Summary
- AEP added 6GW of signed load agreements in the second quarter, primarily in Texas, taking contracted load growth through 2030 to 69GW.
- The contracted customer set includes hyperscalers, data centres and industrial companies; AEP does not provide a data-centre-only or AI-only share.
- The company secured a further 3GW of gas-turbine capacity, taking total secured capacity to approximately 13GW for possible deployment through 2031.
- Another 10GW of turbine capacity is being evaluated through 2035 and must not be treated as committed construction.
- AEP expects fully executed take-or-pay electric-service agreements to support up to $16 billion of residential-customer cost offsets, but these are prospective rather than realised bill savings.
- The conversion chain still requires regulatory approval, sites, fuel, transmission, construction, customer energisation and metered demand.
A turbine reservation turns one shortage into a schedule
Large gas turbines are not bought from a warehouse when a utility discovers that demand has arrived. Manufacturing slots, major components and delivery dates can be committed years ahead. By securing capacity, AEP seeks to avoid reaching the late 2020s with signed customers but no equipment available to build supply.
The extra 3GW announced during the quarter takes secured capacity to about 13GW. The company describes it as available for potential deployment through 2031. “Potential” is doing important work. Reservation can establish access to equipment without selecting every site, receiving every permit or taking a final investment decision on every plant.
That optionality has economic value. AEP can move faster if load connections, regulatory decisions and system studies support construction. It can also preserve time while technologies and customer schedules change. The value depends on the commercial terms: deposits, cancellation rights, price adjustment, assignment and delivery obligations are not detailed in the results release.
Calling 13GW approved generation would therefore be wrong. The equipment is neither evidence of completed power stations nor a guarantee that gas supply and transmission will arrive. It is one secured input in a much longer production chain.
The demand contract is stronger than a forecast, but weaker than a meter
AEP’s 69GW represents contracted load growth through 2030 after 6GW of signed agreements were added in the second quarter, primarily in Texas. A signed electric-service agreement is a higher-quality demand signal than an inquiry: it can impose payment and performance obligations.
The release says fully executed agreements use take-or-pay protections. Such a provision can require payment even if the customer consumes less than an agreed level, helping the utility recover infrastructure costs. Yet the public aggregate does not reveal minimum volumes, duration, credit support, exit charges or the conditions under which payments can be adjusted.
Contracting does not remove construction risk on the customer side. A data-centre developer may have to secure land, permits, servers and financing; an industrial buyer may face its own market cycle. A contract can allocate the consequence of delay without preventing delay.
Metered demand is the final physical test. A facility can connect in phases and remain below its reserved load for years. The financial test is separate: minimum payments could protect other customers even when utilisation is low, but only if the agreement is enforceable against a creditworthy counterparty for the life of the utility assets.
Sixty-nine gigawatts is not a measure of AI
AEP identifies hyperscalers, data centres and industrials within the customer set. It does not say how much of the 69GW belongs to each. Describing all of it as data-centre demand, or all data-centre demand as AI, would invent a denominator.
The mix matters because load shapes and economic risks differ. A computing campus can run continuously but ramp capacity as halls open. A factory may have production shifts, commodity exposure and different outage tolerance. A hyperscaler can include cloud, storage, network and conventional enterprise workloads alongside AI.
Geography also changes the system response. The additional 6GW was primarily in Texas, but the 69GW total spans AEP’s broader service territory and does not supply a site-by-site schedule. ERCOT and PJM have different market, planning and regulatory arrangements; one turbine or transmission response will not fit every jurisdiction.
Investors need a matrix rather than one headline: contracted megawatts by customer class, service territory, expected connection year, minimum-payment coverage and required network stage. That would allow the demand book to be compared with supply and capital plans without disclosing customer names.
Equipment and demand can miss each other in both directions
Securing turbines against contractual demand is a form of matching, but the disclosed aggregates do not establish that every unit of equipment corresponds to a particular customer or location. Thirteen gigawatts of possible turbine deployment is far smaller than 69GW of contracted load growth, and the two figures use different horizons and definitions.
The gap does not imply a 56GW deficit. Existing generation, renewables, storage, market purchases, demand response and transmission imports can all contribute. Some customers will connect later, consume below reserved capacity or supply part of their own energy. Conversely, new turbines may serve system reliability rather than one customer.
Timing can fail in either direction. Demand may connect before generation and grid work are ready, leaving a reliability problem or a delayed customer. Equipment may arrive before approvals, fuel or site construction, tying up capital. A customer may postpone after a plant has moved too far to cancel economically.
The useful schedule must match annual increments: contracted connection dates, available generation capacity, turbine delivery, permit status, transmission completion and expected metered ramp. Aggregate endpoint figures obscure these interfaces.
Ten gigawatts under evaluation is an option on a later problem
AEP is also evaluating another 10GW of turbine capacity through 2035. Evaluation should remain on its own line. It can include market discussions, engineering options or possible reservations; the disclosure does not say it is ordered, approved or included in a committed build programme.
Adding 10GW to the secured 13GW and calling 23GW of new gas plants would collapse the decision boundary. The first amount is equipment capacity secured for possible deployment. The second is an opportunity still being considered. Neither is equivalent to operating nameplate capacity.
The longer horizon exposes technology and policy risk. Plants contemplated for the 2030s will operate under fuel-price, emissions, market and regulatory conditions that cannot be fixed today. A reservation can preserve an option, but exercising it still needs a project-specific case.
That case should compare the turbine with alternatives on firm capacity, construction time, fuel infrastructure, transmission, reliability contribution and total cost. The SEC-filed release does not publish those comparisons, so it supports no conclusion about the final generation mix.
The $16bn household offset is a forecast that needs a bill denominator
AEP says fully executed take-or-pay electric-service agreements support up to $16 billion of expected cost offsets for residential customers. The mechanism is plausible: large new customers can contribute to shared fixed costs and fund infrastructure that would otherwise be recovered more broadly.
But “up to”, “expected” and “support” all signal a future estimate. The amount is not cash already returned to households and is not a realised reduction in bills. It depends on customers connecting, paying, remaining in service and bearing the costs attributed to them.
To test the claim, AEP should report the calculation by jurisdiction. The numerator would include large-load revenue or direct contributions net of incremental operating and financing costs. The denominator would be the residential customer base and the bills that would have applied without the new demand.
The estimate also needs a time profile. Sixteen billion dollars over many years cannot be compared with a one-year bill. Regulatory decisions will determine how and when benefits flow. If investment and revenue arrive in different periods, customers could face higher costs before the expected offset appears.
Stress cases are equally important: delayed connection, lower utilisation, counterparty default, plant cost overrun and fuel-price change. Take-or-pay provisions can absorb some of these risks but not necessarily all of them.
The $78bn capital plan must be reconciled project by project
AEP maintained its $78 billion capital plan for 2026 through 2030 and identified more than $10 billion of incremental capital opportunities. Neither figure should be relabelled as the cost of the 69GW demand book or the 13GW turbine reservation.
The plan covers a wider utility programme. Some projects may support ordinary replacement, reliability, transmission, distribution, generation or new customers. Incremental opportunity is not yet the same as approved expenditure. The filings do not present a customer-by-customer allocation.
Regulators will decide which assets are prudent, how costs enter rates and what protections large users must provide. A take-or-pay contract can back a particular investment, while assets benefiting the broader network may be shared. The allocation can differ across AEP’s jurisdictions.
The best reconciliation would link signed load to specific classes of investment without revealing commercial identities: customer-funded connection, dedicated network upgrades, shared transmission, generation and system-wide reliability. It would show security and minimum payments alongside the cost they protect.
The operating scorecard begins where the announcement ends
For demand, AEP should track signed capacity, scheduled connection, energised capacity, measured peak, average consumption and payments. For supply, it should track equipment reservation, delivery, site approval, construction, fuel readiness, grid connection and commercial operation.
The two scorecards should meet by year and region. A surplus in one place cannot automatically serve a deficit in another. Nor can a turbine delivery compensate for a delayed transmission line.
Realised residential offsets belong on the same dashboard: dollars credited or costs avoided, affected customer count and period. That would distinguish an enforceable contractual benefit from a modelling assumption.
AEP has made a consequential move. It is not merely forecasting that data centres and factories may need power; it is signing service agreements and reserving scarce generation equipment. That improves evidence on both sides of the future power balance. It also commits the utility to manage two calendars whose gaps can be expensive. The success measure is not 69GW or 13GW on a slide. It is the portion that becomes timely, approved, fuelled, connected and paid for without shifting an unpriced risk to households.