ChangXin Memory Technologies
State-backed IDM (integrated device manufacturer) — designs and fabs commodity + specialty DRAM in wholly-owned Hefei fabs; sells wafers/chips/modules to OEMs, cloud/server, and mobile; transitioning from pre-IPO/state-funded to publicly listed via Shanghai STAR Market.
Chronological pre-IPO valuation progression. The ~$19.5B (2024) mark is a closed primary round; the ~$42B (Oct 2025) figure was the pre-earnings-blowout IPO target; the ~$280-420B (mid-2026) range is analyst-floated pending an unpriced STAR Market debut, re-rated up after Q1'26's +719% revenue / ~1,688% profit surge. valuationB values are USD-billions at ~7.15 CNY/USD (2026-06 point uses the ~$350B midpoint of the $280-420B range).
Earnings, margins, COGS & capex
CXMT flipped from years of heavy losses (CNY -16.3B in 2023, ~CNY -7.9B in 2024) to its first full-year profit in 2025 (CNY ~1.9B / ~$275M on CNY 61.8B revenue), then posted a blowout Q1 2026 — CNY 50.8B revenue (+719%) and ~CNY 25B net profit attributable (~$3.6B; ~CNY 33B / ~$4.9B total) as the AI-driven DRAM shortage sent ASPs vertical (DRAM contract prices +~95% QoQ per UBS). H1 2026 is guided to CNY 110-120B revenue and ~CNY 50-57B attributable net profit. The trajectory is real and registration-disclosed but sits at a cyclical peak; the 2023-2024 losses and ~CNY 36.6B accumulated deficit are the honest counterweight.
Income statement — where each revenue dollar goes
% of revenueOf every $1 of revenue, ~59¢ is cost of goods and ~0¢ operating expense, leaving ~41¢ of operating profit (~25¢ net).
Revenue trend
Margins
sharply up: -2.19% (2023) to 5.0% (2024) to 41.0% (2025), higher in Q1 2026
peak-cycle; comparable to incumbents' supercycle memory-segment margins
from a CNY ~-1.6B loss in Q1 2025; FY2025 was first-ever annual profit (~CNY 1.9B)
COGS structure
Dominated by wafer-fab depreciation (huge capacity buildout), equipment, materials, and yield/scrap; COGS/ASP dynamics drove the gross-margin swing from ~5% to ~41% as ASPs spiked and utilization hit ~95.7%. Not itemized in public sources beyond the margin bridge.
Capex
Heavy, ongoing fab investment in Hefei; capacity expanded ~100k to ~290k wpm (2024-2025) toward ~300-350k (end-2026) and 400k+ after, funded by state capital and now IPO proceeds (CNY 29.5B earmarked for DRAM mass-production and technology upgrade). Absolute capex figures not disclosed.
Latest earnings
n/a (pre-listing; no analyst consensus) — headline was a ~1,688% profit surge vs a year-ago loss
H1 2026 revenue CNY 110-120B (~$15.4-16.8B); H1 attributable net profit ~CNY 50-57B (~$7.4-8.4B). SemiAnalysis models FY2026 revenue potentially >$50B, ASP-led rather than share-led.
- Q1'26 revenue
- CNY 50.8B (~$7.1B), +719% YoY
- Q1'26 net profit
- CNY ~25B attributable (~$3.6B); ~CNY 33B total (~$4.9B)
- Capacity utilization
- ~95.7%
- Global DRAM share
- 7.67% by revenue (Q4'25); ~11% by wafer capacity
- Monthly capacity
- ~290k wpm (end-2025), targeting ~300-350k (end-2026)
Growth drivers
- AI-driven global memory shortage lifting DRAM/HBM ASPs to multi-year highs (DRAM contract prices +~95% QoQ in Q1'26 per UBS) — primary driver of the 2026 revenue/margin explosion
- Capacity expansion from ~100k to ~290k wpm with a path to 400k+ wpm using IPO proceeds
- Migration up the stack: DDR4/LPDDR4 to DDR5 (8000 MT/s), LPDDR5, and HBM3 mass production targeted late 2026
- China domestic-substitution tailwind — OEMs (Lenovo, Xiaomi, OPPO, vivo) and clouds (Tencent, Alibaba, ByteDance) increasing localized memory to de-risk from US controls
- Anchor supply deal — multi-year (up to 5-year) server-DRAM agreement with Tencent worth >CNY 20B (~$2.94B)
Bull & bear
A national-champion memory maker inflecting to profitability at the exact moment an AI-driven shortage makes DRAM the scarcest commodity in tech — with a captive domestic market, state funding, and an IPO to bankroll a run at the global top three.
- Registration earnings are extraordinary and real: Q1'26 revenue +719% to CNY 50.8B, ~70% operating margin, ~95.7% utilization — filed results, not a projection
- Structural domestic-substitution demand: Chinese clouds and OEMs must localize memory under US controls, giving CXMT volume even if it competes on price
- Only scaled Chinese DRAM option — scarcity/strategic premium supports the CNY 2-3T valuation talk regardless of Western comps
- Clear capacity and roadmap path (~290k to 400k+ wpm; DDR5 at 8000 MT/s; HBM3 late-2026) into the highest-value AI-memory tiers
- Deep-pocketed backers and IPO proceeds mean it can outspend the cycle where private peers can't
A capital-intensive, technologically-trailing memory maker being IPO'd at a CNY 2-3T valuation on peak-of-cycle earnings, structurally exposed to US export controls and to the incumbents' ability to end the party at will.
- Peak-cycle framing: memory is violently cyclical and CXMT lost CNY 16.3B (2023) and ~7.9B (2024); ~70% margins and the 1,688% surge will not persist through a downturn
- Valuation is rich — CNY 2-3T (~$280-420B) embeds a scarcity/national-champion premium on peak-cycle earnings for a technologically-trailing maker; on mid-cycle normalized earnings it is expensive versus Micron/SK Hynix multiples despite being far smaller than Micron's ~$1.1T cap, and it is not fundamentals a US investor can underwrite
- Technology gap is real: one to two nodes behind on DRAM and years behind on HBM (HBM only ~5k wpm today, scaling toward ~60k wpm vs SK Hynix's far larger base) — the highest-margin AI-memory tier is where it's weakest
- Existential tooling dependency on ASML/AMAT/Lam/KLA/TEL under active US-led export controls; a tightening could cap the entire capacity plan
- Incumbents can defend share with capacity and price, and US-listing risk overhangs customers/supply — CXMT is on the Pentagon 1260H list and faces potential Commerce Entity-List designation
- Not investable by most US persons anyway — A-share float, state control, opaque governance
What it is worth
Cross-check of the analyst-floated CNY 2-3 trillion post-IPO valuation against Western DRAM comps (Micron ~$1.1T market cap, SK Hynix) and against peak-cycle vs mid-cycle earnings. Note: this is an A-share, state-controlled, thin-float name — the valuation embeds a national-champion/scarcity premium and cannot be underwritten on incumbent multiples; it is NOT a US-investable security and is discussed as context only, not a buy/own call.
<CNY 1.5T (~$200B)
memory downcycle compresses ~70% margins toward mid-cycle, export-control tightening caps the capacity/HBM roadmap, and the peak-earnings multiple re-rates hard; debt (~CNY 118.8B) and accumulated losses come back into focus.
~CNY 2-2.5T (~$280-350B)
lists at record raise, near-term earnings strong but the market applies a cyclicality discount as the cycle matures; share gains continue on capacity, HBM remains sub-scale.
~CNY 3T+ (~$420B+)
supercycle extends, HBM3 ramp succeeds, domestic-substitution locks in structural share toward the low-to-mid teens % of capacity by 2027; peak earnings sustain and A-share scarcity premium holds.
IPO raises ~CNY 29.5B ($4.3B) — the largest China A-share IPO of 2026 — implying an analyst-floated ~CNY 2-3T (~$280-420B) market cap, a fraction of Micron's ~$1.1T. The figure re-rated sharply up from a ~CNY 300B (~$42B) IPO target reported Oct 2025 (pre-Q1'26 earnings blowout). On peak-cycle Q1'26 annualized earnings (~$14-16B net) the headline multiple looks modest, but that is the trap: memory is cyclical and 2023-24 were loss years. On mid-cycle normalized earnings the same price is expensive versus Micron/SK Hynix given CXMT's technology and HBM gap. The premium is policy/scarcity-driven, not free-cash-flow-driven.
SWOT
Strengths
- Only large-scale DRAM maker in China with a real, ramping product line (DDR4/DDR5/LPDDR5, HBM3 next) — structural scarcity value at home
- World's #4 DRAM supplier (7.67% revenue share, ~11% capacity) with peak-cycle operating margins (~70%) near the incumbents
- Deep state and strategic backing (Big Fund II — Anhui/Beijing/Guangdong SOEs; Alibaba Cloud, GigaDevice, CMB/ABC/CCB) and a captive domestic-substitution demand base
- IPO fills the war chest — ~CNY 29.5B ($4.3B) to fund the push toward 400k+ wpm
Weaknesses
- Earnings sit at a cyclical peak — memory is brutally cyclical, and 2023-2024 were CNY -16.3B and ~-7.9B loss years; ~CNY 36.6B accumulated deficit and ~CNY 118.8B debt
- One to two nodes behind Samsung/SK Hynix/Micron on process and years behind on HBM (HBM capacity only ~5k wpm today, targeting up to ~60k wpm / ~20% of output through 2026 — still a fraction of SK Hynix's scale)
- Heavily dependent on foreign wafer-fab equipment (ASML/AMAT/Lam/KLA/TEL) precisely where US-led export controls bite
- Opaque, state-directed governance and A-share-only listing — profitability is very young (first annual profit only 2025)
Opportunities
- AI compute demand keeps DRAM/HBM tight, extending the ASP supercycle
- Land HBM share for domestic AI accelerators (Huawei Ascend et al.) locked out of foreign HBM by export controls
- China domestic-substitution mandate could hand CXMT structural volume regardless of price
- US Entity-List forbearance (Apple is reportedly lobbying the White House to keep CXMT off the Commerce Entity List) could open consumer-product adoption (e.g., handsets)
Threats
- US BIS Entity-List restrictions and NDAA Section 5949 procurement bans cap access to leading-edge tooling and US-government-linked customers; CXMT was re-added to the Pentagon Section 1260H list on 2026-06-08 after a brief February removal
- Incumbents (Samsung, SK Hynix, Micron) can flood capacity and crush prices to defend share once the cycle turns
- Memory downcycle would compress the ~70% margins violently, exposing the debt load
- Escalating US-China tech decoupling; potential fresh export-control tightening on DRAM/HBM tooling
Moats, dependencies & bottlenecks
Moats
Strong (domestically) High while US-China decoupling persists Only scaled Chinese DRAM maker; state capital, procurement preference, and substitution mandate create a captive home market.
~290k to 400k+ wpm buildout funded by state + IPO; capital intensity is a barrier but every incumbent has more.
Weak vs incumbents One to two nodes behind Samsung/SK Hynix/Micron; HBM only starting — this is the missing leg of a durable moat.
Designed into Lenovo/Xiaomi/OPPO/vivo and Tencent/Alibaba/ByteDance; qualification stickiness, reinforced by localization pressure and the Tencent multi-year server-DRAM deal.
Dependencies
Applied Materials, Lam Research, KLA, Tokyo Electron) Supply / capital equipment Leading-edge DRAM tooling is export-controlled by the US and allies; the entire capacity/technology roadmap hinges on continued access or domestic-tool maturity (Naura, AMEC, SMEE).
NDAA Sec. 5949, Pentagon 1260H) Added to the BIS Entity List (2024) with tooling restrictions; briefly removed from the Pentagon 1260H list in Feb 2026 then re-added on 2026-06-08. Apple is reportedly lobbying to keep CXMT off the Commerce Entity List. The policy environment is volatile.
Buildout depends on Big Fund II / SOE backing and the ~CNY 29.5B raise; a funding or policy shift would stall expansion.
Current margins and valuation are a function of the supercycle; a downcycle removes the earnings that justify the multiple.
Revenue concentration Substitution-driven demand is a strength but concentrates revenue in a few Chinese buyers (Tencent, Alibaba, ByteDance, Lenovo, Xiaomi).
Advantages
- Sole scaled Chinese DRAM supplier — structural scarcity and policy support at home
- Deep state + strategic-investor funding and now public-market access
- Captive domestic-substitution demand insulated from Western competition
- Low-cost capacity able to undercut on commodity DRAM to win share
- Peak-cycle cash generation to reinvest through the buildout
Weaknesses
- Technology gap of one to two nodes on DRAM and multi-year gap on HBM
- Earnings and valuation anchored to a cyclical peak
- Existential exposure to US/allied equipment export controls
- Very short profitability history; large accumulated losses and debt
- Opaque governance; A-share-only, state-controlled, minimal free float
Bottlenecks
- Leading-edge lithography and advanced-node tooling access under export controls
- HBM3/HBM3E capability and capacity — years and scale behind SK Hynix/Samsung/Micron
- Advanced-packaging/TSV capacity needed for competitive HBM
- Yield maturity at newer DDR5/LPDDR5 nodes vs incumbents
- Talent and IP for sub-16nm-class DRAM scaling
Top signals & trends
Top signals
Bullish (execution/access-to-capital) · Largest China A-share IPO of 2026; validates state commitment to scaling CXMT. Debut expected ~late-Aug to Sept 2026.
Bullish near-term / caution on durability · Real supercycle economics, but peak-cycle by nature.
Bullish (demand validation) · Anchors server/cloud DRAM volume domestically; reported late June 2026.
Bearish (adoption/regulatory friction persists) · Reverses the earlier easing; Apple reportedly lobbying to keep CXMT off the Commerce Entity List. Entity-List tooling curbs remain.
Bullish (share gain) · Among the fastest-growing DRAM suppliers by capacity into 2027 (projected low-to-mid teens % of capacity).
Bearish (supply-side) · Direct threat to the 400k+ wpm and HBM roadmap.
Trends
Strongly positive (near-term) · Primary driver of 2026 revenue/margin; the core reason CXMT is profitable and IPO-able now. UBS expects the upcycle to last into at least late 2027.
Positive (structural) · Policy-mandated localization of memory hands CXMT durable domestic demand.
Negative (supply) / positive (demand) · Cuts tooling access but also fences off the domestic market for CXMT.
Huge TAM but CXMT's weakest area; late-2026 HBM3 ramp is the key catch-up bet.
Positive for CXMT / negative for incumbents' pricing power · CXMT's low-cost capacity threatens the Samsung/SK Hynix/Micron pricing structure over time.
Ecosystem & competitor graph
Suppliers feed the company; customers pull from it. Line thickness shows the strength of each tie (supply-chain dependency, customer earnings contribution). Hover to isolate a tie.
Lithography; DUV access constrained and EUV blocked by export controls — the hardest bottleneck.
Deposition/etch/process tools; export-controlled to CXMT.
Etch/deposition, critical for DRAM scaling; controlled.
Process control/metrology; controlled.
Coater/developer/etch; subject to allied controls.
Domestic tool substitutes CXMT increasingly relies on as foreign access tightens.
>CNY 20B (~$2.94B) multi-year server-DRAM supply agreement; anchor cloud/server buyer.
Cloud customer and strategic investor.
Cloud/AI infrastructure buyer localizing memory supply.
PC/server OEM in CXMT's supply chain.
Handset OEMs sourcing LPDDR/DDR for domestic substitution.
#2 DRAM, HBM leader (dominant HBM3E share to Nvidia); the technology benchmark CXMT is years behind on HBM.
#1 memory maker across DRAM/NAND/HBM; deepest R&D and capacity; can defend share on price.
#3 DRAM, only US-based maker; the primary US-investable memory proxy and closest DRAM/HBM business comparator (though ~2-4x larger, at a ~$1.1T market cap).
Taiwan DRAM maker; smaller, commodity/specialty focus — a tier CXMT already outscales.
Taiwan specialty/niche DRAM and flash; overlaps at the low end.
Chinese memory/MCU firm and CXMT strategic investor/partner — niche DRAM/NOR, more ally than rival.