Back to essays
Storage · Cycle

Is the memory cycle peaking? Six core signals

2026-07-24 · 5 min read · audio 1:54
Memory cycle
Audio version · MiniMax TTS · male-qn-jingying

Prices themselves tell the story

Look at a few numbers (Q2 2026 contract prices, TrendForce):

Note the detail: HBM rises faster than DRAM, and DRAM rises faster than NAND. This is not an average move. It is a transmission of price pressure from the top of the supply chain downstream. HBM serves AI compute cards, where demand is most inelastic; DRAM serves servers, less inelastic; NAND serves SSDs, the most elastic. The further upstream, the tighter the supply; the further downstream, the earlier the rise.

Conclusion: this round is structural scarcity, not broad-based tightness.

Why a "super cycle" rather than an ordinary one

FeatureOrdinary cycleThis cycle
Core driverPC / mobile / serversAI compute
Demand elasticityHigh (price rises kill demand)Low (AI must buy)
Supply response12 to 18 months to expand24 to 36 months (technical bottlenecks)
Duration3 to 4 years5+ years (industry consensus)
Consumer-side price rises will naturally cool the market. Not a crash — a gentle easing.

Six core signals

Six dimensions to judge where we are in the cycle:

Signal 1: Price moves — still rising

Signal 2: Supply gap — still widening

Signal 3: Big-three expansion — deliberately not

Signal 4: Domestic-firm earnings — exploding

Signal 5: Domestic substitution — accelerating

Signal 6: Stock prices — partly priced in

Six signals: five green lights, one yellow.

Final verdict

We are in the middle-to-late stage of this super cycle. Not the top, but no longer the bottom either.

One sentence for those still boarding

This cycle is not "rally then crash". It is "rally then plateau, then gentle cooling." Investors do not need to call the top — follow earnings, follow capacity, follow AI compute growth.