Sources: CXMT plans to start HBM3E production despite yields of just 20% to 30%; its HBM3E performance is closer to that of SK Hynix's previous-gen HBM3 chips
China's CXMT plans to start production of HBM3E, an advanced type of high-bandwidth memory chip used with AI processors …
Context & Ripple Effects
CXMT's HBM effort has advanced from the early-stage HBM work reported in 2024 to small-quantity HBM3E output in September 2026. The planned production start puts a concrete yield and performance constraint on that progression.
The gap is visible against SK Hynix, which began mass production of 12-layer HBM3E in 2024. CXMT's separate preparation to enter NAND also frames HBM as part of a broader attempt to compete across memory categories.
First-order effects
- A 20% to 30% yield limits CXMT's usable HBM3E output and raises the manufacturing burden as it begins production.
- CXMT's HBM3E performance being closer to SK Hynix's prior HBM3 generation preserves SK Hynix's lead in the newer HBM3E product tier.
Second-order effects
- CXMT's planned 2027 expansion becomes chiefly a yield-improvement exercise: capacity growth will not translate proportionately into available HBM3E chips unless the production process improves.
- SK Hynix gains a clearer distinction between an emerging Chinese HBM supplier and its mass-produced HBM3E offering, even as CXMT establishes itself as a future competitor.
Third-order effects
- If CXMT improves yields, its moves in HBM, server DRAM and NAND would shift its challenge to established memory suppliers from a single product line toward a broader memory portfolio.
- The episode underscores that AI-memory competition is determined not only by reaching a chip generation, but by the ability to manufacture it at sufficient yield and performance.
The trend: AI-memory supply is broadening beyond incumbent leaders, but manufacturing yield is the gatekeeper between a technical milestone and competitive volume.