Bridging the gas turbine gap: the four levers utilities are pulling for 2027 to 2029

Yesterday’s piece set the constraint: heavy-duty gas turbines from the three OEMs are quoting 48 to 60 month delivery windows, with the largest J-class units pushing into 2030 and 2031, and the IRP filings across the major ISOs have not been adjusting their COD assumptions to match. The follow-up question is the practical one. If a utility had penciled in 2 GW of new combined-cycle capacity for a 2028 commercial operation date, and the turbine slot does not exist, what shows up instead?

Four levers are doing the work. None of them substitutes one-for-one for firm dispatchable gas. All four are visible in the last six months of regulatory filings, capacity auction strategies, and utility earnings disclosures. The shape of the 2027 to 2029 reliability picture is being decided by how aggressively each ISO and each individual utility leans on each lever.

Lever 1: Battery storage acceleration

Four-hour and now eight-hour battery storage is the lever utilities are pulling first because it is the only one that can actually be in service inside 24 months. Project cycle time for utility-scale lithium iron phosphate storage in 2026 is running 12 to 18 months from notice to proceed, against 60 plus months for a new combined-cycle. The cell supply chain, while not unconstrained, is not on the same binding curve that gas turbine hot sections are. Tesla, BYD, CATL, LG Energy Solution, and the US-domestic capacity coming online under the IRA Section 45X advanced manufacturing credits collectively support a US storage interconnection queue that is now an order of magnitude larger than the gas turbine queue.

The capacity-market accreditation rules, which historically discounted storage relative to its nameplate, have been moving. PJM’s effective load carrying capability framework, MISO’s accreditation update, and ERCOT’s performance credit mechanism debate are all converging on giving four-hour and eight-hour storage credit closer to what dispatch performance during scarcity events actually shows. The 2024 and 2025 ERCOT summer events, where battery storage discharged at the right hours and avoided the load-shed scenarios that earlier analysis had flagged, are reshaping the regulatory assumption.

The limit of the lever is that storage is not energy-unlimited. A four-hour battery dispatched at full output discharges in four hours and then needs to recharge from something. The case for storage as a gas substitute depends on what is on the grid to recharge it during the off-peak window, and that case is stronger in ERCOT and CAISO than in PJM and MISO, where renewable build is slower relative to load.

Where the lever is being pulled hardest:

• ERCOT. The 2026 interconnection queue holds more than 60 GW of storage capacity in active development, with about 12 GW expected to reach commercial operation by year end. Texas-specific market design (no capacity market, real-time energy and ancillary services revenue, and the Performance Credit Mechanism if and when finalized) makes the storage business case work without a separate capacity payment.
• CAISO. The decade-long storage procurement directive from the CPUC, paired with the post-2020 reliability orders, has produced the highest installed storage base in the country. The summer 2025 and early summer 2026 dispatch data show storage filling the four-hour evening ramp that gas peakers used to handle.
• PJM. The 2025/2026 BRA cleared at $269.92 per MW-day, and storage developers have moved aggressively to capture that revenue. The ELCC accreditation produces a discount but the price level still supports the project economics for projects that can clear interconnection.

Lever 2: Coal and old-gas life extension

The retirement schedule for coal-fired generation across the US in 2024 and 2025 baseline filings included roughly 60 GW of announced retirements between 2026 and 2030. A meaningful share of that capacity is now being deferred or actively unretired. The pattern is most visible in PJM and MISO, where capacity auction prices and reliability-must-run agreements have moved fast enough to make the economics of keeping a 40-year-old coal unit on the system pencil out, at least for a few more years.

The trade is real and the trade is dirty. A coal plant kept online past its planned retirement runs higher heat rate, higher emission factor, and higher fixed O&M than a new combined-cycle would. The emissions consequence is straightforward. The reliability case is that the alternative is a deeper capacity shortfall in the late-2020s window, and the cost of that shortfall in load-shed events would be higher than the cost of the additional tons.

Recent examples that have made it into filings:

• AEP, Duke, Southern Company. Multiple coal units originally scheduled for 2027 to 2029 retirement have been pushed by 2 to 5 years across the three holding companies’ service territories, with the rationale tied explicitly to load growth and to the difficulty of permitting and building replacement gas capacity on the prior schedule.
• MISO RMR proceedings. The reliability-must-run process at MISO has produced specific unit-level agreements that keep coal capacity available beyond the announced retirement date, with the cost recovered through a separate reliability charge on the relevant load-serving entities.
• Old gas peakers. Simple-cycle gas peakers that were on the way to retirement are being deferred on the same logic. Unit-level data is harder to aggregate because the peakers are smaller and less prominent in filings, but the pattern is the same.

The lever has a time-limit on it. The boilers, turbines, and condensers on a 40-year-old coal unit are not going to run reliably for another decade no matter how much O&M is funded. The lever buys 3 to 7 years, and at some point in the early to mid 2030s the units have to come off the system whether the gas turbine queue has cleared or not.

Lever 3: Demand-side reliability programs

The third lever is on the load side, and it is moving faster than the headline coverage suggests. Capacity-market reform across PJM, MISO, ERCOT, and ISO-NE has been treating large flexible load as a capacity-equivalent resource, and the hyperscalers and their data center operators have been quietly building out load-shedding and load-shifting capability that the IRPs of three years ago did not assume.

Three flavors of demand-side reliability:

• Hyperscaler load-shedding contracts. Google, Meta, Microsoft, and Amazon have signed or are negotiating contracts with utilities and ISOs that commit to specific load reduction during reliability events. The contracts in some cases include the data center operator running on-site generation or batteries to bridge the load-shed window, in others the contracted load is genuinely shed for a defined number of hours per year. The reliability value is real and it is now showing up in capacity auction qualification.
• Industrial and commercial demand response. The legacy DR aggregators (Enel X, CPower, Voltus, NRG Curtailment Solutions) have grown enrolled capacity meaningfully over 2024 and 2025, with the highest growth in PJM and ERCOT. The capacity products and the energy products both pay better than they used to, which is the actual driver.
• Behind-the-meter rate design. Time-of-use and critical peak pricing rates that have been on the books in California and a few other states for years are being adopted more aggressively in growth-market states. The peak load reduction from rate design is hard to measure cleanly but it is not zero.

Aggregate national flexible-load capacity available for ISO reliability calls in 2026 is somewhere in the 35 to 45 GW range, depending on accounting choices, up from roughly 25 GW in 2020. That is a meaningful contribution to the 2027 to 2029 picture even if it is not enough alone.

Lever 4: Behind-the-meter and direct hyperscaler deals

The fourth lever is the one that has shown up most visibly in headlines: hyperscalers and large industrial customers contracting directly for generation, sometimes behind the meter, sometimes on the utility side of the meter through long-dated PPAs that change the procurement signal.

The pattern includes the 2025 Three Mile Island restart agreement with Microsoft, the multiple data center PPAs against new and existing nuclear in 2025 and 2026, the Talen and Constellation hyperscaler arrangements, behind-the-meter gas generation announcements (with the well-documented air permitting fights that follow them), and a growing list of direct-procurement equity investments by hyperscalers into IPP projects.

The lever works because the hyperscalers have balance-sheet capacity to pay above the marginal capacity-market price for firm power that lands on their schedule. The lever’s effect on the broader reliability picture is mixed. On one hand, hyperscaler-procured generation shows up on the grid and serves load. On the other hand, the generation is contractually tied to specific load, the capacity-market accreditation treatment varies, and the procurement absorbs OEM delivery slots that would otherwise be available to traditional utility buyers. The net effect on reliability for non-hyperscaler load is harder to pin down than the gross capacity number suggests.

What this stacks up to

The four levers, summed across the ISOs, are doing real work. ERCOT is the cleanest case for “storage and load growth and coal stay-on do most of it” because the renewable build is large and the storage queue is large. PJM is the hardest case because load growth is high, the storage build is meaningful but smaller, the coal life-extension lever is being pulled hard, and the gas substitute is not coming on the original schedule. MISO sits in between. SPP is closer to ERCOT in storage queue dynamics but with more conservative IRP filings. CAISO is the case where the lever stack is most mature and the gas turbine problem is least binding, because the system has been running on storage and renewables and demand response for several years already.

The implication for capacity auctions is that the supply curves are going to look qualitatively different than the historical model suggests. Less of the cleared capacity will be new combined-cycle, more will be storage with accredited capacity values that have moved up, more will be coal-and-old-gas with explicit reliability designations, more will be demand-side. The clearing prices have already been doing what economic theory says they should. The composition of what clears at those prices is the part that is shifting.

Risks to the read

• Storage interconnection bottleneck. The storage queue is large but interconnection study cycle time has been the binding constraint at PJM, MISO, and parts of ERCOT. If interconnection reform delivers slower than expected, the storage lever weakens.
• Coal unit forced outage. Life-extended coal units run higher forced outage rates than the planned-retirement curve assumes. A single bad summer with multiple unplanned trips would expose the gap.
• Demand response performance. Capacity-accredited demand response has historically underperformed nameplate during actual scarcity events in a few well-documented cases. If 2027 or 2028 produces a multi-day heat event and the DR call-rate drops below the accredited value, the planning math falls apart on the day.
• Hyperscaler load-growth deceleration. The base case in most IRPs is for sustained data center load growth. A material slowdown in AI-driven capex or a faster move of compute to lower-cost geographies outside the US would reduce the pressure and lengthen the OEM delivery curve faster than the supply response.

Positioning implications

• Battery storage developers and integrators (Fluence, Tesla Megapack, Powin, NextEra, Vistra Energy storage segment) are in a multi-year demand-pull environment. Pricing power is real but moderated by cell supply.
• Coal-and-old-gas owners with reliability-credit revenue (NRG, Vistra, Talen, AEP Energy, Constellation) carry life-extension optionality that is not always priced into base-case modeling.
• Demand response aggregators (Enel X, CPower, Voltus) are positioned for sustained enrollment growth and improving product economics.
• Utility holding companies in fast-growth load territory with IRP exposure to the gas turbine constraint (Dominion, Southern, Duke, AEP, Xcel) have a public capacity-plan deliverability question that is going to surface in 2027 and 2028 commission filings.
• Hyperscaler-aligned IPPs (Constellation, Vistra, Talen) carry the strongest project-completion advantage given direct procurement scale.

The frame: the gas turbine constraint is binding through 2029, and the planning answer is not one substitute but a four-lever stack that compensates imperfectly. The capacity-market mechanisms are doing the right thing on price but the composition of what clears is shifting in ways that current planning documents understate. The 2027 and 2028 capacity auction cycles will make the composition shift visible. The reliability question for the late-2020s is whether the four levers, together, deliver what the IRPs assumed a single lever (new gas combined-cycle) would.

Sources

• PJM Interconnection, 2025/2026 and 2026/2027 Base Residual Auction results posts.
• MISO, Planning Resource Auction outcomes and Reliability-Must-Run docket filings, 2025 and 2026.
• ERCOT, 2026 State of the Grid and quarterly Operations Reports.
• CAISO, Summer Loads and Resources Assessment, 2025 and 2026.
• NERC, 2026 Summer Reliability Assessment.
• Wood Mackenzie and S&P Global Commodity Insights, North American storage and capacity outlook series, 2026.
• Utility holding-company integrated resource plans, Texas PUC, Arizona Corporation Commission, Georgia PSC, Virginia SCC, North Carolina UC, dockets 2025 and 2026.
• EIA Form 860 retirement schedule data, 2024 and 2025 vintages.
• Hyperscaler power procurement announcements: Microsoft and Constellation TMI restart agreement (September 2024), and subsequent 2025 and 2026 disclosures from Talen, Vistra, and Constellation earnings calls.