Data Center Load Pipeline vs Contracted: How to Discount It

Six utilities and one turbine manufacturer put large-load numbers on the record in the same earnings window, and no two of them counted the same way. A pipeline total, a forecast of load served by a future date, an executed agreement and a turbine slot reservation measure different things. Dividing one by another does not establish the probability that a particular utility project will connect.

The useful output of reading all of them together is not an average. It is that a large-load gigawatt figure quoted without its conversion tier is unusable, and that two utilities both saying “pipeline” are frequently not describing comparable objects.

The utility large load pipeline discount, six ways

Here is what the quarter actually disclosed.

PG&E reported 12,710 MW of total data center pipeline at June 2026, compared with 5,090 MW in March. Its Q2 presentation breaks June into 8,200 MW in application and preliminary engineering, 3,880 MW in final engineering, 490 MW with an interconnection construction agreement and 140 MW in construction. Even the preliminary stage requires an application and a study fee; final engineering requires a signed work performance agreement and payment covering design and procurement activities. Our earlier 7:1 comparison treated a forecast as contracted load. The presentation’s stage breakdown does not support that comparison as a pipeline-to-contract conversion rate.

CenterPoint submitted more than 17 GW to ERCOT’s Batch Zero process and expects around 14 GW to qualify, roughly 1.2:1, with about $900M in customer cash commitments and deposits sitting behind the submissions.

DTE reported 2.4 GW of executed agreements against another 5 to 6 GW it describes as development pipeline, so somewhere between 3:1 and 3.5:1 depending on which end of its own range you take. The executed tier is almost entirely two named customers, which is rare disclosure: Oracle at 1.4 GW, approved and under construction with demand ramping in 2027 and 2028, and Google at 1 GW still before the Michigan Public Service Commission.

Xcel discloses in three explicitly labeled tiers: roughly 2 GW contracted or under construction now, roughly 4 GW contracted by end of 2027, and roughly 3 GW of incremental line of sight ramping into the mid-2030s. The middle tier carries a date, which makes it the most plannable disclosure in the set.

Exelon identifies 11 GW of high-probability load, split 9 GW at ComEd and 2 GW in the Mid-Atlantic. Of that, 4 GW is backed by signed, FERC-approved Transmission Security Agreements supported by about $1 billion in collateral. The other 7 GW predates the TSA process.

AEP publishes the largest number in the set and the one that looks cleanest on its face: 69 GW of contracted load growth through 2030, up 6 GW in the quarter from a 63 GW base. We will come back to that one, because it is not what it appears to be.

GE Vernova, on the equipment side of the same market, signed 20 GW of new gas equipment contracts in the quarter. Of those new contracts, 2 GW were orders and 18 GW were slot reservations. The company also converted 10 GW of previously held reservations into orders while shipping 3 GW. The 18-to-2 comparison describes the mix of new equipment contracts in that quarter; it is not a utility load conversion rate, nor the conversion rate of previously reserved turbine slots.

AEP’s contracted data center load GW is really two numbers

AEP’s 69 GW is the most instructive disclosure of the quarter, and the instruction is buried in a qualifier.

On the earnings call, CFO Trevor Mihalik described the figure as 69 GW of contracted load additions through 2030 “all supported by a combination of fully executed ESAs and LOAs.” Those are two different instruments. An electric service agreement is a service contract. A letter of agreement is a commitment to proceed toward one.

The split matters because Mihalik also disclosed that 45 GW of the 69 sits in AEP Texas and is supported by fully executed LOAs. Meanwhile CEO Bill Fehrman attached the company’s headline customer-benefit claim specifically to the other instrument, projecting fixed cost offsets for residential customers of up to $16 billion at its vertically integrated utilities “supported by fully executed take-or-pay electric service agreements.” Those vertically integrated utilities are not in Texas.

So the single largest contracted figure any US utility published this quarter contains its own two-tier structure, and the take-or-pay terms that make contracted load meaningful for equipment buyers attach to the smaller of the two tiers. That is not a criticism of the disclosure. AEP said exactly what it did, in public, on the record. It is a warning about what happens two hops downstream, where “69 GW contracted, backed by take-or-pay” becomes a single sentence in a trade summary and then a planning input in somebody’s territory model.

Read the qualifier. It is almost always there, and it is almost always dropped first.

Collateral is the only tier that cannot be walked back for free

Announcements are free. Letters of agreement can carry obligations that depend on their terms. Collateral puts money at risk, which is why it is the sharpest tool available for reading data center load pipeline vs contracted claims.

A developer that has posted cash against a FERC-approved agreement has converted a press release into a balance-sheet position, but the procurement milestone still depends on the agreement, engineering approvals and project schedule. Collateral alone does not prove that equipment has been ordered.

Two utilities now publish the collateral directly, and comparing them is worth doing. Exelon reports about $1 billion behind 4 GW. AEP reports collecting nearly $2 billion in cash or collateral in a single month, covering all required credit support for the full 45 GW in its Texas Batch Zero filing. Divide those out and the per-kilowatt commitment behind Exelon’s TSA tier is roughly five times what sits behind AEP’s Texas tier.

Neither utility is wrong. They are running different instruments under different regulators, and ERCOT’s Batch Zero credit requirement is calibrated to a different risk than PJM-region transmission security. That is the point. Even collateral-backed load, the strictest tier available, is not one uniform thing across territories. The deposit structures themselves are contested and moving, as the ERCOT interconnection pause and its deposit terms demonstrated three weeks later when a governor froze the queue mid-cycle with hundreds of millions already posted.

The instrument decides who buys the substation

This is where the exercise stops being media criticism and starts paying for itself.

A take-or-pay electric service agreement can require payment even when the customer consumes less than its commitment. That can protect revenue, but the phrase alone does not establish who owns, funds or purchases the substation. Read the construction agreement, tariff, cost-recovery provisions and termination security to determine how the equipment is paid for and when procurement can proceed.

For a municipal utility or cooperative negotiating its first large-load interconnection, that distinction is the whole negotiation. AEP’s earnings disclosures identify instruments to investigate further; an earnings release does not replace the tariff or executed agreement. The cost-allocation approach Pennsylvania took with its large load tariff is the regulatory counterpart, and the equipment scope that follows from a data center substation design decision is where the money actually goes.

DTE supplies the sharpest example of what a schedule commitment is worth. The company has publicly said it plans to refrain from filing another electric rate request until at least 2028, contingent on the Oracle project coming online as planned. That converts one customer’s interconnection schedule into a regulatory promise with a named date. Every long-lead item on that project now sits on the critical path of an affordability commitment rather than a construction plan, which tells you something about how hard that utility will work to hold a delivery slot.

How to read utility capex and data center load claims: four questions

The checklist is short enough to apply to the next press release without us.

One. Which tier is this? Executed, contracted, in negotiation, or inquiry. If the release does not say, the number is a pipeline figure regardless of the word attached to it.

Two. Is there a named instrument, and what does it obligate? Take-or-pay ESA, letter of agreement, transmission security agreement, or nothing. Ask whether the obligation survives the customer changing its mind.

Three. What did the project have to post to be counted? Check deposits, security and withdrawal terms, then ask for actual progression and cancellation data. These disclosures do not establish a universal conversion rate for either paid or free queues.

Four. Did the number move, and did the utility explain the move? DTE’s $30 billion capital plan was flat quarter over quarter, which reads as restraint until you notice the prior plan was $24 billion over 2025 to 2029. The plan rose 22% one quarter earlier, attributed primarily to a single data center customer. Flat since the increase is a different claim from flat.

Run those four against any utility disclosure and the data center load pipeline vs contracted gap stops being a mystery and becomes a number you can plan against. Skip them and you inherit whoever summarized it first. We applied the same discipline to three utilities’ hyperscaler capex strategies in Q1, and the pattern held then too.

The open term nobody has priced yet

One thing is missing from every one of these agreements, and it has a deadline.

FERC ordered NERC on July 16, 2026, in Docket RD26-7-000, to file an initial set of mandatory reliability standards for integrating computational loads by December 31, 2026. The same order asks NERC for proposed registry criteria for computational-load entities by that date, and for an informational workplan on the next phase of standards by March 1, 2027. The order points back to NERC’s own incident reviews. The first of those describes a July 10, 2024 event in which a 230 kV line fault was followed by about 1,500 MW of data center load disconnecting on the customer side.

Registration is what makes a NERC standard enforceable, with civil penalty exposure attached. The order sets no size threshold. If the criteria NERC files reach a given facility, that large computational load becomes a registered entity in its own right rather than a customer of one. None of the public reporting on the agreements signed this quarter says who carries the registration and compliance obligation at the point of interconnection, and most of these contracts were drafted when the load could not be a registered entity at all.

That is a real open contract term with a known closing date. A cooperative or municipal utility courting large load should be asking now, before those criteria are filed, whether its interconnection agreement allocates reliability-compliance responsibility to anyone. The broader FERC large load interconnection rulemaking is the other half of the same clock.

Large-load disclosures move every quarter and the qualifiers move with them. The Feeder is our free weekly briefing on what changed in the filings and what it means for equipment timing. Sign up here.

Related Reading

• ERCOT Data Center Interconnection Pause: The Deposit Trap
• Hyperscaler Utility Capex 2026: Three Q1 Strategies
• Data Center Substation Design Is a Procurement Decision