Tech Disruptors: Lazard’s Bilicic on Power Demand in the AI Era

The market is likely underpricing how slowly the grid can respond relative to AI buildout. In the next 1-3 months, the first winners are not the consumer-facing AI names but the bottleneck owners: transmission contractors, switchgear suppliers, gas-turbine vendors, and merchant generators with spare capacity. The second-order effect is that data-center demand can steepen forward power curves in constrained regions faster than utility earnings can re-rate, because permitting and interconnection lag actual load announcements by quarters.

The more interesting setup is a spread trade between infrastructure that can monetize scarcity and regulated utilities that must spend first and earn later. Utilities with heavy capex plans may face higher financing costs and slower regulatory recovery, while power equipment and EPC names can convert order visibility into margin expansion much sooner. Over 6-18 months, the structural winners are likely to be grid-enablers rather than pure utilities: transmission, transformers, substations, and dispatchable generation.

Consensus may be too linear on "AI = utility upside." The missing issue is that load growth can be bullish for power prices but still bearish for equity multiples if it forces balance-sheet strain, longer project timelines, or customer attrition from delayed interconnections. The thesis is falsified if hyperscaler capex slows, if regional power prices fail to tighten by the next summer peak, or if regulators approve rapid rate-base recovery that neutralizes the financing headwind.