Rising data-centre expansion and prolonged turbine lead times are pushing the US power system towards a critical shortage, prompting policy shifts and flexible solutions amid record-high capacity prices and supply shortfalls.
The strain on the US power system is no longer a theoretical risk; it is now showing up in market outcomes and utility planning. PJM Interconnection’s July capacity auction cleared at the regulator-approved ceiling for a third straight year, yet still left the grid operator short of its reliability target and attracted very little new supply. At the same time, large data-centre developers are still seeking vast amounts of electricity, creating a demand profile that is rising faster than the equipment base can expand. According to Goldman Sachs, U.S. data-centre load could rise sharply by 2027, while turbine makers are already booked many years ahead.
That imbalance matters because the bottleneck is industrial, not financial. GE Vernova said on its July earnings call that it is already taking reservations for 2031 delivery, while Siemens Energy reported multi-year lead times and Mitsubishi Heavy Industries has also described a backlog stretching into the latter part of the decade. The companies’ reported order books are not directly comparable, but they point in the same direction: global demand for large gas turbines is outstripping factory output, with one industry estimate putting annual manufacturing capacity well below recent order volumes.
PJM’s auction results underline how quickly that shortage is affecting the grid. The 2028/2029 delivery-year auction cleared at $325 per megawatt-day, the top level allowed under the current rules, and still fell 6,831 MW short of PJM’s own reliability requirement. Only 525 MW of new generation and uprates cleared the market, a thin response to a system facing rising peak demand and a heavy pipeline of large-load requests. PJM has since proposed measures that would force some new data centres to bring their own power or face curtailment during shortages.
The economics of building new gas generation are changing quickly. BloombergNEF has tracked a steep rise in combined-cycle project costs, while other industry reviews show that plants due later this decade are being priced far above those scheduled to come online earlier. The fuel price itself is not the issue. The problem is the scarcity of castings, skilled labour and production slots, which has extended lead times and pushed turbine prices higher even as Henry Hub gas remains comparatively subdued.
There is, however, a growing counterpoint to the most aggressive data-centre forecasts. Exelon recently cut its own high-probability load outlook sharply and said it was filtering out speculative projects, while Texas ordered an audit of its interconnection queue after requests ballooned to levels far above the state’s record peak demand. Goldman Sachs itself expects only part of the announced data-centre build-out to arrive on time. That suggests some of today’s turbine demand may never become firm orders, even if the queue remains crowded.
For now, the practical conclusion is simple: the near-term answer to AI power demand is not a rapid wave of new gas plants. It is a patchwork of existing assets running harder, faster-to-install alternatives such as reciprocating engines, and policy decisions by state regulators and grid operators over who gets connected first. If the demand proves real, the equipment will arrive too late. If it proves overstated, the cancellations will come after capacity prices have already reset higher. Either way, the constraint is now visible.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
