts to show up, the shortage trade should be reduced quickly.

Key Decisions for Investors

• Long SKHY on 5-8% pullbacks; use a 1-3 month horizon and trail the stop if HBM spot premium compresses below ~2x contract or Q/Q operating profit growth decelerates sharply.
• Pair trade: long SKHY / short SSNLF for a relative-value expression on HBM share and pricing discipline; thesis breaks if Samsung keeps taking HBM share and SKHY’s revenue mix stops improving over the next 1-2 quarters.
• Do not chase NVDA purely on this memory print; treat HBM scarcity as a potential shipment bottleneck. Use any NVDA strength to fade near-term if channel checks show Blackwell lead times extending rather than improving.
• Set a watchlist for semicap beneficiaries (AMAT, LRCX, KLAC) rather than buying immediately; add only if memory vendors start guiding up capex, which would be the first credible sign the cycle is moving toward normalization.