Speed to power: solutions to US data centres’ challenges

How best to measure the breakneck pace of US AI and data centre roll-out? Well, there’s hyperscaler capital expenditure, which has increased four-fold since 2023 and is set to exceed US$750 billion this year, more than either the global spend in renewable power or upstream oil and gas.

Hyperscaler electricity consumption is another metric, skipping along at 23% a year so far this decade – way ahead of total US power demand, which did not grow from 2006 to 2024. Perhaps best is the 201 GW of new data centre capacity that US utilities have committed to connect to their grids, which would increase electricity demand by 30% if it’s all built.

Speed to power is the buzz phrase. All hyperscalers want their data centres plugged into reliable power supply as soon as it is ready to run their large language models and satisfy consumers’ growing reliance on them. The problem is that the US electricity industry, regulators and the entire supply chain alike are not yet capable of moving at the pace, or scale, required.

Chris Seiple, Vice Chair Power and Renewables, and I spent last week in New York talking with power sector investors and financiers about the challenges and opportunities ahead.

What are the biggest challenges for the power sector?

Among multiple challenges there are three big ones. First, the timetable for getting a data centre connected to the grid keeps getting longer. It varies by state but in Virginia, a hot spot for data centre location, the wait is 54 months at best. In Ohio, it’s 84 months – that is, no power supply until 2033 for a project going ahead today.

Second, public opposition to data centre development is growing. Lawmakers have called 400 moratoriums on data centre proposals. One-fifth of projects under development have been delayed by a moratorium, some by more than one.

Third, there’s an acute lack of essential components in the supply chain. Gas-fired power plants will dominate data centre power supply and the waitlist for the heavy frame gas turbines has stretched out to 278 weeks – roughly early 2032. Not much use if you want your data centre up and running before 2030 and weren’t early in the queue to order. Wire infrastructure is not much better with waitlists for transformers coming in at 162 weeks and circuit breakers at 137 weeks. Skilled labourers are also emerging as a bottleneck – not an easy problem to solve quickly.

Is it easier to build generation in regulated or unregulated markets? 

Unregulated markets are easier in some ways in that the data centres are free to choose their power supplier and often have more flexibility in developing behind-the-meter generation (BTM) solutions. The main challenge is the disconnect between the planning for large loads and the planning for generation interconnections, and connecting the two.

In short, there are numerous independent decisions being made in unregulated markets that can get in the way of building out the grid in the most cost-effective way and that makes sure supply growth keeps up with demand growth. Regulated markets can mitigate some of the planning uncertainty and grid inefficiency.

What’s the timetable to deliver the power? 

Realistically, there are no solutions that deliver new supply at any scale in 2027 or 2028 beyond what has already been committed. The task for regulators, the electricity industry and hyperscalers is to ensure that supply is connected by the time that demand ramps up beyond 2028. And the focus should be on delivering the 201 GW capacity already committed – adding more capacity to the pipeline before the current logjam gets moving will only hinder progress.

What solutions are there for power supply challenges?

First, BTM projects – which account for 15% of the pipeline by project count and 40% of planned capacity – skip the need for grid connection and the regulatory hurdles that go with it. Most BTM plans will be gas-fired capacity, backed increasingly by new high-discharge batteries. These projects will be expensive, complicated from an engineering perspective, and few data centres view the BTM solutions as a long-term plan, preferring to connect to the grid at some point.

The ‘ERCOT solution’ could be a pragmatic way forward for unregulated markets. The Texas system operator analyses the planned locations and scale of large load demand in the market, then makes assumptions where developers are likely to build generation. It can then make an ‘80/20’ decision on investment in transmission and connection and hope it all works out.

Interruptibility of demand is emerging as another solution. Data centre developers are trying to figure out how dedicated BTM resources, such as generators or batteries, might offer more flexibility to the grid. At the same ime, regulated utilities are working to communicate to data centre developers what type of flexibility will best optimise getting new load connected faster.  Lastly, deregulated market rules are evolving to accommodate flexible interconnect.

Is AI-driven demand changing utilities? 

Yes. For one thing, hyperscalers are currently price-insensitive. That’s a new experience for regulated utilities, which can leverage that price insensitivity to deal with affordability concerns and to find innovative ways to get load connected faster or to pursue new generation technologies, such as advanced geothermal or long-duration energy storage.

After operating in a low-growth bubble for decades, utilities are coming up with solutions for a new breed of customers with large loads in grids that are already constrained.

Make sure you get The Edge

Every week in The Edge, Simon Flowers curates’ unique insight into the hottest topics in the energy and natural resources world. 
Sign up today using the form at the top of the page.