UBS Maintains $1,625 Target for Micron: DRAM/HBM Upgrades Offset NAND Downgrades
Nashnova编辑部
UBS holds its $1,625 Micron target but rewires the math beneath it: DRAM and HBM upgrades push peak earnings into 2028–2029, while NAND cuts only flatten the near-term slope — profit gravity shifts later and higher.
The price target didn't move — so what actually changed?
The 2029 EPS forecast rose from $121 to $165, but the valuation multiple dropped from 15× to 11× — the two offset, landing the target back at $1,625.
This means → UBS is more bullish on Micron's earnings power but expects the market to pay less per dollar of profit by then. Confidence and caution cancel out exactly.
In plain terms = same price tag, completely different recipe inside — more profit, but each dollar of it is valued less.
What's wrong with NAND — is it peaking?
UBS cut its Q3 industry NAND contract-price growth forecast from ~28% to ~20%, and Micron's from ~35% to ~23%; FY2026–2028 NAND revenue estimates were trimmed by 4%, 17%, and 9%.
Yet demand was not cut — 2026 and 2027 industry NAND bit-demand growth was actually raised to 23% and 26%, with server and storage SSDs offsetting weakness in phones and PCs.
This means → NAND's issue is a less steep price ramp, not deteriorating demand. On the supply side, of YMTC's planned ~45K additional monthly wafer capacity, ~30K are shifting to DRAM — which actually supports NAND prices near term.
Why is DRAM the more important variable?
UBS expects Q3 DDR contract prices up ~20% quarter on quarter, Q4 up another ~11%; the supply deficit is projected to last at least through Q2 2028.
DDR long-term agreements split pricing into fixed and floating components: Micron gives up some upside when the market surges but gains downside protection when prices retreat. In plain terms = the LTA trades a slice of peak profit for steadier revenue in a downturn — not permanently higher prices, but lower volatility.
DRAM upgrades outweigh NAND cuts for Micron because DRAM's revenue base is larger, operating leverage is stronger, and HBM consumes large volumes of DRAM wafers — the stronger HBM demand gets, the tighter effective supply for conventional DRAM, reinforcing a positive price loop.
Nvidia cut specs — why did HBM demand go up?
Nvidia's Vera Rubin 300, constrained by HBM supply, will launch with 384 GB HBM4 instead of the previously expected 768 GB, upgrading to 512 GB HBM4E around Q3 2027.
This means → lower capacity per system, but the same HBM supply supports more system shipments — total demand actually rises.
UBS raised its 2027 industry HBM demand forecast from ~58.7 billion Gb to ~61.5 billion Gb; 2025-to-2027 demand runs roughly 17.5 billion, 32.2 billion, and 61.5 billion Gb — a ~2.5× increase in two years. HBM4E pricing may exceed $30/GB, and UBS lifted the 2027 industry HBM ASP year-on-year increase from 67% to 79%.
HBM sells at a premium — does that mean higher margins?
Not necessarily. A 2027 model snapshot: HBM ASP is ~$29.61/GB versus ~$18.39/GB for conventional DRAM — roughly 1.6×. But HBM unit cost is ~$6.51/GB versus ~$0.96/GB — a gap of nearly 6.8×.
Through-silicon vias — vertical channels drilled through a chip to pass signals between stacked layers — plus stacking, testing, advanced packaging, and yield losses all push costs up. HBM gross margins sit roughly at 70%–near 80%, while conventional DRAM in extreme shortage can reach even higher.
In plain terms = HBM is a high-end custom product — premium-priced but complex to make with higher scrap rates. Conventional DRAM is the commodity; when it hits a supply crunch, its margin can actually be more extreme.
How wide is the 2028 disagreement?
On a calendar-year basis: UBS's 2026 EPS forecast of $106.91 beats consensus by ~3.6%; 2027's $210.07 beats by ~28.2%; 2028's $262.22 beats by ~59.1%.
This reflects a widening gap in assumptions — the further out you look, the more UBS and the Street diverge on how long the supply deficit lasts, how fast prices retreat, and how large HBM4E's contribution grows. By 2028, the two sides are operating under different cycle frameworks entirely.
Whether the model delivers ultimately hinges on HBM4 and HBM4E yields — the share of good chips out of total production. Whichever curve moves faster — price or cost — determines whether revenue and profit arrive in sync.
Content is for reference only, not financial advice.