Chinese DRAM maker ChangXin Memory Technologies (CXMT) reportedly quoted Apple prices on par with the three global leaders, Samsung Electronics, SK hynix and Micron Technology, when Apple asked for a memory price quote, and did not agree to a discount. It also reportedly turned down a DDR5 procurement inquiry from a Taiwanese server maker. The likely backdrop is strong domestic demand for memory in China, which CXMT's production capacity cannot keep up with.
Until now, Chinese semiconductor makers have generally viewed deals with global giants like Apple as a way to demonstrate product reliability and build a presence in overseas markets. But CXMT is already stretched just supplying domestic customers, so it appears to have little need to cut prices to win overseas orders. The report suggests China's DRAM industry may be shifting from a stage of winning markets through price competition to one in which, backed by domestic demand, it can choose whom to supply.
Refusing Apple a Discount, and Passing on a Taiwanese Server Maker
According to a report by the Taiwanese semiconductor outlet TechNews on October 25, 2026, Apple asked CXMT for a memory price quote, but the two sides did not agree on price because CXMT quoted the same level as the three global leaders. The article cites a senior industry analyst, who described CXMT as unwilling to offer large discounts to overseas customers.
Earlier, Korea's Maeil Business Newspaper reported on August 6 that DRAM supply negotiations between Apple and CXMT had failed to reach agreement on price. The talks reportedly covered LPDDR5X for use in next-generation iPhones and other products, and CXMT is said to have held prices at or above those of Samsung and SK hynix. However, neither Apple nor CXMT has officially explained the details of the negotiations, so it cannot be ruled out that conditions other than price also contributed to the lack of agreement.
TechNews also cites a vice president of a Taiwanese server maker who said CXMT declined to supply DDR5 when he approached the company about procurement. The executive said he did not know whether the refusal stemmed from Chinese government policy or from other circumstances. It is therefore not possible to conclude that CXMT is uniformly restricting supply to overseas companies.
Still, it is quite plausible that CXMT's limited spare capacity lies behind the failed talks. If its production slots are already filled with domestic orders, there is little economic benefit in cutting prices to meet the demands of large overseas buyers. Understanding these negotiations requires looking at China's domestic demand and CXMT's production capacity together.
Capacity Strained by Domestic Demand Alone, and Pricing That Doesn't Rely on Discounts
The prospectus CXMT filed in May 2026 for its initial public offering (IPO) on the Shanghai Stock Exchange states that its supply capacity falls far short of demand in the Chinese domestic market. Behind this are China's push for semiconductor self-sufficiency and growing demand for memory in cloud services, smartphones and PCs.
CXMT's customers include major tech companies such as Alibaba Cloud, ByteDance and Tencent, as well as device makers such as Lenovo, Xiaomi, Transsion, Honor, OPPO and vivo. Sales are made both directly and through distributors. According to the prospectus, the top five customers accounted for 74.12%, 67.30% and 68.08% of revenue in the last three fiscal periods. Roughly 70% of revenue is concentrated in a small number of large customers, which shows how much the business depends on its major accounts. Note, however, that these figures are based on revenue and do not directly show the share of wafer capacity allocated to each company.
The tight supply-demand balance is also affecting CXMT's results. The company attributed the sharp rise in first-quarter 2026 revenue to higher shipment volumes and an improved product mix, as well as to the supply shortage and price increases across the DRAM market. When demand exceeds supply, manufacturers tend to prioritize supply to existing customers and profitability over winning new customers through discounts.
Huang Mao-yuan of market research firm Yole Group told TechNews that about 90% of CXMT's production capacity is used for Chinese domestic brands. He also analyzed that meeting the demand of the entire Chinese market would take a decade-scale effort. These are estimates and views from an outside analyst, not customer-by-customer supply figures published by CXMT.
China's government-backed push to localize semiconductors is also likely one reason for prioritizing domestic customers. Even without invoking policy, though, there is little reason to offer large discounts to overseas companies when large domestic customers are placing enough orders. The reports that talks with Apple stalled may reflect the size of the domestic demand that underpins CXMT's bargaining power.
LPDDR6 and the New-Generation "G5" DRAM Move Into Mass Production
Recent product announcements also suggest CXMT is closing in on the established overseas DRAM makers. On September 7, 2026, the company announced that it had begun mass production of LPDDR6, a next-generation low-power DRAM, and that it had been adopted in Xiaomi's foldable smartphone, the Xiaomi 18 Fold.
According to the announcement, this LPDDR6 is a 16GB package combining 16Gb DRAM dies and uses a 1295-ball FBGA package. Its maximum data rate reaches 12,800Mbps, but when paired with the SoC in Xiaomi's device it reportedly runs at 10,667Mbps. Because the maximum transfer rate differs from the rate used in actual devices, the maximum spec value cannot be taken as real-world performance.
CXMT says this LPDDR6 offers up to 60% higher peak bandwidth and 20% lower power consumption than LPDDR5X running at 10,667Mbps. However, the bandwidth improvement is a comparison based on theoretical peak values, and the power figure comes from the company's internal testing. The same improvements have not been confirmed by independent third-party measurements on actual devices. In addition, data rate and bandwidth are not the same metric; the latter also depends on factors such as data bus width. Some reports mention a pairing with Xiaomi's in-house processor, the "XRING 03," but CXMT's September 7 announcement does not name the processor.
On September 20, CXMT also announced the start of mass production of "G5," its fifth-generation DRAM manufacturing platform, and of 24Gb LPDDR5X products built on it. According to the company, G5 yields more than 50% more total dies per wafer than the previous-generation G4. However, this figure is a comparison converted to 8Gb capacity and represents the die count before sorting good dies from bad. It does not mean yield improved by 50%, and the number of good dies that can actually be shipped has not been disclosed.
CXMT also offers 16Gb and 24Gb DDR5 products for servers and PCs, and its lineup includes modules supporting up to 8,000Mbps. While it is narrowing the gap with the major makers in advanced DRAM product specifications, the competitiveness of its mass-production technology cannot be judged from published peak performance alone. What matters is whether it can sustain stable yields over the long term and meet the quality and volume that large customers require.
Toward 360,000 Wafers per Month in 2027: Capacity Expansion Plans
To meet growing demand at home and abroad, CXMT is expanding its production capacity. According to a forecast by research firm Isaiah Research, cited by TechNews, three new fabrication lines will come online in sequence during 2026, bringing year-end wafer capacity to 300,000 wafers per month. In 2027, expansion of the Shanghai manufacturing site is also expected to contribute, taking capacity to around 360,000 wafers per month.
| Period | Monthly wafer capacity | Basis of figure |
|---|---|---|
| End of 2024 | 160,000 wafers | Disclosed actual |
| End of 2025 | 220,000 wafers | Disclosed actual |
| End of 2026 | 300,000 wafers | Isaiah Research forecast |
| 2027 | 360,000 wafers | Isaiah Research forecast |
The figures for the end of 2024 and 2025 are based on disclosed results, while those for 2026 onward should be distinguished as forecasts from an outside research firm. In particular, nominal equipment capacity, the number of wafers actually started, and the number of good dies ultimately shipped are different metrics. Even if new fabs raise processing capacity, product shipments will not necessarily increase by the same proportion right away.
CXMT operates 12-inch DRAM fabs in Hefei and Beijing. In its IPO prospectus, citing Omdia statistics, it said it has grown into the world's fourth-largest DRAM maker by production capacity, shipment volume and revenue. The initial prospectus also planned to raise 29.5 billion yuan, with proceeds earmarked for upgrading production lines, advancing DRAM processes and R&D on cutting-edge technology. The company subsequently listed on the STAR Market of the Shanghai Stock Exchange on July 27, 2026, so the 29.5 billion yuan figure should be treated as a pre-listing fundraising plan.
The capacity expansion is a development that established DRAM makers such as Taiwan's Nanya Technology cannot ignore. If CXMT can supply advanced DDR5 and LPDDR products stably and in volume, competition could intensify. However, a rise in monthly wafer output alone will not determine market share. Evaluation must also take into account manufacturing cost, yield, per-product performance and customer quality qualification.
HBM: A Technology Gap Remains, and U.S. Export Controls Are a Barrier
Although CXMT is rapidly raising its profile in general-purpose and mobile memory, it still faces technical challenges in HBM (high-bandwidth memory), which is essential for AI accelerators. Yole Group's Huang Mao-yuan told TechNews that CXMT's HBM technology is at least two generations behind the global leaders. This is an outside analyst's assessment of the technology level and does not directly indicate a generation gap in product specifications or measured performance.
Manufacturing HBM requires advanced stacking technology, in which multiple DRAM dies are stacked vertically and connected through through-silicon vias (TSVs) and similar structures. On top of the demands of ordinary DRAM, it adds manufacturing challenges such as die thinning, forming fine interconnects, ensuring stacking accuracy and designing for heat dissipation. Because defects arising at each step affect the yield of the final product, being able to make individual DRAM dies is not the same as being able to mass-produce HBM reliably.
CXMT also faces constraints from U.S. export controls on advanced chipmaking equipment. Between regulations from the U.S. Commerce Department's Bureau of Industry and Security (BIS) and export controls in countries such as the Netherlands, it is difficult for Chinese fabs to obtain some advanced equipment, including EUV (extreme ultraviolet) lithography tools. EUV lithography is not strictly required for all HBM manufacturing, but it narrows the options for developing advanced DRAM processes and choosing manufacturing methods. Repeating multiple patterning on existing ArF immersion lithography tools increases the number of process steps and masks, adding to manufacturing cost and the burden of yield management.
That said, it cannot be concluded that CXMT's HBM development remains at the prototype stage. The Information reported on August 31, 2026 that the company had begun small-volume production of HBM3E and planned to ramp up in 2027. Evaluations by Chinese AI chipmakers for adoption are reportedly under way, but CXMT itself has not officially disclosed details on HBM3E production scale or commercial shipments. A report that small-volume production and customer evaluations have begun should be distinguished from stable mass production with solid yields and sustained commercial shipments.
CXMT has already officially announced the start of mass production of LPDDR6 and G5-generation LPDDR5X, and it is advancing technology development and business scale in general-purpose and mobile DRAM. For HBM, however, issues remain to be solved, including the maturity of stacking and bonding technology, yield, customer qualification and constraints on procuring manufacturing equipment.
The reports on the price negotiations with Apple suggest that CXMT, backed by Chinese domestic demand, is approaching a position in which it can grow its business without yielding to discount demands from global giants. However, a supply shortage in the domestic market and technical competitiveness in the global market are separate matters. Going forward, key measures of the gap with the three global leaders will be how far actual supply grows as new fabs come online, and whether CXMT can achieve stable mass production of advanced DRAM and commercial shipments of HBM.