Is AI Hiking Your Energy Bill? Nic Carter Says 'The Data Doesn't Support It'

Castle Ventures partner Nic Carter argues that claims AI data centers are causing a broad spike in electricity bills are overstated, citing data showing real residential power rates have fallen over the past decade and that high-data-center states like Virginia, Texas, and Nevada still have low power prices. He also said AI query power needs have dropped 200x since 2022 and that hyperscalers are increasingly self-generating electricity, which could limit grid and ratepayer impact. The piece also highlights growing policy scrutiny, with House appropriators agreeing to let the Energy Department regulate data centers and Goldman Sachs forecasting U.S. data center power demand will rise to 66 GW in 2027 from 31 GW in 2025.

Analysis

The market is still pricing the AI power story too simplistically. The near-term read-through is not “higher electricity bills everywhere,” but a more nuanced redistribution of value: utilities and grid-infrastructure vendors with load growth in permitted, low-cost jurisdictions should benefit, while politically constrained power markets face a larger regulatory premium than a pure demand shock. In other words, the cleanest winners are not the hyperscalers themselves but the firms that monetize transmission, interconnection, and generation buildout where siting friction is low.

For GS, the implication is second-order but meaningful: if data center capex accelerates toward the 2027 load forecast, financing activity around power, grid upgrades, and on-site generation should expand. That supports advisory, project finance, and structured lending volumes more than it supports a directional equity call, and it can lift fee pools with a lag of 2-4 quarters as deals are underwritten and syndicated. The main risk is that tighter federal oversight or state-level cost allocation rules compresses returns on new projects, which would shift spending from balance-sheet-light growth to slower regulated investment.

The contrarian point is that the market may be overestimating residential backlash and underestimating how quickly hyperscalers can internalize energy supply. If AI inference efficiency continues improving at a rapid clip, the demand curve can grow while the marginal grid burden stays manageable, delaying the inflationary pass-through investors are expecting. The real tail risk is not immediate higher bills, but policy intervention that forces developers to fund more of the local infrastructure upfront; that would be bearish for marginal data center developers but bullish for utilities, EPCs, and equipment suppliers that can capture mandated capex.

Key Decisions for Investors

• Add a tactical long in utility/grid beneficiaries tied to new-load growth; prefer companies with transmission and interconnection exposure over pure generation. Time horizon: 3-12 months. Risk/reward: limited downside if data-center headlines fade, asymmetric upside if capex gets approved.
• For GS, lean modestly long into any pullback on the view that project finance, M&A, and structured power deals expand as AI capex migrates from hype to underwriting. Horizon: 2-4 quarters. Risk/reward: medium upside, lower beta than direct AI names.
• Pair trade: long regulated utilities / grid equipment, short unprofitable data-center developers or highly levered AI infra names. Horizon: 6-9 months. Thesis: policy costs and power-siting friction will separate capital-light winners from capital-hungry losers.
• Use long-dated call spreads on a grid-capex beneficiary rather than outright calls to avoid overpaying for momentum. Target 9-12 months to allow policy and capex approvals to convert into orders.
• Watch for any federal rulemaking that shifts infrastructure costs onto developers; if that emerges, rotate further into equipment suppliers and away from merchant power exposure. Trigger window: days to weeks after policy headlines.