Data center flexibility (intro)
How flexibility can accelerate your grid connection
This is an introduction to data center flexibility. The idea is simple. A data center doesn’t run at full power in every hour of the year. Instead, the compute load varies and can be (to some extent) controlled. As I explain below, moving just a small portion of this load around can have major positive impacts for the grid, unlocking a faster grid connection.
Side note: I’m working on a bigger guide with some practical examples around on-site generation, flex and storage - so more on this to come.
Demand flexibility opens up grid capacity
Let’s start with a simple case study from Belgium. Belgium’s transmission system operator Elia publishes grid capacity maps - showing where new generation (like wind and solar) or load (like data centers) can be connected considering the current and future state of the grid.
Looking at Elia’s map for 2027, there’s literally one spot in the country where you can connect a 100 MW data center. And 100 MW isn’t even that big. So if say that one site isn’t ideal (or is already taken), you need to wait a few years for some grid upgrades, which will unlock 9 more sites by 2031.
Fortunately, there’s option number two. If you allow for 5% of demand flexibility - which could mean shifting or reducing the load, or replacing it with on-site generation - there’s suddenly 16 free spots for new industrial loads in 2027, many up to 300 MW.
Of course you wouldn’t want to shut down your data center for 18 days in a year (the equivalent of 5% of all hours). But the yearly average load of data centers is 50%. So there’s clearly space for optimization. Note: some recent studies assume 80% average load, which still leaves some overhead.
Can data centers act as variable grid assets?
Is this flexibility concept technically feasible? In short: yes.
A recent trial by Nebius, Emerald AI, EPRI and National Grid showed that a test AI cluster in London could slash load by 30% in 40 seconds in response to sudden grid stress, while keeping critical jobs. Even better, it could sustain 10-hour load reductions of 10-40% and still deliver 99% performance on highest priority jobs.
Google recently announced hitting 1 GW of demand response in their data centers, combining on-site generation, storage, and the ability to shift and reduce load. With onsite battery storage, you can shape the daily load profile to match the grid’s needs, potentially getting an accelerated connection agreement. In the case of hyperscalers, it’s also possible to move loads across regions if one is say expected to have particularly severe weather or high demand conditions.
Another recent study by Camuus, encoord and Princeton’s ZERO Lab looked at the business case for flexibility and on-site generation+storage. Six sites were analyzed, with below 1% yearly curtailment and up to 25% of temporary compute load flex.
These two solutions shortened the grid connection time from 5-7 years to 2 years, which means 3 or more years of extra revenue. At $8 million per MW a year, this added up to $2-5 billion per site. Removing $1 billion in costs, this still adds up to $1-4 billion net returns.
The added benefit is almost completely removing the need for grid expansion - so a vastly reduced impact of data centers on consumer bills and infrastructure buildout.
Smarter grid connection agreements
What is described above is called a non-firm or flexible grid connection. In summary: you can connect to the grid more quickly, but in return you might need to reduce or shift your load every now and then.
There’s an even easier arrangement called phased grid connection. Often data center developers will apply for a grid connection capacity that matches the target load of the site. But reaching full capacity takes time. Looking at some of the largest projects, they often start running at 10-20% of their nominal power, reaching 100% after a year or more (or never). Data from Ireland and Norway seems to show a similar pattern - with data centers using only around 30% of their contracted grid capacity (blocking the other 70%). With a phased connection, this could be avoided, accelerating the connection process for the first say 30% of the load, and ramping that up to 100% over the course of 1-3 years.
Finally, strategic siting is perhaps the smartest solution of all. I covered that in a recent post. It basically means placing the data center in locations where grid connection is readily available - often next to generation assets. You can do that by combining multiple data layers: grid infrastructure and capacity, power supply options, PPA availability, weather, policy incentives, social support - which is what I’m building with enersite.app. As described in the post linked below, governments are already doing some of that - by directing investments to preferred sites (like windy Scotland or next to nuclear plants in France). It’s good for the grid and the developer.
To summarize: there’s quite a few win-win options available. The best solution is probably a combination of all the above: picking a great site, going for a phased connection, and investing in on-site clean power + storage. Plus allowing for demand response. That way you maximize revenues, cut development times, lower the grid and power system impacts, and stabilize costs (unless you go for gas…).
Thanks for reading!
As I mentioned in the beginning, more is coming on these topics, so if you’re interested, subscribe to my newsletter: