The hardware behind the future


Shanghai Surprise: CXMT and YMTC Aim Two New Fabs at the DRAM Shortage

China’s DRAM and NAND champions are both building for 2027, and one of them is entering a business it has never manufactured in.

Shanghai Surprise: CXMT and YMTC Aim Two New Fabs at the DRAM Shortage

ChangXin Memory Technologies is building a Shanghai DRAM fab two to three times the capacity of its Hefei site, with equipment installation slated for the second half of 2026 and volume production targeted for 2027.

Nikkei Asia reports the plan, and Electronics Weekly and Tom's Hardware add the wafer numbers. Hefei runs at roughly 180,000 wafers per month and Beijing at about 120,000, so on those figures Shanghai lands somewhere between 360,000 and 540,000 wafers per month of new DRAM capacity in a single location.

That number deserves the usual discount. Equipment installation starting in the second half of 2026 is not tools qualified, yield stabilised and product shipping, and CXMT has published no phasing for how any of it ramps.

Process node is the constraint nobody involved is advertising. Electronics Weekly puts Hefei's most advanced node at 16nm, roughly ten years behind Samsung, SK hynix and Micron, and its most advanced product at LPDDR4 with LPDDR5 in preparation. Capacity at a trailing node does not chase the same sockets as leading-edge DDR5; it chases the commodity tail the incumbents are walking away from.

There is a genuine opening there. The big three have pushed wafer starts and capital toward HBM, and Tom's Hardware reports that Samsung and SK hynix have both told customers supply tightness is likely to run into 2027. Conventional DRAM is the segment being vacated, and it is the only segment CXMT can serve at volume today.

Yole puts CXMT at 11.1 percent of worldwide installed DRAM capacity, rising to 13.9 percent by 2027. Installed capacity is neither bit share nor revenue share, and at 16nm the distance between those three numbers is wide. Most of the output goes to domestic customers in any case.

The more interesting move is Yangtze Memory Technologies. YMTC is a NAND house, and its third Wuhan fab is planned as a roughly even split between NAND and DRAM, a business it started designing for two years ago and is only now ready to manufacture. Wuhan fabs one and two total 140,000 to 150,000 wafers per month with a target of 200,000 during 2026, and fab three is expected to add 30,000 initially, eventually rising to 100,000.

A NAND vendor standing up DRAM from nothing is a harder problem than pouring a shell. Cell architecture differs, test and binning differ, and server DRAM qualification cycles at a new supplier run long enough to push real revenue well past the day the fab opens. Electronics Weekly also reports that YMTC and a Wuhan assembly company intend to build HBM, which is another step again and sits beyond anything either firm has qualified.

Bootnote: none of the wafer figures here come from either company. They come from Nikkei Asia's sourcing, relayed through Electronics Weekly and Tom's Hardware, and neither CXMT nor YMTC publishes capacity. Yole's installed-capacity estimate is the only number with a named methodology behind it, and even that measures what the tools could produce rather than what shipped.

Leave a Reply