The Next Phase of the AI Boom

Switchyard Partners CEO Joe Kaiser said electricity availability may become the primary constraint on the next phase of the AI boom, potentially driving data centers toward stranded-power locations. He argued that sharply reducing compute required per task will be critical to sustaining AI investment, while market incentives could help establish AI guardrails. The segment highlights a growing infrastructure and power-supply risk for AI-related capital spending.

Analysis

The investable implication is not simply higher utility demand: it is a geographic repricing of power access. AI capacity additions will favor locations with curtailed renewables, underutilized thermal assets, nuclear-adjacent generation, or faster interconnection queues; this creates value for merchant generators and transmission developers with deliverable power, while penalizing hyperscaler projects dependent on congested load zones. The bottleneck is likely to shift from GPUs to firm power contracts, substation equipment, transformers, and permitting—areas where supply cannot scale on a normal semiconductor cycle.

Over the next 1-3 months, the market may continue to reward GPU-linked names while underpricing electrical balance-of-plant suppliers. Eaton (ETN), Vertiv (VRT), GE Vernova (GEV), Quanta Services (PWR), and Hubbell (HUBB) have more direct exposure to the physical deployment constraint, but valuation discipline matters after substantial AI-related rerating. Watch announced data-center lease commitments in secondary power markets and utility integrated-resource-plan revisions as confirmation that load forecasts are becoming contracted demand rather than speculative pipeline.

The contrarian risk is that lower compute intensity reduces power demand per AI task faster than total inference volume expands. That would compress the scarcity premium embedded in data-center infrastructure multiples, particularly for VRT and power-equipment names priced for sustained double-digit order growth. Conversely, a utility-rate backlash, transmission permitting delays, or reliability-driven restrictions on large-load interconnections could make existing powered data-center capacity more valuable but delay new equipment revenue by 6-18 months.

There is no clean directional trade from this commentary alone; the key missing data are regional power-price spreads, interconnection queue timing, and whether hyperscaler power purchases are binding commitments. Treat this as a watchlist theme rather than a catalyst until quarterly bookings and backlog conversion demonstrate that AI load is pulling through grid capital expenditure.

Key Decisions for Investors

• Build a watchlist for long ETN/PWR versus short a broad technology proxy (QQQ) only after utility load forecasts or hyperscaler capex guidance explicitly cite power/interconnection constraints; target a 3-6 month horizon, with thesis invalidated by backlog growth decelerating below management guidance.
• Prefer GEV selectively on evidence that gas-turbine, grid, or service-order backlog is converting to revenue; use a 6-18 month horizon and avoid chasing sharp AI-sympathy rallies absent disclosed order economics.
• Monitor VRT and ETN valuation versus forward order growth: if bookings soften while multiples remain elevated, consider downside hedges through puts or a relative short against QQQ. The falsifier is continued acceleration in organic backlog and margin guidance.
• Track US regional transmission organizations with large data-center interconnection queues—especially PJM and ERCOT—as an early signal. A material rise in queue withdrawal rates or large-load connection restrictions would favor existing powered assets but weaken near-term equipment-delivery expectations.