인사이트

Investment opportunities across the full AI training and inference infrastructure value chain, from semiconductors to power, cooling, and networking. Analyzing 12 sub-themes spanning GPU/HBM to transformers and cooling from a unified perspective.

The power crisis triggered by AI data centers. Data center power demand is set to more than double by 2030, creating extreme bottlenecks in nuclear, gas turbines, and grid infrastructure. Full energy value chain analysis spanning generation, transmission/distribution, power distribution, and cooling.

The real bottleneck in AI clusters is not compute but data movement. After GPU/memory, the next investment phase is 'Interconnect' emerging as the bottleneck that determines utilization and ROI. I/O (SerDes) power, heat, and physical area constraints are growing rapidly, and the solution is the evolution of optical architecture from Pluggable to LPO to CPO. The investment strategy has a 3-layer structure: (1) Core (platform definers: switch ASIC/foundry), (2) Tollgate (connectors/lasers/test equipment), (3) Transition (800G/1.6T volumes, AEC/LPO).