When the grid connection is the bottleneck.

Not for availability — there are already an uninterruptible power supply and a standby generator for that. It is about the grid connection: a data centre needs connection capacity for its peak but draws considerably less on average, and storage closes that difference. In Germany this is precisely the bottleneck — in 2025 German data centres consumed 21.3 billion kilowatt-hours at 2,980 MW of installed IT connection capacity, with 5,000 MW expected for 2030.¹

An uninterruptible power supply bridges seconds to minutes until the standby generator starts; it is built for availability, not for energy. Storage is dimensioned the other way round: plenty of energy, but without the availability chain of a UPS. It can take over parts of a backup concept but does not replace it. What applies in a given case belongs in the discussion with the operator and in their safety concept, not on a website.

The capacity applied for then follows average rather than maximum load. In practice the system covers the peaks out of itself and recharges in the troughs, so that a flatter curve arrives at the connection point. This requires a credible load profile for the site and the grid operator's agreement to that mode of operation — without it, the calculation has no effect. In addition, section 17(2b) EnWG has since February 2025 allowed a flexible connection in which output is curtailed for a few hours a year.³

It follows from the load profile, not from IT capacity. The same four quantities matter as at any industrial site: the gap between peak and average, frequency, duration and the time available to recharge. A data centre's profile is flatter and more predictable than a factory's — which makes design easier, but the saving from pure peak shaving smaller too. How the calculation works in principle is shown by the tool on the industrial page.

Beside it, in its own containers on a hardstanding, with its own fire compartment and transformer station. Installation inside the building is technically possible but sharpens fire protection and smoke extraction considerably, and is rarely sensible at the capacities in question here. The footprint is small measured against the campus; what decides the layout are separation distances and heavy-load access.

Efficiency requirements apply to data centres, among them power usage effectiveness. We will name specific limits and dates only once we have evidenced them from the law as it stands — the legal position here has changed several times. Independently of that: storage does not improve PUE. It improves the connection situation and the cost structure; anyone who conflates the two is selling something other than what they deliver.

Yes, with the capacity left after own requirements. This requires prequalification, a marketing contract and, depending on the product, Grid forming inverters. Two limits apply as at any behind-the-meter site: the site function takes priority, and trading raises the cycle count and with it ageing. As the foundation of an investment decision this does not work; as a contribution to operating costs it does.

Yes, on a smaller scale — and often with one decisive difference: at such sites the grid connection frequently already exists. An edge site next to an existing storage or generation project can share its connection; that is the same co-location logic as with photovoltaics, only with compute load instead of feed-in. It does not shorten the technology, but it shortens the route to the grid — and here that is the longest item.