HSBC: AI Investors Should Focus on Korean Memory, Taiwan Semiconductors, and China's Computing Infrastructure
nashnova research
HSBC's strategy team says AI investors should pivot to Korean memory chips, Taiwan semiconductors, and mainland China's compute infrastructure — three pockets where demand is strong, capacity is tight, and valuations haven't stretched the way US AI names have.
Why does HSBC say it's time to look beyond US AI stocks?
The core call: the next leg of AI returns sits in emerging-market tech segments with solid demand but uncrowded valuations.
This means → HSBC sees parts of the US AI trade as already expensive; the risk-reward is shifting to Asia.
The report is led by Alastair Pinder, HSBC's chief EM and global equity strategist, and targets three lanes: Korean memory, Taiwan semis, mainland China compute infrastructure.
Korean memory just sold off — is it still a buy?
Korean memory stocks pulled back sharply, but HSBC says fundamentals haven't deteriorated: capex-driven demand persists and shareholder returns are rising.
HSBC sees "signs that the bulk of selling pressure has abated." This means → lower volatility → lower cost of equity → a floor under sector valuations.
In plain terms = the weak hands have already exited; the remaining holder base is stickier, which actually opens room for a re-rating.
What is the "next phase" for Taiwan semiconductors?
HSBC expects Taiwan's AI trade to broaden from GPUs — graphics processors that power AI training — into custom ASICs (application-specific integrated circuits designed for particular workloads).
The driver: hyperscalers are increasingly adopting custom chips rather than buying only off-the-shelf GPUs.
HSBC notes that "sustained capacity constraints should support both pricing and utilization." This means → even as packaging investment accelerates, demand outpaces supply in the near term, keeping Taiwanese fabs' pricing power intact.
How does the mainland China thesis differ from Korea and Taiwan?
HSBC favors domestically focused AI semiconductor and hardware stacks, citing a triple tailwind: policy support + import substitution + inference-workload growth.
Inference — the stage where a deployed AI model actually runs and serves user requests — is scaling fast, lifting compute demand, while restrictions on foreign GPUs accelerate the shift to domestic alternatives.
This reflects a distinct investment logic: mainland earnings are not directly tied to US hyperscaler capex cycles, offering diversification away from the Korea-Taiwan exposure.
Can less efficient domestic chips actually be a tailwind?
HSBC flags a counterintuitive point: mainland China's homegrown chips are less efficient, but that may actually expand demand across packaging, networking, and power infrastructure.
In plain terms = weaker chips mean you need more of them — more packaging, more networking, more power — so the upstream and downstream volumes actually grow.
HSBC also cautions that whether this logic continues to hold in coming quarters will be the key test of the allocation call.
市场有风险，内容仅供研究参考，不构成投资建议。