AI Data Center Bubble: Can Power Shortages Hide Overbuilding?

A power shortage sounds like evidence against an AI infrastructure bubble: if capacity is scarce, how can the industry be overbuilding? The apparent contradiction disappears when location, timing and project economics are considered separately.

Scarcity is specific#

A facility can have equipment but lack a completed grid connection. Another location can have electricity but lack a customer willing to accept its latency, price or operational requirements. Those projects do not automatically solve each other's constraints.

The IEA's 2025 analysis of energy supply for AI examines the electricity sources supporting data centers. Its projections are scenarios, not guarantees for the revenue or utilization of an individual property.

A bottleneck can protect prices and delay revenue#

In our analytical framework, limited available capacity can support pricing for a connected operator while delaying cash receipts for a developer still waiting to open. The same industry constraint can therefore help one company and hurt another.

A hypothetical project with annual fixed commitments of $20 million incurs $10 million over a six-month delay before considering other costs. This simple example is not a claim about any named operator; it illustrates why deployment timing matters.

Ask what is actually secured#

Distinguish land ownership, a proposed power agreement, an executed contract and an operating connection. Read who pays for upgrades and whether the customer can change the schedule.

Then test whether commitments on equipment and financing continue even if the customer deployment slips. A headline capacity figure is not a substitute for this sequence.

The bubble question remains financial#

An investment can be useful and scarce yet overpriced. Conversely, a delayed project need not be permanently uneconomic. The outcome depends on the cost of waiting, eventual pricing and the capital committed.

Power constraints are therefore neither proof of an AI bubble nor a universal rebuttal. They are variables to incorporate into project cash-flow scenarios.

More in this analysis series#

Start with AI stocks versus railroad mania. Continue with AI Chips vs Railroad Tracks: Depreciation Changes the Bubble Debate.

Analysis published October 1, 2026. Historical comparisons and hypothetical examples are not forecasts. Educational information, not investment advice.