汇丰与投资者团队在五天内走访了硅谷27家科技公司并总结核心要点。报告指出，当前AI驱动的供应短缺是全面性的而非单一环节瓶颈，需求持续超过供给，主要瓶颈集中在数据中心容量与电力、基板/代工晶圆/先进封装、光组件以及存储四大领域；产能瓶颈预计至少持续至2027年，甚至更久。AI相关需求未见放缓，需求正从训练向推理、智能体、企业、主权AI、存储及物理AI扩散；受成本、灵活性与主权考量影响，市场焦点转向开源模型，开源与闭源模型预计共存。AI变现仍处早期阶段，软件公司尚无法量化变现路径，收入贡献相对于使用量与投资仍较小。在供给受限背景下，半导体与科技公司估值倍数预计短期受限，进一步重估主要取决于自身盈利超预期表现。
We held meetings with 27 tech companies in Silicon Valley, along with a group of investors, across five days. There is a broad, AI-led supply crunch rather than a single isolated constraint, with demand continuing to run ahead of supply across much of the semiconductor and data center ecosystem. The major bottlenecks span data center capacity and power; substrates, foundry wafers, and advanced packaging; optical components; and memory, and are expected to persist through 2027, with a possibility of extending beyond that. There are no signs of AI-related demand slowing down, with demand broadening from training towards inference, agents, enterprise, sovereign AI, storage, and physical AI. The focus is also shifting towards open-source models due to cost, flexibility, and sovereignty, although both open-source and proprietary models are expected to co-exist. AI monetization remains a key debate, with software companies unable to quantify the monetization story as revenue contribution rem