AI Data Centers: Power Ordered Is Not Power Delivered

The question behind every megawatt headline

A developer announces a 500 MW campus. An equipment supplier receives an order. An investor sees a power shortage. These statements can all be true while the campus still has no usable computing capacity.

The missing question is simple: what has actually been delivered, to which site, and under what operating conditions?

This guide offers our research framework. It does not estimate a particular project's completion date or recommend a security.

Start with the unit and the boundary

Megawatts measure power; megawatt-hours measure energy over time. A connection rating, a generator's nameplate output and IT load inside a data hall are different boundaries. Do not add them together or compare them without reconciling what each number measures.

A campus may also be built in phases. The ultimate site plan can be much larger than its first energized building. Separate the full development ambition from the phase that customers can use today.

Primary-source context

The U.S. Department of Energy's data-center resource hub explicitly notes that its energy-use estimates model equipment shipments and do not directly address growth in grid or onsite supply. A demand projection therefore does not establish that the electricity infrastructure will be ready.

DOE's electricity demand hub discusses grid modernization, onsite generation, storage and demand flexibility among potential responses. These are different delivery paths, not interchangeable proof of uninterrupted service.

Follow five milestones

1. Announced demand. Identify the site, phase, expected load and intended date. A press release establishes an intention.
2. Connection arrangements. Look for documented service terms, required network upgrades, cost responsibilities and any curtailment conditions. Study progress is different from permission to energize.
3. Equipment delivery. Track transformers, switchgear, cooling and backup systems against the actual project schedule. An order establishes demand for a supplier; shipment establishes a later milestone.
4. Installation and commissioning. Look for completed tests and customer acceptance for the relevant scope. One accepted subsystem does not demonstrate that the entire site is ready.
5. Usable operation. Ask what IT load can run, at what availability, and whether customers have begun using and paying for it.

Our interpretation: the most informative update moves a named project from one milestone to the next. Another announcement of the same ultimate campus size adds less evidence.

An illustrative example

Hypothetical numbers, not company data: a 500 MW campus plan has a 100 MW first-phase connection. Only 60 MW of that phase has been commissioned. If that 60 MW measures total facility power and power usage effectiveness is 1.20 at the stated operating point, the corresponding IT load is 60 ÷ 1.20 = 50 MW.

That calculation is a boundary check, not a forecast. Actual utilization, redundancy, operating restrictions and future phases need their own evidence. Calling the site “500 MW operational” would skip several steps.

What to monitor next

• Phase-level energization and commissioning dates, with updates to earlier targets.
• The definition of reported MW: site plan, contracted service, energized facility or IT load.
• Required grid upgrades and equipment delivery tied to the same phase.
• Service restrictions and the operating duration of any proposed onsite supply.
• Customer acceptance, utilized IT load and the supplier's revenue and cash conversion.

What would weaken the constraint case?

More completed connections, shorter comparable equipment lead times, wider supply alternatives and improving customer delivery schedules would weaken a scarcity thesis. Persistent postponements may support the existence of a constraint while reducing near-term supplier earnings. The industrial problem and the investment outcome need separate tests.

Apply this framework to Vertiv vs Eaton, then read our backlog guide before translating orders into financial expectations.

Sources and scope

Primary sources: the two DOE hubs linked above; checked October 4, 2026. The five milestones and numerical example are Constraint Alpha's analytical framework. No project-level shortage estimate, valuation or expected return is calculated.