SK Hynix Sees AI Memory Demand Visibility Through 2029

SK Hynix's demand visibility for DRAM and HBM capacity now extends to 2029, while Samsung expects memory supply tightness to persist through 2028 — those were the headline numbers from a Cathay Securities Investment Trust forum in Taipei on the 4th, held during "Taiwan Week" with Daiwa-Cathay Capital Markets.

The event, titled "From Silicon Valley to Silicon Island: The Investment Map in the AI Era," brought together Daiwa Securities' South Korean technology expert Sung Kyu Kim, Cathay's Taiwan-Korea technology analyst Li Han-lin, Taiwan semiconductor analyst Huang Chiu-wen and fund manager Liang En-yi. Cathay CIO Cheng Li-cheng, Daiwa Managing Executive Officer Naoki Suzuki and Daiwa-Cathay President Takeshi Mitsuda opened the sessions.

Why is memory no longer a supporting player?

Sung Kyu Kim argued that AI workloads are shifting from model training to inference, agentic AI and physical AI. That changes memory's role: it now directly gates system performance rather than playing second fiddle to compute.

As AI chip specifications climb, demand for both HBM capacity and transmission bandwidth rises in tandem. Future AI agents will handle longer conversation contexts and run more complex tasks, which should further lift DRAM and NAND demand.

The four major cloud service providers — Microsoft, Amazon, Meta and Google — keep ramping AI capital expenditure, and that capex is the main engine behind the memory upcycle, Kim said.

Can supply actually keep pace?

Probably not soon, according to Li Han-lin. Process miniaturization is approaching physical limits and Moore's Law is slowing. DRAM process upgrades have become markedly harder, and the growing complexity of HBM and high-layer NAND stacking means capacity expansion now demands more capital, more equipment and longer manufacturing lead times.

Li also sketched the division of labor across the strait: South Korea leads in memory — DRAM and HBM — while Taiwan controls advanced process technology, advanced packaging and the AI server supply chain. The two industries are highly complementary and together form an indispensable piece of global AI infrastructure.

Where does demand spill next?

Huang Chiu-wen traced how AI demand has widened from GPUs to CPUs, memory, communications and power systems as workloads move beyond the data center. Three areas stand out:

• Advanced packaging. As GPU and custom ASIC specs upgrade, integration and packaging complexity rise. TSMC (2330.TW) keeps expanding CoWoS capacity, which should pull through demand for packaging and testing equipment, IC substrates, facility engineering and semiconductor materials.
• Optical communications. With AI clusters scaling and data transmission moving toward 800G and 1.6T, optical links and co-packaged optics (CPO) are drawing market attention.
• Power management. Rising rack power consumption is pushing supply architecture toward high-voltage direct current, opening specification-upgrade opportunities for power equipment makers.

Huang added a caveat practitioners should hear: industry growth doesn't guarantee equal gains for every supplier. Whether spec upgrades lift unit prices and margins — and whether a company has real technical barriers, mass production capability and cost control — decides who converts the opportunity into profit.

How should portfolios rotate?

Fund manager Liang En-yi, who runs the Cathay Active Momentum High Dividend Fund (00400A) and the Cathay Taiwan High Dividend Fund, said the long-term AI trend is intact but the beneficiary list isn't fixed. As demand spreads from chips and servers into packaging, optics and power, allocation has to move with it.

His "one offense, one defense" approach works like this: 00400A combines growth momentum with dividend criteria, screening roughly 40 to 60 names from more than 1,900 listed and OTC companies, and adjusts holdings on fundamentals and industry trends. The Taiwan High Dividend Fund leans on earnings quality, dividend records and valuation, using income and long-term value as the anchor.

Cathay Securities Investment Trust said investors can access the Taiwan-Korea technology story through the Cathay Taiwan-Korea Technology ETF (00735), or weigh other allocations against their own objectives and risk tolerance — while staying mindful of semiconductor cyclicality, industry concentration and market volatility. The funds and ETFs were named at the forum only; fee rates and holdings are governed by each fund's prospectus.