Morgan Stanley: Memory Contract Prices to Peak in Q4, Storage Cycle Elongating Rather Than Crashing
N.R. Finch
Morgan Stanley warned on July 21 that memory contract prices will peak in Q4 2026, with earnings-upgrade momentum already falling from 92% to 77% — the storage cycle is stretching, not crashing, but valuation pressure has arrived first.
Why does Morgan Stanley say prices are peaking?
The core evidence is not price-curve extrapolation but fading earnings-upgrade momentum — net earnings revision rates have dropped from a 92% peak to 77%, leaving less room for consensus estimates to keep rising. This means → analysts are still bullish on memory profits, but the conviction is weakening.
SK Hynix's forward one-year EPS has recently declined; its price-to-book ratio fell to 2.5×. Samsung's P/B dropped to 1.7×. Both have pulled back sharply from recent highs but remain above long-term averages.
In plain terms = share prices are already pricing in "peak profit growth" — fundamentals haven't broken, but valuations cracked first.
What are inventories and prices signaling at the same time?
Q2 DRAM and NAND inventory levels rose, driven mainly by memory module makers — not a surge in end demand, but mid-stream stocking.
Three signals are firing together: year-on-year price growth slowing + inventories turning up + earnings momentum fading. This means → the market's focus is shifting from "how much more can profits grow?" to "how long can the growth rate last?"
This reflects a classic memory-stock pattern — fundamentals still healthy, but valuations compress first because markets always price the direction of change, not the current level.
AI capex is surging — why can't memory prices just keep climbing?
Morgan Stanley calls AI infrastructure spending the single most positive factor stretching this cycle. The report tracks large-model training and inference intensity, hyperscaler capex trajectories, and flags Q2 2026 hyperscaler capex as the key verification point.
But the report draws a clear logical gap: strong AI demand ≠ sustained memory price increases. Infrastructure monetization does not equal excess compute; better, cheaper models do not automatically mean hyperscalers will raise capex indefinitely.
In plain terms = AI is indeed burning cash on infrastructure, but the pace and direction are set by the hyperscalers — they can hit the brakes at any time, and memory prices turn with them.
Can long-term contracts lock out cycle risk?
Memory makers have signed long-term supply agreements with downstream customers, seemingly smoothing cyclical swings. But Morgan Stanley cites pandemic-era precedent: long-term agreements do not inherently eliminate cycle risk.
When prices and supply-demand shift, contracts can be renegotiated — or force customers into involuntary inventory buildup. This means → contracts lock volume, not price — once the market turns, the agreements themselves can amplify inventory pressure.
Why are HBM and commodity memory heading down two different paths?
Samsung, SK Hynix, and Micron are shifting more capacity toward HBM — high-bandwidth memory designed for AI chips — objectively freeing up room in commodity DDR5.
But HBM supply remains tight, constrained by advanced process, advanced packaging, and bandwidth hard limits. Morgan Stanley estimates the total addressable market — spanning capacity, bandwidth, power, and system architecture — at roughly $25 billion.
This reflects a supply-structure split: HBM undersupplied, DDR5 potentially oversupplied — two product lines in the same industry heading toward opposite supply-demand dynamics.
What is Morgan Stanley's actual conclusion?
The cycle stretches rather than crashes — AI capex is the core force extending the upcycle, but post-peak pricing pressure on earnings growth is already showing up in valuations.
On the equipment side, ASML EUV shipments are projected to rise from 92 units in FY2027 to 104 units in FY2028, matching continued advanced-manufacturing expansion — but that does not mean all memory makers benefit equally.
In plain terms = Morgan Stanley is not bearish on the memory industry, but it is saying "the best is passing" — the next phase hinges on whether HBM and commodity DRAM can each sustain their own supply-demand balance, and that will drive divergence among memory stocks.
Content is for reference only, not financial advice.