Despite the Federal Reserve raising interest rates last week, US stocks have rallied against the trend; memory stocks continued to surge in pre-market trading today.
$SanDisk Corp.(SNDK)$ climbed above 1,800—hitting a more than two-month high—while $Micron Technology(MU)$ returned to the $1,030 level, with $Western Digital(WDC)$ and $Seagate Technology PLC(STX)$ also rising. Everyone is talking about the "return of the super-cycle."
However, the real signal tonight lies not in stock prices, but in the continued sharp rise of contract memory prices.
1/ In the third quarter, average selling prices (ASPs) rose by 20–30% quarter-over-quarter for DRAM and over 15% for NAND, with single-digit growth expected for the fourth quarter. So far, this is the familiar old story. What truly shifts the narrative is this specific point: "Hyperscale cloud providers have signed new contracts agreeing to pay higher DRAM prices in Q1 2027 than in Q4 2026."
2/ Let’s break this down: past memory cycles followed a classic "boom-bust" pattern—sellers raise prices, buyers scramble for stock, capacity expands, and prices collapse. This time, however, buyers (the likes of $Alphabet(GOOG)$ , AWS, $Meta Platforms, Inc.(META)$ , and $Microsoft(MSFT)$ ) are proactively locking next year's prices into contracts. Effectively, they are using long-term agreements to transform a cyclical commodity into an asset that behaves like an inflation hedge. This isn't just a story about demand; it represents a fundamental shift in buyer structure.
3/ Just how tight is the supply-demand gap? The supply-demand sufficiency ratio is projected at 79% for DRAM and 80% for NAND in 2026, only barely returning to 100% in 2027. Meanwhile, HBM consumes roughly three times the wafer capacity of standard DRAM per bit (according to Citi). The more the "Big 3" manufacturers shift production lines toward HBM, the greater the shortage of standard memory—HBM for AI servers already accounts for over 50% of total DRAM shipments. This supply chain dynamic provides the physical basis for locking prices into contracts.
4/ $Bank of America(BAC)$ has raised its forecast for the 2030 global memory TAM (Total Addressable Market) from $1.8 trillion to $2.0 trillion, projecting a CAGR of 21% for DRAM and NAND combined between 2027 and 2030. Its proprietary "Memory Market Prosperity Index" registered a reading of 180 in July; given a historical median of 100 and a peak of only 120 during the 2017–2018 super-cycle, the current level of market prosperity is 50% higher than the widely acknowledged peak of the previous cycle.
The prevailing market view is "the cycle is back, so buy memory stocks." My take is different: it is not merely a cyclical resurgence, but a scenario where the cyclical nature itself has been fundamentally altered—effectively dismantled by half. The real question isn't whether to buy, but *which* side of the market to bet on:
Highest elasticity: SNDK/WDC (pure-play NAND/HDD). They are the only ones capable of filling the capacity vacuum left as production shifts toward HBM; their stock surged from $93.5 to $1,791 (a 20-fold increase) over the past year. While they offer the greatest short-term upside, their valuations are already stretched, and they lack significant long-term contract structures to lock in future revenue.
Most long-term contracts: MU/ $SK hynix(SKHY)$ . Micron has announced $98 billion in minimum contract revenue across eight customers. The depth of these contracts determines who suffers the least damage during a "soft landing." Bank of America forecasts a roughly 10% price pullback in 2028 from 2027 highs (a soft landing scenario); for Micron, this implies a step-down in profit margins, whereas for SNDK, it could trigger a sharp reversal in share price.
Decision Matrix: If you trade based on short-term sentiment, buy SNDK but set a clear exit point (keep an eye on Micron’s September 30 earnings report). If you subscribe to the logic that "contract-locked pricing equals structural inflation," the answer is MU—trading at a TTM P/E of 22.7x with a consensus target price of $1,547 (52% upside); its valuation reflects a much longer time horizon than SNDK's. However, two risk factors warrant attention:
-
① CXMT announced yesterday the mass production of its fifth-generation DRAM, with per-wafer output increasing by 50% compared to the previous generation; once volume truly ramps up, it will first undermine the price floor for standard DRAM long-term contracts.
-
② Once Samsung and Micron both successfully master HBM4 production (Samsung’s yield reportedly rose from under 60% to 80%), the narrowing HBM premium will force capacity to shift back to standard memory, prematurely easing the "crowding-out effect."
In short: this price hike is not driven by demand, but by buyers proactively locking in contracts. The cyclical nature of the memory market hasn't vanished; it has simply been absorbed by cloud providers leveraging their creditworthiness.
Comments