Huxiu identifies three AI tracks that could generate excess returns

Three tracks, one theme: scale, scarcity, and specialization

It has been reported that Chinese business outlet Huxiu has singled out three AI investment tracks likely to deliver excess returns: foundational models (large language and multimodal models), industry‑vertical AI applications, and the underlying compute and tooling infrastructure. The logic is straightforward. Foundation models offer scale and broad product leverage. Vertical AI can monetize specialized data and workflows. Infrastructure captures the scarce compute engines and deployment pipelines.

Who stands to gain — and why now?

Major Chinese firms are already positioning themselves. Baidu (百度), Alibaba (阿里巴巴) and Tencent (腾讯) have all invested heavily in large models and cloud stacks. Huawei (华为) and domestic chipmakers are racing to plug gaps left by Western export controls. It has been reported that these dynamics—massive upfront model costs, entrenched data moats, and constrained access to advanced chips—create high barriers to entry that can amplify returns for winners.

Geopolitics matters

Western export controls and broader US‑China tech tensions are central context for Western readers unfamiliar with China’s tech landscape. Sanctions and trade policy have reportedly accelerated on‑shore substitution: domestic GPUs, custom accelerators, and localized data ecosystems. That changes the risk/reward calculation for investors and startups alike. Does limited access to Western silicon raise costs? Yes. Does it also create captive markets for Chinese suppliers? Potentially.

Beware the hype

Huxiu’s take is bullish but not unconditional. High capital intensity, fast‑moving regulation, and technological uncertainty mean outcomes will be uneven. Reportedly, the greatest practical edge will go to firms that combine model scale, proprietary vertical data, and tight control of deployment infrastructure. Winners may emerge — but will they be the usual big cloud names, or a new generation of specialized players? Time, policy and hardware will decide.