UBS Initiates Coverage on CXMT with Buy Rating, Target Price RMB 70
0xBroomberg
UBS initiated coverage of CXMT (长鑫科技) with a Buy rating and a RMB 70 target, implying 33% upside; the bank sees CXMT at the intersection of tight DRAM supply, AI-driven demand, and China's push to localize semiconductors.
Where does the RMB 70 target come from?
UBS uses a P/B-ROE model — pricing off return on equity to derive a fair price-to-book — and arrives at 8.6x 2027E P/B.
This means → UBS values CXMT at a roughly 40% premium to A-share wafer-fab peers, which trade at an average 6.1x 2027E P/B.
The key assumption: average ROE of 37% across 2028–2030, with an 8.5% cost of equity. If ROE falls short, the premium unravels.
Why are UBS's profit forecasts so far above consensus?
UBS projects CXMT net income of RMB 139.7bn, 332.8bn, and 428.2bn for 2026–2028 — a 75% CAGR, roughly 19% above Wind consensus on average.
The main driver is depreciation dilution: depreciation as a share of revenue drops from 40–119% in 2022–2025 to just 6–10% in 2026–2028.
In plain terms = CXMT spent heavily building fab lines and carried enormous depreciation; once capacity ramps and revenue scales, fixed costs get spread thin — UBS forecasts gross margins rising to 82–89%.
How is the product mix shifting — and who is buying?
UBS expects server DRAM to grow from 26% of revenue in 2025 to roughly 51% by 2028, while mobile DRAM drops from 59% to 38%.
This means → CXMT is pivoting from a mobile-memory supplier to a server-memory supplier — and server DRAM commands higher prices, so the mix shift alone lifts margins.
Localization orders are already in hand: ByteDance reportedly signed a $7bn+ DRAM procurement deal, Tencent signed a RMB 20bn+ long-term server-DRAM supply agreement, and Alibaba and Lenovo are also customers.
How long can the DRAM pricing upcycle last?
UBS sees the DRAM supply shortage lasting at least through Q2 2028, driven by sustained AI infrastructure spending and constrained capacity additions by major producers.
DDR contract prices are projected to rise from $0.40/Gb in 2025 to $2.24/Gb in 2027 — a two-year CAGR of 137%. HBM — high-bandwidth memory built for AI chips — is expected to climb from $1.52/Gb to $3.13/Gb.
This reflects a structural dynamic: Samsung, SK Hynix, and Micron are shifting capacity toward HBM → conventional DDR supply tightens further → CXMT, still mainly a DDR/LPDDR producer, benefits directly from commodity DRAM price increases.
Can capacity expansion support meaningful share gains?
CXMT's monthly DRAM wafer capacity is forecast to rise from 240k wafers at end-2025 to 466k by end-2028, driven by expansions in Hefei and Beijing plus new fabs in Shanghai and Hefei.
Global DRAM bit-supply share is expected to climb from roughly 7% in 2025 to about 10% by 2028; China server-DRAM market share from 12% to about 20%.
On process technology, CXMT has advanced from G1/G3 to G4 and is moving toward G5 — each node migration means more bits per wafer, so bit shipment growth outpaces capacity growth.
What is the biggest risk to this thesis?
UBS frames CXMT's investment case as a "Davis Double Play" — earnings growth and multiple expansion occurring simultaneously.
But the bank itself flags that in 2029–2030, as DRAM prices normalize, EBIT margins are likely to retreat to 68–74%.
Put simply = today's elevated margins rest on a pricing upcycle; once the cycle turns, both earnings and valuation could compress — how long margins hold and when the cycle inflects are the key tests for this thesis.
Content is for reference only, not financial advice.