AI Investment Theses

China AI infrastructure demand will increasingly route toward domestic accelerators and supporting semiconductor infrastructure as export controls, procurement policy, and sovereign compute priorities make foreign GPU availability less reliable.

Hyperscalers, sovereign buyers, and AI-native platforms are likely to sustain a multi-year infrastructure buildout across accelerators, networking, foundry, memory, and datacenter physical infrastructure.

HBM supply is likely to remain structurally tight because AI accelerator roadmaps keep increasing memory bandwidth requirements faster than high-yield advanced memory and packaging capacity can normalize.

Inference workloads should become a larger and more persistent source of AI infrastructure demand as consumer AI, enterprise copilots, search, agents, and vertical applications move from pilots to repeated production usage.

Power availability, cooling density, and grid interconnect timelines are becoming binding constraints on AI datacenter deployment, shifting incremental value toward electrical and thermal infrastructure suppliers.