CXMT Just Passed Micron, SK-Hynix Gross Margins. The Lead Is Narrower Than It Looks
CXMT reported an 87.59% second-quarter gross margin, narrowly topping the latest Micron SK-Hynix benchmarks after entering 2025 with a far weaker cost structure. The comparison marks a striking financial reversal for China’s largest DRAM producer. It does not establish that CXMT has surpassed either rival in technology, scale, or durable profitability.
The claim began circulating after CXMT released its first interim report as a public company on August 29, 2026. Its first-half revenue reached RMB 150.3 billion, up 873.64% from a year earlier. Net income reached RMB 77.6 billion, while the first-half gross margin rose to roughly 84.8%.
CXMT’s second-quarter margin was higher still. Reporting based on the filing calculated a 79.15% first-quarter gross margin and an 87.59% second-quarter result. That compared with Micron’s 84.6% GAAP margin and SK Hynix’s approximately 83.2% margin in their latest reported quarters.
Those numbers deserve attention because CXMT had a 13% consolidated gross margin in the first half of 2025. Its original prospectus described a manufacturer still climbing out of losses, high depreciation costs, and limited economies of scale.
The reversal is real. The implied competitive conclusion remains unsettled.
What CXMT Actually Reported
CXMT moved from a marginally profitable producer to an 80%-plus-margin manufacturer within one year, primarily as DRAM prices and shipment volumes surged.
The company’s August 29 disclosure covered the six months ending June 30, 2026. Revenue reached RMB 150.3 billion, compared with about RMB 15.4 billion during the same period in 2025. Net income reached RMB 77.6 billion, reversing the prior-year loss.
The first-half gross margin was reported at about 84.8%. The second-quarter figure reached 87.59%, according to calculations from the interim results. That was 8.44 percentage points above the first-quarter level.
The latest interim results attribute the improvement to sharply higher DRAM prices and stronger sales volume. CXMT said global DRAM supply remained tight as demand expanded across AI servers, mobile devices, and conventional computing.
The scale of the change becomes clearer beside its listing documents. CXMT reported a 13% consolidated gross margin for the first half of 2025. Before inventory write-down reversals, the margin was only 5.02%.
Its core DDR and LPDDR products looked weaker under that adjusted measure. DDR products generated an 18.9% margin, while LPDDR products produced only 0.71%.
The company’s IPO prospectus said depreciation, production ramp costs, and incomplete economies of scale had weighed on profitability. CXMT was still spending heavily to close manufacturing and product gaps with established DRAM suppliers.
That historical starting point matters. An 87.59% quarterly margin is not a routine improvement from an already mature base. It represents a sharp reversal from a manufacturer that had recently struggled to cover production costs.
The second-quarter result also arrived shortly after CXMT entered public markets. Its shares began trading in Shanghai on July 27, following an offering that raised RMB 57.92 billion. The stock gained 466% on its first trading day, according to coverage of the Shanghai listing.
Public investors therefore received an unusually strong first earnings report. Revenue, margins, and net income all reflected one of the most favorable DRAM pricing environments in recent history.
Yet gross margin measures the difference between revenue and the direct cost of producing goods. It does not account for research spending, administrative expenses, financing costs, or future capacity investments.
It also does not measure manufacturing yields directly. Yield refers to the share of usable chips produced from each wafer. Two suppliers can report similar margins during a shortage while operating with different yields, product mixes, and long-term cost positions.
CXMT’s result confirms that it captured exceptional economics during the current shortage. It does not disclose enough operational detail to show whether those economics will persist when supply conditions normalize.
The Micron SK-Hynix Comparison Is Real but Uneven
CXMT topped both reported gross-margin benchmarks, although the three figures cover different reporting periods, accounting conventions, and product portfolios.
Micron reported fiscal third-quarter results on June 24 for the period ending May 28, 2026. Its revenue reached $41.46 billion, while GAAP gross margin was 84.6%.
Micron also reported an 84.9% non-GAAP margin. The adjustment mainly excluded stock-based compensation from cost of goods sold. Its official quarterly filing shows both figures, allowing a direct distinction between the two measures.
Against Micron’s GAAP result, CXMT led by 2.99 percentage points. Against Micron’s non-GAAP result, the gap was 2.69 points.
SK Hynix reported calendar second-quarter revenue of KRW 79.319 trillion and gross profit of KRW 65.991 trillion. Those figures imply a gross margin of approximately 83.2%.
CXMT’s reported second-quarter margin was therefore about 4.4 percentage points higher. SK Hynix’s reviewed interim statements provide the revenue, cost-of-sales, and gross-profit figures behind that calculation.
The comparison is mathematically sound, but it is not perfectly matched. CXMT and SK Hynix reported the three months ending June 30. Micron’s quarter ended more than one month earlier, on May 28.
That timing difference matters in a fast-rising price cycle. A company recognizing more June shipments could capture higher contract and spot prices than one whose reported quarter ended in May.
The companies also sell different combinations of products. CXMT’s principal business remains centered on conventional DDR and low-power LPDDR memory. LPDDR contributed 65.86% of its principal-business revenue during the first half, while DDR contributed 31.60%.
Micron combines DRAM and NAND results at the consolidated level. Its cloud, core data center, mobile and client, and automotive units each contain different products and customer agreements.
SK Hynix also combines DRAM and NAND, but it has a stronger position in high-bandwidth memory. HBM stacks multiple memory dies to deliver the bandwidth required by AI accelerators.
HBM consumes more wafer capacity and requires advanced packaging. It also involves customer qualification cycles that differ from those for ordinary PC or smartphone memory.
These differences do not invalidate CXMT’s number. They limit what the number proves.
A quarterly gross margin answers a focused question: how much gross profit did the company retain from each unit of revenue during that period? CXMT’s answer was better than those of Micron and SK Hynix.
It does not answer whether CXMT has lower costs for equivalent products. It also does not show whether customers view its most advanced products as substitutes for HBM4 or other premium memory.
This is why the Micron SK-Hynix comparison should be treated as a financial milestone, not a complete competitive ranking. CXMT has joined the same margin range during a favorable quarter. It has not disclosed enough evidence to show parity across process technology, yields, product breadth, or customer access.
A Memory Shortage Turned Scale Into Margin
CXMT’s margin expansion reflects a combination of higher prices, rising output, and a more valuable product mix, rather than one isolated manufacturing advance.
DRAM is famously cyclical. Suppliers make large capital commitments years before they know the exact balance between future demand and available capacity.
When production exceeds demand, prices can fall below the cost of manufacturing chips. When supply becomes constrained, existing capacity gains exceptional pricing power.
CXMT experienced both sides within a short period. Its prospectus recorded a negative 112.71% gross margin before inventory write-down reversals in 2023. The comparable measure improved to negative 4.03% in 2024 and 5.02% during the first half of 2025.
The company’s main DRAM average selling prices then rose substantially. Reporting based on its filings indicates that prices for its two major product groups increased by 61% and 24.46% in 2025. Tightening supply accelerated the improvement during 2026.
Price increases affect gross margin quickly because semiconductor manufacturing has high fixed costs. Once a factory is operating, each additional usable chip spreads depreciation and other production expenses across more saleable output.
Higher utilization therefore improves unit economics. Better yields can amplify that effect because more chips from each wafer become revenue-generating products.
CXMT also shifted toward newer memory. DDR5 gained share as customers adopted higher-performance systems. LPDDR5 expanded across premium smartphones and other mobile devices.
The company’s interim report described LPDDR6 products with data rates reaching 12,800 megabits per second and capacities reaching 16 gigabytes per chip. CXMT said it had begun customer sampling and was preparing for mass production.
Some reporting went further, claiming that mass production had already started for an upcoming Xiaomi device. CXMT’s own disclosed language appears more cautious, so the customer qualification and volume ramp remain important verification points.
This product transition can support margins because new generations initially command higher prices. It also raises execution risks.
Advanced products require tighter manufacturing tolerances. Customers need confidence in reliability, power consumption, thermal behavior, and sustained output.
A favorable market can conceal some cost differences. If nearly every qualified DRAM supplier has more orders than available output, customers have fewer alternatives and producers gain pricing leverage.
That condition appears central to CXMT’s second-quarter result. The company said the global DRAM market would remain undersupplied during the second half of 2026.
Micron described similar forces. Its fiscal third-quarter DRAM revenue increased 211% from a year earlier, driven mainly by an approximate 140% increase in average selling prices and a roughly 30% increase in bit shipments.
Micron said higher selling prices were the primary reason its consolidated margin rose from 74.4% to 84.6% sequentially. Favorable product mix and manufacturing cost reductions provided additional support.
SK Hynix also linked its results to sustained AI infrastructure demand. High-performance server products led the price increases, while long-term supply agreements gave the company more visibility into customer demand.
All three manufacturers are therefore benefiting from the same market mechanism. AI investment has absorbed premium memory, while capacity allocation has tightened conventional DRAM supply.
CXMT’s reversal stands out because its starting point was much lower. The company did not need to overtake Micron or SK Hynix in HBM technology to post the highest quarterly gross margin.
It needed scarce output, rising average prices, better factory utilization, and enough qualified products to capture demand. During the second quarter, that combination produced extraordinary profitability.
What the Margin Does Not Prove
One quarter above Micron and SK Hynix does not establish a permanent cost advantage, especially while prices, inventory accounting, and product mix remain unusually favorable.
The first uncertainty concerns durability. CXMT’s margin changed from 13% in the first half of 2025 to about 84.8% one year later.
That speed shows operational leverage, but it also exposes dependence on market conditions. A durable structural advantage usually produces more stable performance across several stages of a cycle.
The second uncertainty concerns inventory accounting. Semiconductor companies reduce inventory values when expected selling prices fall below recorded costs. When conditions improve, later sales of that inventory can produce accounting benefits.
CXMT’s prospectus explicitly separated reported margins from figures calculated before inventory write-down reversals. For the first half of 2025, the gap was almost eight percentage points.
The current interim figures deserve the same scrutiny. Investors need the complete cost disclosures to determine how much of the latest margin came from current production economics and how much came from accounting effects.
The third uncertainty is product comparability. CXMT generated most of its principal-business revenue from LPDDR. SK Hynix and Micron have greater exposure to premium AI memory and broader storage portfolios.
A higher consolidated margin does not mean CXMT can produce equivalent HBM at a lower cost. It can instead reflect stronger pricing for the products CXMT already ships.
HBM remains especially important because AI accelerator demand is reshaping the industry. SK Hynix began mass shipments of HBM4 during the second quarter and said it would expand production in the second half.
Micron also reports substantial cloud and data center exposure. Its core data center business posted an 87% gross margin during fiscal Q3, slightly below CXMT’s consolidated second-quarter result.
That detail shows why company-wide comparisons can obscure the real opponent. CXMT is competing strongly in conventional and mobile DRAM. The established suppliers still hold deeper positions in the highest-end AI memory market.
CXMT’s listing plan reinforces this distinction. The company allocated RMB 29.5 billion across DRAM technology upgrades, next-generation research, and manufacturing-line improvements.
Public documents did not identify a dedicated HBM project within those named uses. Reporting on the capital program also noted that CXMT had not disclosed HBM yields or near-term production commitments.
The fourth uncertainty involves output quality. CXMT has not published detailed yields for its newest LPDDR products. It has also not disclosed enough volume data to confirm how quickly customer samples will become large commercial shipments.
Sampling is meaningful because customers generally will not begin evaluation without credible hardware. It is still different from sustained, high-volume production across multiple buyers.
The fifth uncertainty is external access. United States export controls restrict China’s access to some advanced semiconductor equipment. CXMT must continue improving density, yield, and energy efficiency within those constraints.
This challenge does not prevent progress. It can raise the time and capital required to match production methods used by competitors with broader equipment access.
The sixth uncertainty is scale. CXMT has grown rapidly, but the established memory companies retain much larger global operations, customer networks, and intellectual-property portfolios.
CXMT held about 7.67% of global DRAM sales in 2025, according to figures cited from its listing materials. That is enough to influence supply, particularly inside China, but it remains far below the combined position of Samsung, SK Hynix, and Micron.
Scale matters when the cycle turns. Large suppliers can redirect output between products, negotiate long-term contracts, and spread research expenses across a wider revenue base.
CXMT’s 87.59% margin proves that the company captured the current upswing. The missing evidence concerns resilience during weaker pricing.
The decisive test will come when average selling prices stop rising. If CXMT can preserve strong margins while expanding newer products, its result will look structural.
If margins fall quickly with DRAM prices, the second quarter will look more like a peak-cycle event. That would still represent major progress, but it would support a narrower competitive conclusion.
Three Signals Will Show Whether CXMT’s Lead Can Last
Customer qualification, post-shortage margins, and disclosed investment priorities will determine whether the second-quarter result becomes a new baseline or a temporary peak.
The first signal is commercial LPDDR6 volume. CXMT says its product reaches 12,800 megabits per second and supports capacities up to 16 gigabytes per chip.
Those specifications place it in the next generation of low-power memory for phones, servers, and vehicles. The stronger evidence will be repeat orders, several named customers, and sustained production volumes.
A successful rollout would strengthen the case that CXMT’s margin reflects product progress as well as pricing. Delays, limited customers, or undisclosed yields would weaken that interpretation.
The second signal is gross margin after DRAM prices stabilize. CXMT, Micron, and SK Hynix all benefited from tight supply during their latest quarters.
Future filings should separate selling-price effects, product-mix changes, production costs, and inventory adjustments. That detail will reveal whether CXMT’s cost structure has permanently improved.
A margin that remains competitive through slower pricing growth would strengthen the claim of structural parity. A steep decline would indicate that scarcity supplied much of the advantage.
The third signal is CXMT’s capital allocation toward advanced server memory. Conventional DRAM can generate substantial cash, but AI infrastructure is directing industry investment toward HBM and specialized server products.
CXMT’s IPO funded broad production and research projects. Investors still lack a detailed, independently verified timetable for competitive HBM volume.
A disclosed production plan, customer qualification, and measurable shipment ramp would narrow the product gap with SK Hynix and Micron. Continued silence would leave CXMT’s competitive position concentrated in conventional and mobile memory.
Readers should also watch the incumbents’ response. Micron is investing at record levels and using multiyear customer agreements to improve demand visibility. SK Hynix is expanding HBM4 output while developing HBM4E.
Both companies can defend margins through premium products even if ordinary DRAM pricing weakens. CXMT must show that its own mix can evolve quickly enough to preserve profitability.
For PC and smartphone buyers, the immediate implication is more supply from a credible fourth producer. Greater output can eventually moderate memory prices, although the current shortage still favors manufacturers.
For cloud operators and AI developers, the distinction between ordinary DRAM and HBM remains critical. CXMT’s gross-margin lead does not yet mean it can replace the premium memory used beside leading accelerators.
For semiconductor investors, the central question is no longer whether CXMT can make money. The company has answered that during the current cycle.
The harder question is how much it can retain when scarcity eases. Track LPDDR6 shipments, adjusted gross margins, and advanced-memory investment over the next three reporting periods. Those signals will show whether the Micron SK-Hynix comparison marks a lasting competitive shift or an exceptional quarter produced by an exceptional shortage.
