Structural Divergence in Legacy Memory: MLC NAND/NOR Over DDR4, with EMIB as a New Growth Driver

Core Conclusion

Investment opportunities within the Greater China legacy memory segment are diverging. We favor MLC NAND and NOR Flash over DDR4 due to accelerating pricing power and structural supply constraints. Separately, Intel's EMIB packaging technology creates a new, material growth vector for companies like AP Memory (6531.TW), offering silicon capacitors. Our selective Overweight stance is on Macronix (2337.TW), AP Memory, Powerchip (PSMC, 6770.TW), and GigaDevice (603986.SS), while we are Equal-Weight on Nanya (MN) and Winbond (2344.TW).

What the Market May Be Missing

The market may be underestimating the sustainability of the MLC NAND/NOR supply crunch, driven by major player exits and capacity reallocation. Furthermore, the potential revenue contribution from Intel's EMIB ecosystem for AP Memory appears underappreciated, representing a high-value, non-commoditized growth avenue distinct from the broader memory cycle.

Evidence Chain

DDR4 cycle momentum is peaking, limiting equity upside. DDR4 8Gb pricing is expected to exceed USD20 in Q2 2026, representing a 60% quarterly increase. However, the pace of hikes is set to decelerate in the second half of the year due to factors like KV cache compression, server optimization, and the start of LTA discussions. The incremental supply from CXMT also looms. While near-term profitability for DDR4 vendors like Nanya and Winbond will be high, the decelerating pricing momentum caps share price appreciation.

MLC NAND and NOR Flash pricing power is accelerating structurally. Supply is in extreme tightness following Samsung's and Kioxia's phased exits from the MLC NAND market. Macronix is reallocating NOR capacity to support MLC NAND, concurrently reducing NOR supply. This dual constraint is expected to drive Q2 price increases of up to 100% for both memory types, with high double-digit hikes sustained into H2 2026. This provides a more durable earnings tailwind for pure-play vendors like Macronix.

Intel's EMIB-T creates a high-value niche for AP Memory. AP Memory is positioned as the key Taiwanese supplier of silicon capacitors (eDTC) for Intel's EMIB-T advanced packaging. With high-power AI ASICs (e.g., TPU, Trainium) requiring superior power delivery, content per chip could approach USD100. Industry checks indicate Google and AWS are engaged for mechanical testing vehicles, with potential production for 2nm ASICs in 2027/2028. This could drive incremental revenue exceeding 50% versus 2026e levels, with customer prepayments being a key near-term catalyst.

Company-specific drivers underscore a selective approach. Macronix leads the NAND/NOR pricing upcycle. PSMC's value lies in spec migration within legacy DRAM. GigaDevice benefits from NOR pricing and an upcoming ramp in LPDDR4/DDR4 via CXMT's Gen 4 platform in 2027. In contrast, Nanya and Winbond are more levered to the decelerating DDR4 cycle.

Key Risks and Divergences

• Cycle Reversal: Memory prices are cyclical; a sharper-than-expected downturn in end-demand could rapidly reverse pricing gains.
• Execution Risk: Intel's yield and execution on EMIB-T for external clients like Google/AWS is unproven at scale, posing a risk to AP Memory's growth timeline.
• Competitive Response: Accelerated pricing may attract renewed capacity additions or slow the pace of demand.
• Macro Demand: Weaker consumer IoT or PC demand could impact legacy memory volumes.

Valuation and Trade Implications

Our preference hierarchy is Macronix > AP Memory > GigaDevice > PSMC. We raise AP Memory's price target to NT$777 (implying 48x 2026e P/E) and GigaDevice's to Rmb348 (43x 2026e P/E) to reflect the improved pricing outlook and new growth drivers. Macronix offers the most direct exposure to the accelerating flash upcycle. Investors should avoid a blanket approach to legacy memory and focus on these divergent sub-segment dynamics and technological niches.

Appendix: Key Stock Summary