Northbound Funds Net Bought Over HK$1.7 Billion in SMIC, While Hua Hong Semi Saw Net Selling
Nashnova编辑部
On August 17, northbound capital net-bought HK$3.077 billion in total. SMIC led with HK$1.774 billion in net buying while Hua Hong faced HK$778 million in net selling — opposite flows in the same foundry sector, signaling a market split on earnings delivery.
Why are funds flowing in opposite directions for SMIC and Hua Hong?
SMIC's Q2 results beat expectations across the board; Q3 gross-margin guidance points higher. Multiple brokerages upgraded the stock. This means → the HK$1.774 billion inflow is a bet on "earnings delivered plus improving guidance."
Hua Hong posted Q2 net profit of US$38.6 million, just below the US$39.4 million consensus. Q3 revenue guidance also fell short. In plain terms = earnings missed by a sliver and the outlook underwhelmed — enough for profit-taking, with HK$778 million net-sold.
This reflects a northbound stance that is not a blanket bet on domestic foundries — capital is pricing each name strictly on results and forward guidance.
Why are AI-model companies also seeing heavy northbound buying?
Zhipu (02513) drew HK$1.672 billion in net buying; MINIMAX-W (00100) drew HK$306 million.
The backdrop: DeepSeek sharply raised API prices on August 8. Morgan Stanley read this as a positive signal of improving pricing discipline. This means → AI firms are shifting from "burn cash, grab users" toward "monetize model capability," with signs the price war is cooling.
Morgan Stanley flagged three structural shifts happening in parallel: pricing turning rational, open-source licensing tightening, and model parameter counts leaping. In plain terms = the industry is getting more expensive, more closed, and bigger — all pointing to stronger pricing power for leading players.
Tencent and Alibaba — what logic is northbound money buying?
Tencent (00700) saw HK$533 million net-bought. Daiwa noted that near-term capex may stay elevated, but management views AI spending as justified by attractive long-term returns; computing power can already be monetized through resale. This means → Tencent treats AI as a long-cycle investment and is already working to make its compute spend self-funding.
Alibaba-W (09988) saw HK$422 million net-bought. The company open-sourced its Qwen3.8 model series on August 14; the new Qwen3.8-27B has only 27 billion parameters yet outperforms the previous-generation Qwen3.7-Plus. In plain terms = a smaller model delivering stronger results — a clear efficiency signal.
On the net-sell side, what is the market worried about?
YOFC (06869) saw HK$418 million net-sold. Jefferies estimates that 19,400 tonnes of new optical-fiber preform capacity has been announced globally for 2026–2028. Supply may still be tight in 2026, but 2027–2028 is the risk window when capacity hits the market all at once.
Nomura previously warned that if new entrants successfully develop high-end products and expand, incumbents' margins will come under pressure. This means → capital is pricing in the possibility of oversupply ahead of time.
Kingboard Laminates (01888) saw HK$439 million net-sold; GigaDevice (03986) saw HK$297 million net-sold. Cambridge Technology (06166) bucked the trend with HK$191 million net-bought.
What does the northbound reshuffle within semis tell us?
Same day, same sector: heavy buying of SMIC, simultaneous selling of Hua Hong. In plain terms = northbound money is not betting on "domestic substitution" as a theme — it is betting on "who delivers earnings first."
This reflects a screening standard that has shifted from concept to execution — beat expectations and you get capital; miss guidance and you get sold.
Whether this reshuffle signals an emerging consensus on the earnings-delivery thesis for leading domestic foundries will be the key observation point going forward.
Content is for reference only, not financial advice.