How CXMT Became the Most Profitable Memory Chipmaker in Q2 2026

CXMT’s extraordinary second-quarter profitability was primarily a memory-cycle and product-mix story. A supply squeeze in mainstream DRAM raised prices just as the Chinese manufacturer had scaled production and secured large domestic customers. The shift by leading suppliers toward high-bandwidth memory (HBM) for AI servers helped tighten supply of conventional DRAM, creating an unusually favorable market for CXMT’s core products. 45

The 82% margin: a pricing windfall in standard DRAM

QUICK FactSet data cited by Nikkei and other reports put CXMT’s Q2 2026 EBIT margin at 82%, ahead of SK hynix at 76% and Samsung Electronics’ semiconductor division at 70%. The comparison covered major memory manufacturers’ reported results for the period. 14

The key is what CXMT was selling. HBM is specialized stacked memory used alongside AI accelerators, while CXMT’s business has been concentrated in more broadly used DRAM. As Samsung, SK hynix and Micron directed capacity and investment toward AI-related memory, conventional DRAM became tighter. That strengthened pricing for the segment in which CXMT had meaningful available output. 5

This also explains the scale of the turnaround. CXMT’s first-half gross margin reached 84.84%, and its Q2 gross margin was reported at 87.59%, after improving from roughly 79% in the first quarter. With a largely fixed manufacturing cost base, rapidly rising memory prices can push profit up much faster than revenue.

From loss to large first-half profit

CXMT reported first-half 2026 revenue of RMB150.31 billion, up 873.64% from a year earlier. Net profit attributable to shareholders was RMB77.61 billion, reversing a RMB2.33 billion loss in the comparable 2025 period.

Quarterly reporting cited by TrendForce put Q2 net profit at RMB52.8 billion, up 113% from the prior quarter. The same reporting said Q2 gross margin reached 87.59%. A separate report calculated Q2 revenue at about RMB99.5 billion from the company’s semiannual figures.

The numbers demonstrate the operating leverage of commodity memory: when demand exceeds available supply, a producer that can keep fabs highly utilized can convert a large portion of incremental sales into profit. They do not, by themselves, establish a permanent cost or technology advantage.

China’s cloud buyers gave CXMT a large home market

CXMT’s domestic customer base is central to its growth. Reuters reported that the company signed a long-term Tencent supply agreement worth more than RMB20 billion (about $2.94 billion) and identified Tencent, Alibaba Cloud, ByteDance, Lenovo and Xiaomi as major customers in its IPO prospectus.

Reuters also reported a five-year agreement with ByteDance worth more than $7 billion. These customer relationships can support utilization and provide demand visibility at a time when global DRAM supply is constrained. They also reduce the company’s reliance on winning overseas qualifications immediately.

A 10% DRAM share makes CXMT harder to ignore

CXMT’s global DRAM revenue share reportedly rose from 4% a year earlier to 10% in Q2 2026, putting it fourth worldwide by revenue share. 16 That is meaningful commercial progress: a company that was once peripheral to the global market has become large enough to influence supply expectations and customer sourcing decisions.

Still, a revenue-share gain during a strong price cycle is not the same as parity across all memory products. CXMT’s current advantage is most evident in mainstream DRAM and China-based demand. HBM leadership depends on more than wafer volume, including yields, stacking, packaging and customer qualification. Reporting earlier in 2026 indicated that only a small portion of CXMT capacity was allocated to HBM, with the company’s HBM transition still at an early stage.

Capacity expansion could reshape the commodity-DRAM market

Morgan Stanley estimates cited in reporting project CXMT’s monthly DRAM wafer capacity rising from about 180,000 wafers in 2025 to 300,000 in 2026—roughly 13% of global DRAM wafer capacity and 11% of bit shipments. The estimate reaches 500,000 wafers per month by 2028.

Reuters separately reported that CXMT was building plants in Shanghai and Hefei and discussing another project, with the expansion potentially taking output above 600,000 wafers per month when new fabs come online.

These are projections and plans, not guaranteed production. But if executed, they would give CXMT more ability to supply DDR5 and other mainstream DRAM at scale—and more potential influence over the next memory downcycle as well as the current upcycle.

Korea is also winning the AI-memory boom

CXMT’s quarter should not be read as evidence that South Korea’s memory industry is losing the broader AI cycle. South Korea’s semiconductor exports reached $281 billion in the first eight months of 2026, up 169.6% year over year and accounting for 41% of national exports, according to customs data reported by Reuters. 17

By early September, South Korea’s year-to-date exports had reached $709.4 billion, exceeding the country’s prior full-year record. 17 The data point to robust demand for Korean chip exports, including the high-value memory used in AI infrastructure.

The often-cited “Korea discount”—the valuation discount sometimes applied to Korean equities—is a separate question from chip-export performance. The supplied evidence confirms the export boom, but does not provide a current comparable valuation measure showing whether that discount has narrowed or persisted.

What the CXMT result means

CXMT has shown that it can turn a favorable DRAM cycle into exceptional profitability. Its combination of domestic demand, sharply higher conventional-DRAM prices and increasing production scale produced an 82% EBIT margin that exceeded the reported margins of much larger rivals in Q2. 14

The competitive balance remains unresolved. Korean leaders retain an important position in the high-end AI-memory market, while CXMT is becoming a formidable supplier of mainstream DRAM. The strategic risk for incumbents is cumulative: sustained cash flow from future upcycles could fund more Chinese capacity, process development and customer qualification. Yet expanding capacity alone does not guarantee leadership in HBM.

For now, CXMT’s result is best understood as a major commercial breakthrough in commodity DRAM—and a warning that the market’s next competitive contest may be shaped as much by scale and supply discipline as by frontier-memory technology.