Survey of 50 Asian Institutions: Weak Sentiment on Memory Chip Sector, Cloud Capex Becomes Core Variable
Miles Bennett
Over 50 Asian institutional investors voiced a common concern during a Hong Kong roadshow: memory-sector sentiment is weak heading into June earnings, with the widening gap between sell-side memory-market forecasts and actual cloud-provider capex guidance as the core drag on valuations.
Why has cloud capex become the make-or-break variable for memory?
Sell-side analysts have raised their memory total-addressable-market estimates far faster than cloud providers have raised actual capex guidance — expectations are running ahead of the money.
Some investors expect the market to push the corresponding cycle's capex expectations into the $1 trillion–$1.5 trillion range over the next 3–6 months.
This means → if cloud providers ultimately deliver only modest increases, the market will read it as a direct negative for memory. The research team estimates this single variable accounts for roughly 70% of current memory-sector sentiment.
Pricing momentum is fading — are Korean makers' profit forecasts still being cut?
After Q2 2026, both year-on-year pricing momentum and quarter-on-quarter pricing elasticity for memory weakened simultaneously.
Korean memory makers' Q2 operating-profit forecasts have been revised down continuously, deepening the wait-and-see mood.
In plain terms = the pricing engine is losing steam, profit expectations are shrinking with it, and investors are reluctant to commit before June earnings.
Long-term agreements — are investors more optimistic, or just asking different questions?
Compared to 3–6 months ago, the focus has shifted from "can LTAs work at all?" to "how can memory makers use LTAs to lock in long-term relationships with key customers?"
Yet more than half of surveyed institutions remain cautious: the near-term impact of LTAs on spot prices and margins is still unclear, and the market fears a ceiling on short-term profits.
This means → the research team pushes back on cycle pessimism from a full-cycle perspective — the core value of LTAs is not boosting current margins but smoothing out the 50%–70% price crashes that typically hit during down-cycles.
How much can HBM average prices actually rise — why do buy-side and sell-side disagree so sharply?
Over half of buy-side firms expect HBM average selling prices to double year-on-year next year; their logic is that only a per-GB price doubling would lift HBM margins meaningfully above standard DDR5.
The research team sees a reasonable range of 25%–30% YoY, arguing it must be calculated against HBM's cannibalization ratio — how much standard DRAM capacity HBM production consumes.
In plain terms = buyers want "double or it's not worth it"; the research team says "a 30% rise is already the fair ceiling" — and the gap comes down to different assumptions about how much DRAM capacity HBM crowds out.
DRAM vs. NAND — which is tighter, and will Chinese makers disrupt?
Both are tight, but DRAM more so: DRAM self-sufficiency sits at only 50%–60%, versus 70%–80% for NAND.
Investors show no significant concern about Chinese memory competition for now. The logic: if Chinese DRAM makers tilt new capacity toward HBM paired with domestic AI inference chips, that actually relieves supply pressure on commodity DRAM, and global shortages limit the impact on overseas leaders.
Supply-chain sources indicate that China's leading NAND maker will build both DRAM and NAND lines in its new fab next year, splitting cleanroom capacity evenly — a move expected to keep improving NAND supply-demand through 2028.
South Korea's $3 trillion AI investment plan — bullish or bearish?
The market's initial read is negative: fears that the plan will weaken supply discipline, accelerate expansion, and dilute memory makers' pricing power.
The research team counters that capex execution carries many unknowns, and supply expansion is a prerequisite for absorbing AI's massive capital spending — without it, memory cannot maintain a reasonable share of industry value.
This reflects the market's current core framework for memory: the actual magnitude of cloud-provider capex increases will be the key validation point for whether sentiment can turn around.
Content is for reference only, not financial advice.