South Korea’s central bank projects that the country’s monthly memory chip output will rise by roughly 600,000 wafers by 2028, against a projected increase of about 300,000 wafers for China’s CXMT over the same period. On the face of it, Korea widens its lead. The number needs unpacking before it means anything.
Wafers Are Not Bits
A wafer start is an input, and memory is sold in bits. How many bits come off a wafer depends on the node, the die size, and the yield, and the spread between a leading-edge DRAM line and a trailing one is large enough to swamp a headline capacity figure. Two fabs running the same wafer volume can differ by a factor of two or more in shipped bits.
The gap widens further with product mix. Wafers committed to high-bandwidth memory consume capacity at a punishing rate, because HBM stacks consume multiple dies per unit sold and the packaging step adds its own yield loss. A Korean fab tilting toward HBM can raise wafer starts while adding very little to the commodity DRAM supply that sets spot prices. That has been the defining feature of this cycle and it is why conventional memory has stayed tight through a capacity expansion.
What CXMT’s 300,000 Is Really Measuring
The Chinese figure is the more informative of the two, because it is constrained by something other than demand. CXMT can add cleanroom space quickly. What it cannot do quickly is buy the lithography, deposition, etch, and metrology tools that fill it, or replace the service contracts that keep them running at yield. Every wafer of Chinese capacity added under those conditions is evidence about how much of the toolchain has been substituted, licensed, or acquired through channels that were supposed to be closed.
So the useful reading is inverted. If CXMT hits 300,000, the question is not whether China is catching Korea in volume. It is which pieces of equipment made those wafers possible, and that is answered in customs data, equipment vendor revenue by region, and the service and spare parts flow rather than in output forecasts.
Indicators Worth Tracking
- Equipment maker revenue attributed to China, quarter by quarter, and how much of it is trailing-edge versus advanced.
- The DDR5 share of CXMT’s output, which is the real capability marker, since commodity DDR4 volume proves capacity and nothing else.
- Any credible sign of Chinese HBM at volume, which would mark the transition from supply-side nuisance to strategic competitor.
- Korean capex guidance revisions, which move faster than central bank projections and reveal what the producers actually believe about 2028 demand.
A projection out to 2028 in an industry with an eighteen-month planning horizon is a statement about intent, not a forecast. Read it as one.