The memory industry cycle typically unfolds across five distinct phases: a downturn marked by oversupply and inventory digestion where share prices bottom before fundamentals; an early recovery characterized by spot prices rising ahead of contract prices and share prices starting to rebound; a mid-cycle boom where revenue and pricing momentum accelerate, driving simultaneous expansion in earnings and valuations and delivering the strongest share price gains; a late-stage period where pricing momentum decelerates or turns negative while revenue and profits remain elevated, valuations begin to compress, and share prices trade sideways; and a cycle peak where inventory cycles lengthen and share prices decline before fundamentals deteriorate. The defining feature of the late stage is a combination of inexpensive price-to-earnings ratios, upward revenue revisions, yet share prices that have stalled or started to fall.

From a valuation perspective, the market is clearly pricing the sector as if it has already entered the late stage. SK Hynix trades at a forward PE of approximately 3.6 times (Bloomberg, 14 August 2026), while Samsung Electronics commands a forward PE of roughly 4.3 times (Bloomberg, 14 August 2026). This compression in PE multiples is precisely the hallmark of a memory market transitioning into its late cycle. The valuations being assigned by the market have already factored in expectations that the cycle is approaching its peak.

Fundamentally, however, the memory sector remains positioned somewhere between mid-cycle prosperity and the late stage. While third-quarter pricing momentum is indeed decelerating (DRAM price increases have moderated from +65% in Q2 to +17%, and NAND from +60% to +20%), the demand driver behind this cycle, namely AI data centers, is proving more durable than the consumer electronics boom of the previous cycle. Cloud capital expenditure continues to rise, and both OpenAI and ByteDance have resumed training their foundational large language models, which directly fuels demand for HBM and enterprise-grade SSDs.

More critically, memory manufacturers themselves are holding inventory at extremely low levels, with DRAM inventory at just 2-3 weeks and NAND at 4-5 weeks, compared to the typical 7-9 week inventory cycle observed in late-stage memory markets (TrendForce). Singapore's sovereign wealth fund, Temasek, has recently signalled plans to make direct investments in Samsung and SK Hynix, judging memory to be the most undervalued segment within the AI supply chain. Meanwhile, SanDisk presented a financial framework at its recent Investor Day projecting mid-to-high single-digit revenue growth through 2028-2030 with gross margins around 80%, while also locking in over half of FY2027 shipments and approximately two-thirds of FY2028 shipments through NBM long-term supply agreements. Both industrial capital and leading companies themselves believe the current late-cycle pricing is unjustified.

Against this backdrop, E Fund AI (03489) and E Fund Asia Semiconductor (03486) offer two distinct approaches to capitalizing on this opportunity. E Fund AI (03489) provides exposure to leading AI computing and application companies across China and the United States, with a balanced allocation spanning both hardware and cloud providers. E Fund Asia Semiconductor (03486) focuses on the full Asian semiconductor supply chain, including SK Hynix and Taiwan Semiconductor. During this phase where the market applies a late-cycle discount while fundamentals remain elevated, these two products offer investors the means to position for both sustained memory strength and valuation recovery, from the perspective of the entire AI ecosystem or pure hardware exposure respectively.