SK Hynix plans an ADR offering to raise 43 trillion won.

Huatai Securities says near-term focus should be on Hong Kong names that are deeply oversold, carry heavy short interest and show marginal stabilization or improvement in earnings expectations—notably innovative drugmakers and electric utilities—and on high-dividend defensive banks. Innovative-drug fundamentals are still improving; global style rebalancing has triggered catch-up moves and short-covering may continue. Prioritise CXO names with positive guidance and leading drugmakers with strong BD potential. For the AI hardware chain, further near-term increases offer limited risk/reward; the next critical window is the US July earnings season, so adjust exposure according to unrealized gains. AI software is more a source of stock-specific alpha tied to model leadership. On a medium-term view, maintain a balanced allocation between sectors showing accelerating demand and undervalued reversal candidates.

未来数据中心租赁规模超过8500亿美元，意味着AI算力扩张已经从“资本支出驱动”阶段，进入“长期合同锁定”阶段。本质上，科技公司不再是按需扩容，而是在提前锁定未来3–7年的算力供给。 从结构变化看，增量集中在2025–2026年加速兑现：单季度新增承诺同比+204%，环比+31%，说明这一轮扩张不是线性增长，而是明显的“前置集中投放”。这类行为通常对应两个阶段特征：一是AI训练与推理需求已经可见规模化拐点，二是供应端（数据中心、电力、芯片）开始成为约束变量。 公司结构上，Meta与Microsoft的新增加速，反映的是通用模型与企业AI部署的双线扩张；Oracle的领先，则更多来自其作为基础设施中介方，在OpenAI等大模型客户外溢需求中的承接角色。 对市场而言，这种结构意味着AI叙事正在从“增长预期驱动估值”转向“合同与产能锁定驱动现金流预期”。风险点也同步变化：不再是需求是否存在，而是电力、算力交付、资本回收周期是否匹配。