XLU’s AI Power Thesis Is Running Into a Reality Check

Utilities became one of the market’s less obvious AI trades because data centers need enormous amounts of electricity.

That thesis is now facing a reality check.

The source highlights tighter review of data-center interconnection requests in Texas and a sharp reduction in parts of the utility pipeline after customers were asked to put more money behind their applications.

The issue is simple.

Requested power is not the same as committed power.

A developer can submit applications to multiple utilities for the same project.

If every request is counted as future demand, the pipeline can look far larger than what will actually be built.

The reason is that regulated utilities earn returns on invested capital.

New transmission lines, substations and generation can increase the rate base.

Building infrastructure for projects that never materialize can create problems for utilities, regulators and existing customers.

Stricter application fees and collateral requirements are designed to separate serious projects from speculative requests.

When pipelines shrink after fees are introduced, that suggests some of the earlier demand was not firm.

The source points to American Electric Power and Exelon as examples of utilities refining their data-center pipelines.

That does not mean AI power demand is fake.

The amount of real demand is still enormous.

The question is how much of the headline queue should be treated as bankable.

For XLU investors, this matters because several of the fund’s largest holdings are directly involved in the data-center power story.

NextEra, Southern, Duke, Constellation and AEP all have different business models, but the broad sector has benefited from the idea that AI will drive years of electricity demand.

If only part of the requested load becomes real, utility growth expectations may need to become more selective.

The rate structure also matters.

A utility that signs long-term contracts or requires customers to fund dedicated infrastructure has a different risk profile from one that builds first and hopes demand arrives.

Investors should reward the former more heavily.

Power prices are another variable.

Large new loads can raise system costs and create political pressure if residential customers believe they are subsidizing data centers.

Regulators may respond by changing tariffs or requiring hyperscalers to bear more of the cost.

That can protect households, but it can also affect utility returns.

XLU remains a diversified utility ETF.

It is not a pure AI-power fund.

Many holdings still earn returns from traditional regulated electricity and gas operations.

That can provide stability if data-center expectations cool.

At the same time, the ETF’s valuation can be influenced by the growth narrative attached to AI demand.

If that premium fades, XLU may trade more like a traditional defensive sector and less like an infrastructure-growth theme.

Interest rates add another layer.

Utilities are capital intensive and often carry significant debt.

Higher Treasury yields can pressure valuations even when operating growth remains healthy.

The strongest utility setups are therefore those with real contracted data-center demand, manageable financing costs and regulators willing to support the investment plan.

The quality of the counterparty is another major variable.

A signed agreement with a well-capitalized hyperscaler is more valuable than a speculative request from a developer still searching for financing.

Utilities that disclose deposits, collateral and signed transmission agreements give investors a better way to separate high-probability demand from marketing numbers.

That transparency can become a competitive advantage.

The sector may ultimately benefit from the current cleanup because unrealistic queues make planning harder for everyone.

Removing duplicate or speculative requests can help utilities direct capital toward projects that are most likely to be completed.

That could reduce wasted capex even if the headline gigawatt pipeline falls.

For XLU investors, a smaller but more credible demand pipeline may be healthier than a huge number that the market cannot verify.

BTI’s bottom line: AI electricity demand is real, but utility pipelines need to be audited for quality. Interconnection requests are not contracts. As regulators and utilities demand more collateral, the market will get a clearer view of which projects are real. That is healthy for the sector, but it may reduce the growth premium investors have been willing to pay for every utility exposed to the AI power boom.