KKR’s Agrawal Says Market May Be Underestimating AI Power Needs

KKR’s Raj Agrawal said AI’s rapid adoption may drive much higher power demand than the market is pricing in, noting companies and consumers are still very early in usage. His remarks suggest a constructive long-term backdrop for data center power and infrastructure spending, but the article contains no specific financial forecasts or company updates.

Analysis

The market is still pricing AI as a software capex cycle, but the second-order constraint is increasingly physical infrastructure: grid interconnects, backup generation, transformers, and land-constrained datacenter buildouts. That shifts the value pool away from pure model beneficiaries toward the owners of scarce bottlenecks, especially firms with regulated rate bases or contracted, inflation-linked cash flows. In that setup, the cleanest beneficiaries are not necessarily the hyperscalers but the adjacent infrastructure stack that monetizes every incremental MW demand.

The key underappreciated risk is timing mismatch. AI demand can compound quickly, while power supply additions remain gated by permitting, transmission lead times, and equipment shortages that can stretch 18-48 months. That creates a near-term squeeze in power pricing, interconnect queue values, and equipment margins long before end-demand visibility shows up in reported earnings; it also means any slowdown in AI capex would hit the most exposed data-center suppliers first, not the broader theme.

KKR is an interesting read-through because the firm has exposure to real assets that can reprice with this bottleneck, but the more actionable implication is to own the picks-and-shovels around electrification rather than the narrative itself. The contrarian mistake is assuming AI power needs are a back-half-of-cycle story; if utilization rates rise faster than expected, the market may have to re-rate utility growth, gas peakers, and transmission beneficiaries within the next 2-6 quarters. The reverse catalyst is simple: if model efficiency improves faster than demand growth, or if enterprise adoption stalls, the infrastructure premium compresses sharply because those assets are valued on long-duration throughput assumptions.

Key Decisions for Investors

• Overweight KKR as a real-assets proxy to AI infrastructure scarcity over the next 6-12 months; use pullbacks to build a starter position, with upside tied to revaluation of power-linked assets rather than fund flows alone.
• Long XLU / short QQQ as a 3-6 month relative-value trade: if AI drives power scarcity faster than earnings growth in software, utilities and grid-exposed names should outperform growth beta by 5-10% on a relative basis.
• Buy call spreads on utilities or grid equipment beneficiaries with explicit data-center exposure (e.g., ETN, PWR, NRG) into the next 1-2 quarters; risk/reward improves if the market starts pricing interconnect and transformer shortages before consensus revises estimates.
• Avoid chasing the most power-intensive AI infrastructure names at full valuation; if datacenter capex guidance weakens or efficiency breakthroughs emerge, these are the first names to derate over a 1-2 quarter horizon.
• Pair long gas-peaker / power price beneficiaries against short a basket of AI software names with the least pricing power; this captures the bottleneck without taking full directional risk on the AI narrative.