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

HBM3E (36GB) spot pricing is reported at about $2,100—roughly 4–5x the $370–$510 long-term contract range—signaling tight near-term supply for AI memory. SK Hynix’s Q2 operating profit jumped 61% QoQ to a record 60.5 trillion won and operating margin reached 76%, while Q2 revenue rose 257% YoY to 79.3 trillion won. Korea DRAM export volume fell ~13% from May to July, but shipment value rose ~19% (+~37% unit price), and TrendForce expects DRAM contract prices to rise another 13–18% in Q3, implying the upcycle is not yet peaking.

Analysis

This is still a scarcity tape, not an easing tape. The key market mechanism is that pricing is being set by the marginal buyer at immediate delivery, which usually means downstream users are still under-inventoried and will continue to overpay until lead times normalize. That keeps near-term revisions biased up for the dominant premium-memory supplier, while also implying that any buyer with fixed-price product exposure in the AI server chain is absorbing margin pressure or delaying shipment timing.

The bigger second-order risk is supply response lagging the headline profitability. Once a few quarters of excess gross margin persist, Samsung and other memory players have every incentive to fund capacity, and the equity market typically discounts that capex wave long before it dents reported supply. So the trade is strongest over the next 1-3 months on earnings upgrades, but the structural risk window is 6-18 months when new HBM capacity and HBM4 qualification start to compress both pricing and valuation multiples.

Consensus may be underestimating how sticky qualification is in HBM, which means the market can stay tight longer than the usual DRAM cycle. The contrarian miss is that this is becoming an oligopoly, not a monopoly: share is shifting, and that makes the next leg less about one name’s moat and more about who can convert scarce supply into capacity and customer wins fastest. That argues for owning the basket selectively rather than assuming the current winner remains the cleanest way to express the theme indefinitely.

Key Decisions for Investors

• Long SKHY on 3-5% pullbacks; look to add ahead of the next quarterly print. Target 10-15% upside over 1-3 months if contract pricing and margin commentary stay firm; cut if the spot/contract premium collapses materially or management guides to flatter margins.
• Initiate a relative-value pair: long SSNLF / short SKHY over 6-12 months to express the second-source catch-up trade. The risk/reward is attractive if Samsung’s HBM share gains keep compounding; thesis breaks if SKHY holds share above ~55% and Samsung’s ramp slips.
• No aggressive short in NVDA yet; treat it as a watch item rather than a trade. If NVDA commentary starts citing HBM availability as a shipment or gross-margin constraint, that would be the signal to rotate out of SKHY strength and reassess the broader AI hardware basket.
• Use any sharp SKHY rally to buy downside protection rather than outright sell the core position. Cyclical peaks usually show up first in lead times and spot spreads, so a 3-6 month put hedge is the cleaner way to stay long the trend while capping the inevitable capex-driven reversal.