China’s two memory powerhouses team up and splurge $63 billion to ramp up production, while CXMT relies on “crude steel-making” to violently jack up DDR5 output capacity.

Zhejiang Xin Storage and Yangtze Memory together sink 63 billion yuan into capacity expansion. Zhejiang Xin Storage expects monthly capacity to jump to 350k wafers by end-2026, nearing Micron’s estimated同期 capacity of 375k wafers. It injects variables into the global DRAM and NAND supply-demand map.
（Background: Samsung and SK hynix splashed 2,000 trillion won to build a wafer fab! Betting on the next 10 years of memory—Gwangju becomes the biggest winner）
（Background add-on: HBM memory already accounts for 63% of AI chip costs; hynix, Samsung, and Micron pocket the compute-power pricing power）

Table of contents

Toggle

• Without EUV, use brute force to get around it
• The capacity arithmetic that closes in on Micron?
• Follow the money—where it goes, the landscape follows

China’s two leading memory makers, although they can’t get the most advanced equipment due to U.S. sanctions, still want to capture a slice of this wave of exploding demand from AI servers. So Zhejiang Xin Storage teamed up with Yangtze Memory, planning to jointly spend 63 billion yuan in capital expenditures to purchase production lines.

Without EUV, use brute force to get around it

The red line Zhejiang Xin Storage faces is the EUV exposure machine—high-end lithography equipment that can etch finer circuits on wafers. This is the part most tightly locked down by U.S. sanctions. Without it, advanced process theory is, in principle, hard to make progress. That has been one of the reasons external observers have, for a stretch of years, believed Chinese memory would never catch up with the global “Big Three.”

Zhejiang Xin Storage’s solution is very “low-tech”: switch to older-generation DUV equipment, and stack multi-patterning exposure technology. In simple terms, it exposes the same wafer multiple times, trading the number of steps for circuit fineness—a crude “steel refining” approach. The trade-off is yield pressure and higher cost per chip, but it gets what it most wants: shippable products.

The production line has already started supplying DDR5 (8000 MT/s) and LPDDR5X memory modules aimed at AI demand, moving directly into the most supply-constrained market. This also explains why Zhejiang Xin Storage dares to keep expanding production even with higher costs: it’s betting that the demand side is willing to pay for the supply. And the efficiency—being able to build a cleanroom in 12 months—is the confidence that makes this brute-force approach work. While others are still laying foundations, running environmental assessments, and waiting for delivery schedules, Zhejiang Xin Storage is already running equipment for yield. The time gap itself becomes a capacity advantage.

The capacity arithmetic to close in on Micron?

Analysts point out that if the plan is completed on schedule, by end-2026 Zhejiang Xin Storage’s monthly DRAM output capacity will be close to 350k wafers. That compares with Micron’s estimated 375k wafers for the same period, a gap of less than 10%.

The NAND front line isn’t idle either. After Yangtze Memory fills capacity, total expected monthly production capacity will exceed 170k wafers. Behind this figure is the current approximately 150k wafers per month of output from two Wuhan wafer fabs, plus ongoing equipment installation and testing, and the third phase expansion expected to enter mass production in the second half of 2026.

Put the two lines together: DRAM is coming up close, while NAND quietly lifts. In China’s memory “two heroes,” one is making a loud push and the other is quietly grinding away. Different entry points, but both converge toward the same direction—trying to pull global storage-chip market share away, wafer by wafer, from the share originally carved up by Samsung, SK hynix, and Micron.

Where the money flows, the balance shifts that way

What this capacity expansion truly moves is the equipment supply chain. China’s semiconductor equipment domestic procurement rate—plainly, the proportion of domestic equipment in total procurement spending—currently is only about 23.2%, meaning more than 70% still relies on imports.

But even just Zhejiang Xin Storage’s procurement in 2026 is expected to bring nearly 10 billion yuan of new business to local suppliers. First, let the market “nurture” suppliers using demand; then, let suppliers in turn improve yields and raise efficiency. As a result, the proportion of import dependency may be gradually reduced year by year. Once the localization curve is triggered, it becomes hard to turn back.

At the same time, the Chinese government is promoting Zhejiang Xin Storage to transfer DRAM technology IP—i.e., process patents and engineering know-how—to Fujian Jinhua, Shengwei X?U, and Yangtze Memory’s subsidiary Newxin. In the short term, this is to ease domestic shortages; in the long run, it’s preparation to break into the EU and even the U.S. market. When technology is distributed across multiple plants, the effectiveness of sanctions targeting a single point is diluted.

Of course, this chess game is never being played by only China: Samsung Electronics, SK hynix, and Micron are simultaneously ramping up capital expenditures. SEMI (the International Semiconductor Industry Association) predicts the global semiconductor equipment market will jump from 116.6 billion USD in 2024 to 155.6 billion USD in 2027. Even the “pie” itself is getting bigger.

If you zoom out in time, the real damaging power of this capacity expansion isn’t in the short-term shipment volume. It’s in the scenario where Zhejiang Xin Storage brings yield and cost down to a level that can be scaled. Then the global memory pricing power will have a second voice, enough to make rivals think an extra second at the next quotation meeting—reconsidering how to manage inventory and the timing of capacity expansions.