AI Agent Demand Reshapes the Compute Trade Across Asia
TREE NEWS reports: Mainland and Hong Kong equity markets opened higher on Tuesday, with the AI hardware and application complex leading gains. The ChiNext board rose as much as 2% before paring gains, the CSI 50 pulled back from session highs, and Hong Kong’s Hang Seng Tech Index climbed roughly 1%. Large-cap Chinese internet names were the standout: Tencent surged more than 6% intraday, while Alibaba rose nearly 3% after touching a 5% gain. Domestic government bond futures edged higher across the curve, and most commodity futures advanced, led by polysilicon, glass and international copper.
What Happened
Two catalysts converged. First, Meta’s consumer AI agent app Muse has dominated US iPhone free-app download charts since launch, logging roughly 730,000 downloads in its first five days and more than 2.8 million within twelve days, surpassing the early trajectory of competing assistants. Meta shares jumped over 11% in the prior US session, and at least one major bank raised its price target to $796. The market is repricing server CPU and inference demand for an “agent era” in which AI executes tasks — shopping, ride-hailing, scheduling — rather than merely answering prompts.
Second, Alibaba’s cloud unit used its annual Yunqi conference to lay out aggressive AI infrastructure targets. CEO Eddie Wu said customer AI demand is “very strong” and that the company aims to operate more than 20GW of global data center capacity by 2032. Its chip design arm unveiled the Zhenwu V900, claiming three times the compute of the prior M890 generation, with sharply higher annual shipment plans. Tencent separately released an upgraded AI image model, Hy Image 3.5 Preview, now being integrated into products including Yuanbao and its video and design tools — a signal that Tencent is pushing beyond text into multimodal applications for its WeChat ecosystem of over one billion users.
Market Implications
- Compute silicon and CPU names: Hygon Information, Moore Threads, Montage Technology, GigaDevice and CXMT-linked storage plays led gains. The narrative has shifted from training-centric GPU demand to inference and agent orchestration, which is more CPU- and memory-intensive — a favorable mix for domestic chip designers.
- Storage: Sell-side estimates put 2026 second-quarter capital spending by the four largest North American cloud providers up 87% year on year, with storage chips potentially accounting for roughly half of a projected $1.4 trillion in 2027 capex among the top fourteen cloud operators. DRAM contract prices are seen rising 20–30% quarter on quarter as AI-driven structural shortages compound a cyclical recovery.
- Hong Kong internet: Tencent and Alibaba are being repriced as AI infrastructure and application owners rather than pure consumer platforms. Alibaba’s multi-year data center roadmap gives the market a capital-expenditure anchor; Tencent’s model releases give it optionality across its own ecosystem.
- Bonds and commodities: Firmer government bond futures suggest investors are not yet pricing an inflation shock from the AI capex cycle. Copper and polysilicon strength reflects expected power and hardware buildout demand, while shipping equities weakened on freight-rate concerns.
Key Takeaways for Investors
The trade is broadening from a single GPU story into a full-stack theme: CPUs, HBM and conventional memory, advanced packaging, PCB, optical interconnect and AI applications. Investors should watch three things: whether Meta’s Muse engagement holds after the initial download spike, whether Alibaba’s 20GW target translates into near-term orders for domestic suppliers, and whether DRAM pricing confirms the shortage thesis. The risk is that agent-era demand estimates are extrapolated from a few weeks of app-store data; the opportunity is that memory and CPU suppliers have far more room for earnings revisions than the crowded GPU leaders.
