China's CXMT Just Broke the Memory Chip Oligopoly

The number that shouldn’t exist yet

China’s ChangXin Memory Technologies, better known as CXMT, posted an 82% operating profit margin in the second quarter of 2026. That is not a typo. It is a figure that places the company above Samsung Electronics’ semiconductor division, SK Hynix, Micron, SanDisk, and Kioxia — all at once.

By operating profit margin, CXMT became the top performer among six major memory chip producers globally in the April-to-June quarter. By raw operating profit, the company surged from a 29 billion yen loss a year earlier to roughly 19 trillion yen — a swing so violent it reads like a financial anomaly rather than an industrial outcome.

The reason CXMT sits at the top of that list reveals a structural crack in a market that has spent the past decade teaching the world that memory chips are brutally cyclical and that only three countries — South Korea, the United States, and Japan — should be allowed to compete at scale.

Why CXMT’s margin story is so lopsided

CXMT did not leapfrog the incumbents by producing better HBM, the high-bandwidth memory that powers AI servers. It happened because the incumbents pivoted hard toward HBM, and CXMT stayed focused on general-purpose DRAM — DDR5 chips used in PCs, consoles, and enterprise workloads that sit far from the AI hype cycle but dominate volume.

When Samsung, SK Hynix, and Micron redirected wafer capacity to HBM, DDR5 supply tightened. Prices for DDR5 moved sharply upward. CXMT, which holds a much higher share of its revenue in DDR5, felt those price moves almost immediately. HBM contracts, by contrast, tend to be annual agreements with hyperscalers, muting short-term price volatility. As TrendForce noted, HBM profitability had begun trailing DDR5 already in the first quarter of 2026. A Japanese advisor at KPMG FAS put it more plainly: in a cycle like this, the more general DRAM exposure you carry, the more you benefit from the rebound.

CXMT is also not fighting the same cost structure as its competitors. The company is developing its fabrication capacity in Hefei and Shanghai under state-backed financing, which means lower cost of capital than Micron or Samsung can deploy from publicly traded balance sheets. Its free cash flow over January through June reached roughly 22 trillion yen, an improvement of about 28 trillion yen year over year. That cash conversion, combined with lower capital costs, turns a DRAM price move into outsized margin.

The customers behind the numbers

CXMT’s client roster reads like a list of China’s most politically sensitive tech firms: Alibaba, ByteDance, and Tencent. These are the companies the U.S. has tried hardest to restrict from accessing leading-edge memory. Paradoxically, those restrictions may have helped CXMT more than any subsidy.

Western export controls tightened around advanced memory access for Chinese data centers. Rather than forcing these firms to look elsewhere, the constraints funneled procurement inward. For Alibaba and Tencent, buying CXMT DDR5 is no longer a pragmatic compromise — it is the default path. ByteDance, whose infrastructure supports TikTok’s global traffic, faces the same political geometry. CXMT’s demand floor is therefore more secure than a typical DRAM producer’s, which depends on a thin slice of global spending that can evaporate in a downturn.

This customer base also explains the market-cap re-rating. According to QUICK and FactSet data cited in Nikkei, CXMT’s shares — excluding treasury stock — are valued at roughly 87 trillion yen, pushing it ahead of Tencent on paper and making it the largest-cap memory company in China by a wide margin. For comparison, Kioxia trades at about 31 trillion yen. CXMT’s price-to-earnings ratio sits near 20, compared with single digits for the Korean and Japanese incumbents. The market is pricing CXMT not as a cyclical commodity player but as a structural supplier to one of the world’s largest domestic tech ecosystems.

What changes — and what does not

Several things change if this margin sequence holds.

First, the pricing power of the Big Three erodes faster than most models assumed. The memory oligopoly worked because Samsung, SK Hynix, and Micron coordinated capacity cuts during downturns and rode price upcycles together. If CXMT and its NAND counterpart YMTC have enough cash to expand independently, the old coordination mechanism weakens. Nikkei flagged this directly: expanded Chinese capacity paired with aggressive pricing could increase memory price volatility and reshape the existing market order.

Second, the geopolitical leverage equation tilts. The United States restricted Chinese access to leading memory technology partly to preserve a strategic chokepoint. CXMT’s rise suggests that chokepoint is less effective than intended — not because China has closed the gap on HBM, but because it has captured the volume business that funds the capital expenditure required to catch up. Memory economics are circular: you need cash from today’s volume chips to build tomorrow’s leading-edge fabs. CXMT now has that cash.

Third, the incumbents face a timing problem. SK Hynix and Samsung are deep into HBM ramp, and the capital tied up there cannot easily pivot back to DDR5 without accepting yield losses and margin compression. Micron, the only American memory player, faces the same constraint. For the next 12 to 18 months, DDR5 prices are likely to remain supportive, which means CXMT will continue to generate strong cash flows precisely when its Hefei and Shanghai expansions require them. That compounding dynamic is what makes this quarter look like an outlier but could soon look like a trend line.

What does not change

CXMT is not producing HBM at anything close to the volume or performance needed for leading AI training clusters. In that category, the gap to SK Hynix and Samsung remains wide. The company also operates under ongoing U.S. restrictions on equipment, which means its fab expansion may run into friction that its cash flow does not fully resolve.

Its market-cap lead over Tencent is calculated on paper, not realized in trading liquidity, and the Chinese equity markets punish valuation stretches quickly. A single bad quarter could compress the multiple dramatically, given the cyclical nature of DRAM earnings.

And the incumbents are not idle. Samsung and SK Hynix have publicly committed tens of billions to new capacity over the next few years. If DDR5 prices soften by late 2027 or 2028, CXMT’s margin advantage narrows just as its new fabs come online and depreciation begins to bite.

The bigger shift

The headline number — 82% — is dramatic. The deeper story is that China’s memory industry has found a wedge into a market that spent years insulating itself through alliance-style capacity management. CXMT is exploiting a gap opened by the incumbents’ own pivot toward AI memory. That gap may be temporary. The capability it builds in the meantime will not be.

If Chinese memory firms translate this cycle’s cash generation into durable capacity and yield improvements over the next two years, the global balance of memory supply shifts irreversibly. For Samsung and SK Hynix, that means losing the pricing discipline that has sustained their margins. For Washington, it means the export-control architecture is containing the symptom — leading-edge chip access — while the disease, domestic Chinese production, keeps growing.

CXMT’s Q2 result is not the end of the memory oligopoly. But it is the first quarter in decades where the oligopoly did not lead the margin table.