For most of the history of the data center, the battery lived in a room of its own. Power arrived, passed through a centralized uninterruptible supply, and reached the floor already protected. The rack was a consumer. Whatever happened upstream was the facility's problem, and the hardware inside the rack never needed to know about it.
That arrangement is under pressure, and the reason is not fashion. It is that racks have grown to a size where protecting them from the far end of the room has become an awkward piece of engineering.
What changed about the load
A conventional rack presents a fairly civilized electrical load. It draws a certain amount, varies modestly with utilization, and does so at a scale the distribution system was designed around.
An accelerator-dense rack does not behave that way. It draws far more, and its demand moves. Large numbers of devices working on the same job tend to change state together, which means the rack's draw can swing sharply rather than drifting. The distribution system sees a load that is both much bigger and much less smooth than what it was built for.
Our guide to megawatt rack power delivery covers how distribution itself has had to change to carry that. Storage is the other half of the same problem: the further the stored energy sits from a load like this, the more infrastructure has to be sized to deliver it quickly.
Putting the storage in the rack shortens that path to nearly nothing. The energy is already on the bus it needs to support.
What a BBU actually is
A battery backup unit is a module that occupies rack space alongside the rack's power shelves and feeds the same internal bus they do. It is not a scaled-down version of a room UPS. It is a different design for a different job.
The clearest difference is intent. A centralized system is normally expected to carry the load until a generator has started, synchronized and taken over, which is a process measured in a meaningful number of seconds. A BBU is generally not sized for that. It is sized to cover a short disturbance, to bridge a transfer between sources, or to give the hardware enough time to park work and shut down in an orderly way rather than losing state abruptly.
That narrower job is what makes it practical to put inside a rack at all. Storage sized for a full generator sequence would not fit, and would bring a quantity of stored energy into the data hall that most facilities would not want distributed across every cabinet.
What it does and does not solve
The honest framing is that a BBU solves a power-delivery problem and does not solve a facility problem.
It genuinely helps with transients. A rack with local storage is far less sensitive to brief disturbances on the feed, because the local source can absorb what the upstream system would otherwise have to ride through. That is a real benefit, and it is the main reason the approach exists.
What it does not do is keep the room alive. Compute that survives an interruption still needs cooling, still needs a network, and still needs whatever controls the facility depends on. A rack that rides through an event in a hall whose cooling has stopped has bought a few minutes of hashing, not resilience. Our note on UPS systems and availability design covers the centralized side, and the two are complements rather than alternatives.
It also does not remove the need for the facility to be designed around the rack's actual draw. Local storage smooths the edges. It does not change the average.
The lithium question is the operational one
The part that most affects whether this is straightforward in practice has nothing to do with electrical engineering. It is that distributing lithium cells throughout a data hall is a materially different proposition from concentrating them in one purpose-built room.
A dedicated battery room can be designed around its contents: construction, detection, ventilation and access can all assume that this is where the stored energy is. Once storage is in every cabinet, those assumptions no longer hold anywhere in particular, and the design question moves out into the whole floor.
For a facility this raises practical matters that have to be settled rather than assumed. How cells arriving inside customer equipment are documented and handled. What the detection strategy is in a hall where energy is distributed. How the suppression design treats a failure inside a sealed module, which is not the same problem as a fire in open air. And what the procedure is for moving a module in or out once it is installed.
None of this is exotic, and facilities that host this hardware work through it. The point for a tenant is that it is the facility's answer that governs, not the hardware vendor's datasheet.
What this means if you are placing hardware
If you are shipping racks that arrive with their own storage, the questions worth asking before the equipment leaves are narrow and answerable.
- Will the facility accept it. Not every site will take distributed cells, and the answer is better obtained in writing than inferred.
- What documentation is required for the cells, for transport, and for installation, and who is responsible for producing it.
- How the modules may be handled on site, including whether facility technicians will touch them at all or whether that is restricted to your own people.
- What happens on a fault. What the facility does if a module alarms, who is called, and what the isolation procedure is.
- Whether it changes the power commitment. Local storage affects how a rack behaves on the feed, which is worth confirming against whatever capacity you have contracted.
Where this is heading
The direction of travel is reasonably clear even if the details are not settled. As individual racks take on more of the characteristics of a small power system in their own right, more of what used to be facility infrastructure is arriving pre-integrated with the compute. Distribution is already moving that way, as our piece on 800V DC distribution describes. Storage is following the same path for the same reason.
The consequence for anyone buying colocation is that the boundary between "your equipment" and "the facility's responsibility" is less obvious than it used to be. A rack that contains its own storage has absorbed part of what the building used to provide, and the contract and the operating procedures need to reflect that rather than assume the old split.
If you are planning a deployment that brings its own storage into the hall and want to know how it would be handled, talk to our team. It is a far better conversation to have before the racks are built than after they arrive.
Frequently Asked Questions
What is a battery backup unit?
A battery backup unit, usually shortened to BBU, is an energy-storage module that sits inside the rack itself, typically in the same shelf arrangement as the rack's power supplies. It feeds the rack's internal power bus directly. When upstream power is interrupted, it carries the rack for a short period rather than for a full generator-start sequence, which is what distinguishes it from a centralized uninterruptible power supply.
Does a BBU replace a central UPS?
Usually not. The two are designed for different jobs. A BBU covers the brief gap between a supply disturbance and either a transfer to an alternative source or a controlled shutdown, and it does so very close to the load. A centralized system protects everything in the room, including cooling controls, network equipment and anything else whose loss would make the surviving compute useless. Facilities that adopt BBUs generally still maintain central protection, often with a different design brief.
Why did AI racks drive this change?
Because the load profile changed. A rack drawing tens of kilowatts with a relatively steady demand is a different protection problem from one drawing far more, with sharp swings as large numbers of accelerators move together between idle and full computation. Putting storage immediately adjacent to that load shortens the electrical path it has to be delivered over, which matters more as the current involved grows.
What does a colocation tenant need to confirm?
Whether the facility will accept racks containing their own lithium cells, what documentation it requires for them, how they must be transported and handled on site, and how the facility's fire detection and suppression design treats them. These are the questions that decide whether hardware can be installed, and they are far cheaper to settle before the equipment ships.
A Real Energisation Date, Not a Brochure
Rax operates where transmission investment has run ahead of load, and we are direct about staged capacity and real dates. Talk to our team about sizing a deployment against a timeline.
Contact Rax Data & Energy