What's being built

CXMT runs three 12-inch DRAM fabs today — two in Hefei, one in Beijing — for a combined ~300,000 wafers per month. On top of that:

→ A Shanghai fab due to complete in H2 2026, geared toward AI memory including HBM → Further Hefei expansion underway → A proposed second Beijing fab in Yizhuang, the same development zone as the existing plant → Combined, capacity heading beyond 600,000 wafers per month

One detail worth flagging: the reported state support under discussion for the Beijing site is around 60 million yuan (~$8.9 million). Against a 12-inch fab that costs billions, that figure tells you these talks are genuinely early — this is a term sheet, not a groundbreaking.

The number nobody is quoting

Here's the one that actually reframes the story.

CXMT is expected to exit 2026 at roughly 350,000 wafers per month. Micron's projected output for the same period is around 375,000.

Read that twice. On raw wafer volume, China's memory champion is about to draw level with one of the big three — a company that has been making DRAM since 1978.

Manufacturing scale is no longer the constraint. That problem has effectively been solved.

But volume is not value

So why only 7.67% of the market by revenue?

Because CXMT is producing conventional DRAM — DDR5, LPDDR5X — while Samsung, SK Hynix and Micron have shifted their best capacity into high-bandwidth memory for AI accelerators, where margins are in a different universe. SK Hynix is expected to take something like 70% of HBM4 for Nvidia's Rubin platform. That is where the money is.

The gap between 350,000 wafers and 7.67% of revenue is the entire Chinese semiconductor story in one ratio. China has solved scale. It has not yet solved value capture.

And the bridge between the two is precisely what export controls are designed to block. CXMT is targeting HBM production from the end of 2026 and has signalled ambitions toward 30% DRAM share by 2030 across a sixth mega-fab — plans that reporting consistently describes as bottlenecked by access to advanced chipmaking tools.

The gift the incumbents handed over

Here's the part I find genuinely interesting, and it cuts against the usual framing.

CXMT's opening was created by its competitors.

The 2026 memory supercycle is not a demand story alone — it's a reallocation story. Samsung, SK Hynix and Micron pulled capacity out of conventional memory to chase HBM margins. Micron exited consumer memory entirely. SK Hynix reported its capacity essentially sold out for the year. The result: DRAM prices rose 80–90% quarter-on-quarter in Q1, and Samsung flagged an ASP jump of 146%.

That vacuum sits exactly where CXMT's product line sits. Phones, PCs, cars, industrial equipment — enormous volume, now underserved. Reports suggest HP, Asus and Acer have begun limited use of CXMT parts.

The incumbents vacated the floor to chase the ceiling, and handed the floor to Beijing.

Where I'd be careful

Two things temper this.

First, memory is cyclical, and always has been. Every DRAM bust in history has the same cause: capacity commissioned at the top of a price spike, arriving after the peak. CXMT is adding 300,000 wafers per month of capacity into a market where prices nearly doubled in a quarter. That is the textbook setup, not a hedge against it.

Second, $488 billion on 7.67% share is a policy valuation, not an earnings valuation. The market is pricing in a China that closes the HBM gap despite tool restrictions. That may happen. It has not happened yet, and the multiple assumes it as though it already did.

My read: the scale achievement is real and underrated, the valuation is running well ahead of it, and the honest question is whether CXMT can climb the value chain before the cycle turns.

Question for you: if a company can build the wafers but not capture the margin, is it a competitor to Micron and SK Hynix — or a very large, very well-funded commodity supplier? I don't think the market has decided, and $488 billion says it's assumed the answer.

China's Most Valuable Listed Company Has 7.67% of Its Own Market

On July 27, CXMT closed its first day of trading in Shanghai up 466%.

At 49 yuan a share, ChangXin Memory Technologies was worth about 3.3 trillion yuan — roughly $488 billion — making it the most valuable company listed on China's A-share market. Bigger than the banks. Bigger than Moutai. The IPO itself raised 57.92 billion yuan ($8.6 billion), the largest in Asia this year, and the one-day pop was the biggest among the world's ten largest IPOs of 2026.

Now the other number: CXMT held 7.67% of the global DRAM market in 2025.

Address

Boston, Massachusets

Contacts

news@techproduit.com

Get your weekly newspaper