Catalyst N° 021 of 125 28 Mar 2024
The electricity gauntlet has arrived
with Andy Lubershane, partner and head of research, Energy Impact Partners
In this note
The question
Everyone has suddenly woken up to the fact that electricity demand is outrunning our ability to supply it. Now that the problem is mainstream, what actually happens next, and is the alarm overdone?
The answer
Not overdone. If anything the opposite, because the public conversation had absorbed only half the problem. The near-term consequence is uncomfortable for anyone hoping this accelerates decarbonization: the fastest available answers are keeping coal plants open past their retirement dates and building a lot of new natural gas. Every clean alternative is a 2030s solution to a 2020s problem.
03The argument
The metaphor is a gauntlet, the medieval kind where you run between two columns of people trying to hit you. One column is demand, rising fast after roughly two decades of essentially flat electricity consumption across North America and Europe. The other is supply, meaning both the difficulty of building new generation and the separate difficulty of delivering it through transmission and distribution. Utilities are the ones running down the middle.
The critical point is that AI did not create this. Interconnection queues were already long and transmission was already nearly impossible to build; meanwhile EVs and electrified industry were already turning demand upward. Without AI this was a five-to-ten year trend that specialists worried about. Data centers compressed it into an immediate crisis, because they are enormous and because they are price-insensitive but acutely time-sensitive. A hyperscaler will pay almost anything for power delivered soon, and frequently still cannot get it.
Then comes the part that matters most and gets the least attention. When a utility has to serve a large new load in the next two to five years, its first moves are to postpone coal retirements and to build gas. Not because anyone prefers that, but because these new loads need capacity rather than just generation. A data center runs around the clock, so it adds to peak, and meeting peak requires something firm and dispatchable. Gas is the only option that is simultaneously gigawatt-scale, proven, relatively cheap, land-efficient and not buried in siting and permitting problems.
Which sets up the timing mismatch that is the real heart of the episode. Every clean answer is real and none of them arrives soon enough. Nuclear has genuine consensus behind it as a necessary part of the future, but not this decade, and the first projects will be expensive because cost curves require building the same design repeatedly rather than the dozens of competing designs currently queued for approval. Geothermal could in principle work almost anywhere if you could drill deep enough, but the deepest hole ever drilled is around 12 kilometers and getting there economically is another matter, so nearer-term plays target medium depths. Multi-day storage is progressing but utilities want gigawatts of it by the early 2030s, not now. Grid-enhancing technologies are the closest thing to an immediate lever, and even those underdeliver relative to how elegant they sound on paper.
04What you need to know first
- Capacity versus generation
- Generation is energy over time; capacity is the ability to deliver power at the moment of peak demand. Renewables are cheap generation but weak capacity. Almost every “why not just build solar” objection in this space dies on this distinction.
- Clean firm
- Generation that is both low-carbon and dispatchable on demand: nuclear, geothermal, hydrogen, fossil plants with carbon capture. The scarce category, and the one the gauntlet makes valuable.
- Interconnection queue
- The multi-year waiting line to connect a new project to the grid. A major reason supply cannot respond quickly to price signals.
- Dynamic line rating
- Transmission lines are rated for worst-case heat and wind conditions that rarely occur, so most of the time they could safely carry more. Sensors allow real-time re-rating, squeezing more out of existing wires.
05Details worth keeping
- The evidence that this went mainstream: NERC’s December 2023 risk assessment reclassified many regions from low reliability risk to elevated or high; utilities including Georgia Power filed amended resource plans within a year of the prior one; the major national papers each ran the story within about three months; and share prices for owners of clean firm capacity ran up sharply.
- Northern Virginia and Dominion are the saturated frontier case, with Phoenix and others following as secondary regions.
- Concrete coal-delay examples given: Evergy in Kansas, partly attributed to a new Panasonic battery factory, and First Energy.
- Two strategies for future-proofing gas plants built now. Either prepare for carbon capture, which ranges from merely siting near viable sequestration up to designing for retrofit; or plan for the plant’s capacity factor to decline over decades, keeping value through hydrogen firing or by building small distributed units at the grid edge that double as resilience assets.
- The demand-side story was well covered; the supply-side story was not. Renewable PPA prices had been rising and interconnection was getting harder, and both hosts note that article had yet to be written.
06Claims worth citing
All figures as of March 2024 and now roughly two and a half years old. In a fast-moving area this is the section most likely to be stale; verify before using any of it.
- Roughly 20 years of flat aggregate electricity demand in North America and most of Europe preceded the current growth. Lubershane
- Data center scale progression: tens of megawatts, to hundreds, to early talk of gigawatt-scale facilities, described as nuclear-plant-level demand.
- The US had a few gigawatts of geothermal generation, almost all of it in rare locations where heat and water naturally meet near the surface. Lubershane
- Deepest hole ever drilled roughly 12 kilometers, against roughly 10 kilometers needed to make geothermal viable nearly anywhere. Lubershane
- Conservatively rated transmission lines may be genuinely constrained only about 10% of the time. Lubershane
- AWS purchased a data center sited behind the meter at a Talen nuclear plant; Microsoft had hired a head of nuclear. Both cited as signals of hyperscalers paying premiums for clean firm power.
07Where it’s contested
- Lubershane is bearish on AI and bullish on the gauntlet at the same time. He explicitly takes the under on how impactful AI will be over five years and allows that data center development may be in an enthusiasm bubble. He argues the gauntlet is underblown anyway, because electrification of vehicles and heating queues up behind it. Worth carrying, because it shows the load-growth case does not actually depend on AI optimism.
- The price forecast is explicitly disclaimed. He says he is not in the electricity price forecasting business and reasons only from supply and demand. Kann adds a real counterweight: price-insensitive large customers may absorb their own interconnection and capacity costs rather than socializing them, which would blunt the effect on everyone else’s rates.
- Grid-enhancing technologies sound better than they test. Early pilots revealed additional constraints that make the benefits harder to realize than the concept suggests.
- The strongest counterargument, raised by Kann as devil’s advocate, is that data center energy efficiency has improved dramatically before and could again, slowing load growth; paired with tepid EV sales and possible IRA rollback affecting manufacturing. Lubershane considers IRA repeal politically difficult given how much of the investment landed in red states.