Structural Derating Ahead for Greater China DDR4 Memory Names Despite Near-Term Strength

Core Conclusion

Greater China legacy memory stocks with DDR4 exposure face structural multiple compression. Software- and system-level optimizations are beginning to erode long-term DRAM demand, while new capacity, particularly from CXMT, threatens future pricing. Long-term agreements (LTAs) provide a price floor but also cap upside, shifting the investment thesis from cyclical growth to dividend/utility-like returns. In contrast, pure-play legacy NAND vendors are far less affected and offer relative preference.

The Market's Mispricing

The market is extrapolating the current tight DDR4 supply and price cycle linearly, overlooking two converging headwinds: the secular demand impact from system-level cost optimizations and the impending supply wave from new capacity. Furthermore, optimism surrounding LTAs as a protective shield underestimates their role in capping future pricing power and earnings growth.

Evidence Chain

Software and system optimizations are driving a structural reduction in DRAM content per system, with cost being the primary catalyst. System houses and NAND vendors are actively deploying solutions like HLC (Longsys), iGPU-based designs (Phison), and ZNS (T-Head) to lower DRAM requirements. The driving force is a massive cost differential, with DRAM per GB costing approximately 50x that of NAND in Q2, forcing customers to seek efficiency gains. The investment implication is a direct threat to the sustainability of the industry's current high margins, as pricing must eventually converge with a lower demand trajectory.

The peak of the DDR4 supply crunch is likely in sight, with new capacity set to alter the balance. While current fulfillment rates are around 80%, indicating near-term tightness, supply is forecast to increase. Our industry model points to price declines starting in 2H27, driven by new capacity from CXMT and Nanya Tech's own expansions coming online. This forecasts a clear turning point, with pricing momentum likely to stall and then reverse within the forecast horizon, pressuring earnings growth.

Long-Term Agreements provide a downside cushion but definitively limit upside, fundamentally altering the investment case. Historical precedent, such as customers honoring fixed-price MCU LTAs post-pandemic, suggests DDR4 LTAs may offer a multi-year price floor. However, the report explicitly states that even if DDR4 pricing holds for three years, earnings upside is "capped." This transforms the investment narrative, as memory players may transition into dividend plays, warranting a lower, utility-like earnings multiple rather than a cyclical growth premium.

A stark divergence is emerging between the fundamentals of legacy DRAM and legacy NAND. The analysis "repeatedly" emphasizes a "contrarian preference for legacy NAND vs. DDR4." It notes that pure legacy NAND vendors like Macronix are "much less impacted, earnings even stronger" in the coming quarters. This creates a clear relative trade: investors should reduce exposure to old DRAM plays and rotate toward traditional NAND and niche memory names which face less systemic pressure.

Key Risks & Divergences

• Demand Erosion Acceleration: Widespread, faster-than-expected adoption of software optimization techniques (e.g., KV cache compression, TurboQuant) could accelerate the decline in per-system DRAM demand.
• Supply Surge: Faster-than-anticipated capacity ramp from Chinese domestic manufacturers, notably CXMT, could flood the DDR4 market sooner, exacerbating price pressures.

Valuation & Trade Implications

We have systematically cut target prices for all covered DRAM-exposed names—Winbond (2344.TW) to NT$100, Powerchip (6770.TW) to NT$71, GigaDevice (603986.SS) to Rmb301, AP Memory (6531.TW) to NT$607, and Nanya Tech (2408.TW) to NT$278—primarily through valuation multiple compression. The appropriate valuation framework is shifting from cyclical growth to one reflecting capped earnings and dividend returns. Action: Avoid or underweight old DRAM. selectively favor legacy NAND (e.g., Macronix) and niche memory players with less exposure to these structural headwinds.

Appendix: Key Company Target Price Revisions