Data Center Load Study Checklist: What Developers Need Before Utility Submission

A practical workflow for turning tenant demand into a utility-ready load request.

A data center load study is the structured analysis that tells a utility how much power a project needs, when it needs it and how that load will behave on the grid. For developers, it is no longer a back-office engineering package. It is one of the first filters that decides whether a site is real.

Power demand has moved faster than utility planning cycles. Lawrence Berkeley National Laboratory's 2025 interconnection queue work showed active generator and storage queues above 2,600 GW at the end of 2024. Large-load requests from AI data centers add a different problem. They are massive demand projects asking the grid to serve them.

That changes the diligence workflow. A developer cannot send a loose megawatt number and wait for the utility to clean it up. The submission has to describe load, ramp, phasing, redundancy and operating assumptions with enough precision for the utility to run system studies.

Start with the committed IT load

The first number is not the headline campus size. It is the committed IT load by phase.

A 500 MW land position means little if the first phase is 72 MW, the second phase is contingent on tenant leasing and the full buildout depends on a second substation. Utilities study what the system must serve. Developers should separate:

1. Day-one critical load

2. Contracted tenant load

3. Expected expansion load

4. Optional long-term campus load

5. Temporary construction power

6. Non-IT building load

The distinction matters because utilities may treat speculative load differently from contracted load. A request backed by a signed tenant, a dated ramp schedule and evidence of capital commitment gets a different internal conversation from a speculative site control package.

AI can help by extracting requirements from tenant term sheets, design briefs, one-lines and prior utility correspondence. Human judgment still has to decide which load cases are credible enough to submit.

Build the ramp schedule before asking for capacity

The load study should show when demand arrives because transmission upgrades, substation work and feeder construction do not clear at the same speed.

A useful ramp schedule includes:

• Requested service date for each phase

• MW by phase and by year

• Expected commissioning window

• Temporary backfeed needs

• Redundancy assumption, such as N, N+1 or 2N

• Known tenant deadline pressure

• Flexibility if utility upgrades slip

A developer may underwrite a 24-month tenant deadline while the utility needs 48 months for substation expansion. That gap is not a procurement issue. It is site viability.

The AI role is schedule reconciliation. It can compare requested energization dates against similar utility timelines, known equipment lead times and milestone dependencies. The human role is escalation. If the ramp is impossible, the commercial team has to decide whether to renegotiate phasing, seek interim power or drop the site.

Separate firm load from flexible load

Utilities increasingly want to know whether a data center can behave flexibly. That does not mean every AI campus can turn off at will. It means the developer should identify which parts of the load are truly critical and which could be phased, curtailed or supported differently.

The load study should distinguish:

• Firm critical IT load

• Mechanical and cooling load

• Office and support load

• Battery charging load

• Optional expansion load

• Loads that could participate in demand response

This matters in constrained markets. A project with no operating flexibility may face longer upgrade timelines or tougher cost allocation. A project with phased demand, onsite backup and transparent operating logic gives the utility more options.

Do not promise flexibility that operations cannot deliver. AI can model scenarios, but operations leaders have to confirm what can be curtailed without breaching tenant commitments.

Include onsite generation and backup assumptions

Backup power is part of the load story. Diesel generators, gas turbines, batteries and fuel cells all change how the utility evaluates reliability, emissions and outage behavior.

The load study should describe:

1. Backup generation type and rating

2. Runtime assumptions

3. Testing schedule

4. Black-start capability, if any

5. Battery duration and dispatch logic

6. Whether onsite generation will parallel with the grid

7. Whether exports are expected or prohibited

The parallel operation question is especially important. A backup-only generator creates one utility conversation. A behind-the-meter power strategy that may operate alongside the grid creates another.

Document the evidence package

A good load study is not only a spreadsheet. It is an evidence package. The utility should be able to see where every assumption came from.

Developers should attach or reference:

• Tenant requirements

• Conceptual site plan

• Electrical one-line

• Phasing plan

• Cooling basis of design

• Backup power narrative

• Construction schedule

• Prior utility emails

• Nearby substation and feeder context

• Known transmission constraints

Agentic AI can turn a messy folder into a structured submission checklist, flag missing evidence and compare the request against prior utility templates. It should not invent assumptions.

Treat the load study as site selection diligence

The load study belongs before LOI when possible. If the utility answer determines schedule, cost and tenant deliverability, then the load study is not late-stage engineering. It is core site selection.

The practical workflow is simple:

1. Define phased IT load

2. Build ramp schedule

3. Separate firm and flexible load

4. Document backup and onsite generation

5. Assemble evidence package

6. Submit a utility-ready request

7. Track responses, study milestones and upgrade exposure

The best teams look for weak signals early: vague service timelines, repeated clarification requests, missing substation capacity and conflicting upgrade assumptions.

AI compresses the work. It can read documents, assemble the package and maintain the risk register. The decision still belongs to the development team. If the load cannot be served when the tenant needs it, the site is not a data center site. It is land with a power story that has not been proven.