An irreconcilable tension is brewing amid the Artificial Intelligence boom. The paths of demand and supply of compute—the processing power and computation resources needed to train, run, and operate AI models—are diverging. That imbalance, in our view, is creating attractive investment opportunities.
While digital demand for AI services can scale instantly, the physical infrastructure on which AI applications run (e.g., electricity, data centers, chips, memory, and cloud services) can take years permit, connect, and build (see Investing for the Compute Bottleneck). In our view, this dichotomy will grow over time due to two primary reasons:
Demand for compute is set to scale massively with the proliferation of AI agents: These agents perform multi-step tasks and may use 5–50x more compute tokens than a typical chatbot interaction. More activity at the application layer can translate into more demand for the compute infrastructure beneath it (Exhibit 1).
Supply faces bottlenecks across the entire compute pipeline. Data centers are projected to account for roughly half of US electricity-demand growth through 2030, while connecting new projects to the grid can take more than five years.1 Even after power is secured, developers still need permits, specialized labor, electrical equipment, cooling systems, GPUs, high-bandwidth memory, and networking.
In this context, we believe that sustained value will accrue to owners of scarce compute capacity. These include power producers, data center operators, and AI cloud providers that can already turn electricity into computation.
Key Takeaway: AI demand can scale much faster than the physical infrastructure supporting it. We believe this imbalance will favor owners of energized compute capacity and create attractive growth-oriented investment opportunities.
Exhibit 1: More demand at the top of the AI tech stack can create more demand at the bottom of the stack
1. Source: IEA report on energy and AI. Lawrence Berkeley National Laboratory.