CXMT Erases a Decade of Losses as Q2 Revenue Beats Estimates by 54%
Goldman Sachs raises 12-month target to RMB 129 — implying 120% upside — after the domestic memory maker posted Q2 revenue of RMB 99.5 billion (US$13.8 billion), beating consensus by 54% and erasing a decade of accumulated losses in a single half-year.
Changxin Memory Technologies, China's sole large-scale DRAM manufacturer, delivered second-quarter 2026 results that blindsided even the most bullish analysts on the Street. Revenue surged 977% year-on-year and 96% quarter-on-quarter to RMB 99.5 billion (US$13.8 billion) — exceeding Goldman Sachs's own model by 52% and Bloomberg consensus by 54%. For the first half of 2026, CXMT reported net profit of RMB 77.61 billion (US$10.78 billion), ranking it tenth among all A-share listed companies by earnings — a figure that, in isolation, more than wipes out the RMB 36.65 billion (US$5.09 billion) in cumulative losses the company carried into 2026.
Goldman Sachs responded by reiterating its Buy rating and holding its 12-month price target at RMB 129, representing approximately 120% upside from current levels. The bank lifted its full-year 2026 net profit estimate by 14% following the earnings beat, while keeping projections for H2 2026 through 2030 broadly unchanged — a signal that analysts view the outperformance as structural rather than a one-off demand pull-forward.
AI Demand Reshapes CXMT's Revenue Trajectory Overnight
The arithmetic behind CXMT's blowout quarter is straightforward, but the market forces driving it are not. Three converging dynamics collided in Q2 2026: explosive AI infrastructure spending that lifted DRAM pricing globally, a deliberate supply-chain diversification push by Chinese hyperscalers and device manufacturers away from U.S.-aligned memory vendors, and CXMT's own accelerating capacity ramp.
Gross margin for the first half reached 84.84%, compared with negative 2.19% as recently as 2023. The operating expense ratio improved to 5.4% in Q2, better than both Goldman's forecast and Bloomberg consensus, which explains why net profit beat consensus by a staggering 260%. That margin trajectory mirrors what Samsung Electronics experienced in 1987, when a global DRAM shortage triggered by U.S.-Japan trade friction allowed the Korean company to recover all prior semiconductor losses in a single fiscal year — a precedent that underscores how violently cyclical the memory industry can swing in either direction.
According to market research firm TrendForce, global DRAM spot prices surged more than 410% in 2025, with contract prices rising a further 80%-95% quarter-on-quarter in Q1 2026. AI servers require eight times the DRAM capacity of conventional servers, per Micron Technology estimates, and the migration of Samsung, SK Hynix, and Micron toward High Bandwidth Memory (HBM) production has left a structural gap in standard DRAM supply — precisely the segment where CXMT competes most aggressively.
Supply-Chain Nationalism Accelerates CXMT's Customer Penetration
Beyond pricing, CXMT is benefiting from a secular shift in procurement strategy among Chinese original equipment manufacturers. The company's DRAM products have now entered the supply chains of Xiaomi, OPPO, vivo, Lenovo, and Huawei — a roster that collectively represents a substantial share of global smartphone and server shipments.
Goldman's CHIPS 4 sector report projects China's domestic DRAM demand to grow at a 50% compound annual growth rate from 2026 to 2028, reaching US$257 billion by 2028. HBM demand within that market is forecast to expand at an even faster 188% CAGR over the same period, reaching US$32 billion by 2028, driven by AI server deployments. Goldman explicitly identifies CXMT as the primary domestic beneficiary of this demand curve.
The geopolitical dimension amplifies the commercial logic. With Washington restricting exports of extreme ultraviolet (EUV) lithography equipment and tightening controls on advanced deep ultraviolet (DUV) tools in coordination with the Netherlands and Japan, Chinese technology buyers face structurally elevated incentives to qualify domestic memory suppliers regardless of short-term cost differentials. CXMT is the only company in mainland China currently producing DRAM at commercial scale.
LPDDR6 Commercialization Signals Technology Ambition Beyond Current Node
CXMT's technology roadmap adds a forward-looking dimension to what might otherwise read as a pure cyclical earnings story. The company has developed a proprietary LPDDR6 chip with peak transfer speeds of 12,800 Mbps and maximum capacity of 16GB — specifications that represent a material step-up from its previous LPDDR5X generation. Samples have been delivered to key customers for validation, with mass production in active preparation.
Separately, CXMT is pursuing vertical transistor (VCT) architecture and 4F² cell structures as longer-term process alternatives. These approaches are designed to reduce dependence on the most advanced lithography nodes — a strategic hedge against continued export restrictions — though engineering challenges around yield, cost, and reliability remain unresolved.
On process geometry, the gap with global leaders remains measurable. Samsung and SK Hynix are manufacturing DDR5 on 12-14 nanometer D1a/D1α and D1b/D1β nodes, and SK Hynix has already moved HBM3E into mass production while advancing HBM4. CXMT, which began mass-producing DDR5 in late 2024 on a 16nm (1Z) node — bypassing the 17nm intermediate generation — trails by one to two technology generations. That gap is consequential: HBM, the highest-margin memory product in the AI server stack, remains out of reach for CXMT in the near term.
Goldman's Valuation Model Prices In Long-Term Growth, But Warns on 2027-2028 Supply Surge
Goldman's discounted price-to-earnings methodology anchors the RMB 129 target to a 2030 target P/E of 16.6x, derived from the correlation between forward trading multiples and earnings growth rates among comparable companies. The bank applies that multiple to 2030 estimated earnings per share, then discounts back to a 2027 present value using a 12.7% cost of equity — a framework that explicitly prices in CXMT's long-duration growth profile rather than near-term earnings momentum alone. CXMT's assumed 2030-31 average net profit growth rate is 21% year-on-year.
The bull case, however, carries an explicit caveat. Goldman flags that as global new capacity additions from Micron, SK Hynix, and Samsung concentrate in H2 2027 through 2028, pricing momentum will likely decelerate. Micron has raised its fiscal 2026 capital expenditure guidance from US$20 billion to US$25 billion and indicated spending above US$35 billion in fiscal 2027. SK Hynix announced approximately US$40 billion in domestic Korean capacity investment in August 2026. Samsung's 2026 capex plan exceeds US$70 billion. The collective scale of these commitments virtually guarantees a supply response — the only open question is timing and magnitude.
Morgan Stanley's chief semiconductor analyst Joseph Moore introduced the concept of "Chipflation" in a June 2026 research note, arguing that current pricing represents a structural reset rather than a conventional upcycle. BlueBox Asset Management fund manager William de Gale offered a counterpoint: "Every time people start declaring that the memory cycle has disappeared, I suspect things will play out as they have in the past — and then everything deteriorates rapidly."
CXMT's own filings acknowledge the risk. The company notes that if compute-in-memory, on-chip cache, or other alternative architectures achieve large-scale commercialization, they could reduce AI's dependence on conventional DRAM — a scenario that would compress the very demand tailwind that powered this quarter's results.
Impact Assessment: What CXMT's Breakout Means for the Broader Semiconductor Landscape
For investors, the H1 2026 results reframe CXMT from a speculative domestic-substitution play into a company generating cash at a rate that funds its own technology catch-up. The RMB 77.61 billion (US$10.78 billion) in first-half net profit not only erases accumulated losses but provides internal capital for the capacity and R&D investments needed to close the HBM gap.
For the global memory oligopoly — Samsung, SK Hynix, Micron — CXMT's emergence as a credible fourth player in standard DRAM introduces a pricing variable that did not exist three years ago. Its presence in Chinese OEM supply chains creates a floor under domestic demand that is partially insulated from U.S. export control escalation.
For China's broader semiconductor industry, the trajectory from founding to first profit took approximately ten years and required the convergence of an AI supercycle, geopolitical demand pull, and a decade of engineering investment. That combination is difficult to replicate — but the commercial proof point is now on the record.
Related Coverage:
CXMT Rejects Apple's Discount Demand, Signaling a Chip Supply Shift