Beyond the Model Builders
As the initial frenzy around Large Language Models (LLMs) settles into a mature industrial phase, the “easy money” in direct AI application developers has likely been made. The next frontier for the quantitative investor lies in the physical constraints of compute.
The Energy Bottleneck
Training a frontier model in 2025 requires gigawatts of power. Data centers are competing with cities for grid access.
- Nuclear Micro-reactors: We are seeing a surge in valuation for modular nuclear solutions.
- Grid Modernization: Copper and high-voltage transmission equipment manufacturers are the silent winners of the AI boom.
Specialized Compute
While general-purpose GPUs started the revolution, inference-specialized ASICs (Application-Specific Integrated Circuits) are scaling it.
“During a gold rush, sell shovels. During an AI revolution, sell cooling systems and transmission lines.”
Outlook: We remain overweight on utilities with heavy renewable exposure and semiconductor fabrication equipment, while reducing exposure to pure-play software wrappers.