Battery storage for AI data centres is not a template exercise
Energy-Storage.news published a useful piece on April 21 asking how developers should design battery storage for AI data-centre co-location. The headline answer from the panel was “it depends.” That is true. It is also t...
Energy-Storage.news published a useful piece on April 21 asking how developers should design battery storage for AI data-centre co-location. The headline answer from the panel was “it depends.” That is true. It is also the point many decision-makers still underestimate. Independent Solar Consultants works on live high-load projects, including a 60 MW data-centre engagement and another at 30 MW scale, and what we are seeing on the ground is that battery storage for AI data centres is only one part of a wider power-strategy problem.
Independent Solar Consultants is an independent solar consultancy and broader energy-strategy partner. We do not sell kit. We interrogate whether the site, the load, the control philosophy and the grid route actually make sense. That matters commercially because many schemes now being described as “AI data centres” do not behave the same way operationally, and if you simplify that point too early, you can embed the wrong infrastructure strategy into the project from the outset.
The conference panel got one thing absolutely right. Different data-centre use cases need different battery responses. Prevalon’s Frank Rodriguez drew a line between analytics and content-management sites on one side and AI and large language model training sites on the other, saying the latter demand millisecond-level power delivery and tighter focus on frequency and load management. BrightNight’s Nadim Kanan, meanwhile, described customer cases where a four-hour system is acceptable and others where duration can be shorter if the role is mainly to smooth peaky loads. Kehua’s Chris Liu pointed to architectures where supercapacitors handle front-end peaks and the battery handles energy shifting and frequency support, making two-hour duration sufficient in some configurations.
That is why the phrase “battery storage for AI data centres” can mislead as much as it helps. It sounds specific, but in practice it often hides three different questions. First, what is the real duty of the battery? Second, what is the real shape of the site load? Third, what power pathway is the project actually relying on in its first operating years?
Why battery storage for AI data centres depends on the load, not the label
Not every site marketed as an AI data centre is an AI training facility. That distinction matters. A conventional high-load digital facility with stable operating characteristics is not the same as a training-heavy environment with abrupt swings, fast ramp requirements and different tolerance for instability. If the battery is there to support UPS-style resilience, that drives one design response. If it is there to shave peaks, support grid interaction, or bridge power availability before a fuller interconnection pathway opens up, that drives another.
This is not academic. The Energy-Storage.news panel spoke repeatedly about interoperability, state-of-charge buffers, latency, degradation and how the battery is controlled alongside other infrastructure. Those are not details. They are the design. A battery can be too large and sit underused, too small and miss revenue or reliability outcomes, or technically “correct” on paper while commercially wrong for the site.
A standalone answer worth stating clearly is this: battery storage for AI data centres cannot be sized properly until the real operational duty is defined. That should not be controversial, but plenty of schemes still move too quickly toward equipment conversations before the site has earned that certainty.
What the market is missing about co-located battery storage
The market likes the story of co-location because it sounds like a direct answer to the AI power challenge. Storage on site. Solar where possible. Gas where necessary. Optional grid participation. It sounds joined up. Sometimes it is. Sometimes it is just a collection of technologies wearing the same hard hat.
What gets missed is that co-location only creates value when the controls philosophy, contractual structure, degradation assumptions and redundancy logic are aligned with the actual power objective. The conference panel addressed this from several angles. On degradation, Raafe Khan raised whether data-centre operating profiles sit outside standard OEM assumptions. Rodriguez described a recent project where the battery acts as a shock absorber for huge load swings, requiring a multi-layer control strategy to preserve long-term performance. BrightNight said the market is still not seeing bespoke battery warranty structures built specifically around data-centre duties.
That point matters commercially. If your long-term service assumptions are borrowed from another application without enough scrutiny, your opex model may be cleaner than your real-life battery behaviour. The issue is not whether the battery “works.” The issue is whether the whole commercial case remains sound after years of actual site duty.
Another answer worth stating plainly is this: a co-located battery is not the strategy; it is one asset inside the strategy.
That wider strategy now sits against a U.S. market where interconnection remains slow. LBNL says around 10,300 projects representing about 1,400 GW of generation and 890 GW of storage were still actively seeking grid connection at the end of 2024, and median queue duration for the projects that do get built has stretched to more than four years in the available data. FERC also said on April 16 that it plans to act by June 2026 on large-load interconnection reforms.
Those numbers do not tell you exactly what your site should do, but they do tell you this: a developer waiting for the grid to become simple is not reading the room.
What experience shows on live high-load projects
On live projects, the first mistake is often false certainty. A client or partner says “AI,” and suddenly the conversation jumps to massive power, oversized storage and a future-facing narrative. We pull that back. We want to know what the processors are doing, how the facility is phased, what the actual ramp and redundancy logic is, whether the grid timeline is real, how the planning pathway interacts with temporary and permanent infrastructure, and whether the site has been honest about where resilience really sits.
We are currently involved in large data-centre work at 60 MW and 30 MW scale. The lesson is not that every large-load project needs the same answer. The lesson is that they almost never do. On one anonymised scenario, a development team may think it needs a heavy storage-led response because the language around AI has pushed them in that direction. But once the real use case, phasing, control architecture and connection risk are unpacked, the right answer may be staged infrastructure, different buffering logic, or deeper front-end investigation before battery duration is fixed. That kind of correction is valuable because it happens before expensive commitments harden.
A third standalone answer worth stating is this: the fastest way to make battery storage look sensible is to ask the wrong first question. If you begin with “how many megawatt-hours do we need?” instead of “what problem is the battery solving here?”, you are already drifting.
What does this mean for businesses like mine?
If you are developing or funding a large data-centre site, this means energy strategy needs to sit in the first room, not the second. It should be there alongside land, planning, utilities, civil design and financing, because each of those areas is now affected by how the site will actually secure and manage power.
For infrastructure investors, the implication is equally direct. Battery storage may improve resilience, speed-to-power and optionality, but only if it is tied to a credible operating model. For delivery teams, it means you need clarity on whether the battery is being asked to smooth, bridge, back up, optimise, arbitrage, or all of the above. For planners and development managers, it means the grid narrative and the behind-the-meter narrative must join up early enough to inform the real build pathway.
A recent Reuters report on NiSource’s agreements with Alphabet and Amazon shows just how bespoke these power arrangements are becoming. That is not a side story. It is evidence that the commercial structure around large loads is changing with the technical structure.
Typical approach versus an independent consultancy approach
| Factor | Typical Approach | ISC Approach |
|---|---|---|
| Site classification | Treat “AI” as one category | Separate true AI duty from generic high-load digital use |
| Battery sizing | Start with MW/MWh assumption | Start with the problem, duty cycle and controls logic |
| Grid pathway | Assume grid timing will sort itself out | Stress-test queue, bridge-power and phased energisation options |
| Redundancy | Borrow generic uptime language | Define what resilience means for this site and this customer |
| Revenue stack | Add merchant optionality as a bonus | Test whether market participation conflicts with core duty |
| Procurement timing | Rush into equipment framing | Hold design until load and power pathway are proven |
This is why we describe ourselves as the engineer on your side of the table. Independent Solar Consultants exists to protect investments, not move boxes.
This is not only a U.S. story
The article itself is U.S.-based, but the underlying tension is global. In the U.S., long interconnection queues and large-load reforms are driving behind-the-meter and hybrid approaches. In New Jersey, Iron Mountain’s Edison site shows a more distributed version of the principle, pairing a 23 MWh battery with a 7.2 MW rooftop solar system and advanced controls to support reliability and local grid flexibility.
In the UK, the pressure is showing up differently but with the same core theme: power availability is shaping project logic. Recent reporting on a proposed Buckinghamshire AI data-centre scheme highlighted the extent to which on-site generation is now being considered as a response to power constraints.
The pattern is wider than one region. Large-load developments are forcing energy strategy to become part of site logic, not an afterthought.
When ISC would say this is right, wrong, or premature
We would say a co-located battery strategy is right when the site has a properly defined duty, a clear control philosophy, a realistic grid or bridge-power pathway, and a resilience objective the battery can actually serve.
We would say it is wrong when the storage design is being used to cover for poor site classification, weak operational thinking or unrealistic connection assumptions.
We would say it is premature when the team is asking for fixed infrastructure answers before it has resolved what the facility really is, how it will phase, and what the energy system is meant to do in practice.
The market is moving fast. That is obvious. What matters more is that the wrong early assumptions are becoming more expensive. The Energy-Storage.news panel was right to resist one-size-fits-all answers. Our view is that developers should take that warning even more seriously than the headline implies.
Battery storage for AI data centres is a moving target because the load, the grid, the commercial model and the uptime expectations are all moving at once. The developers who do well in this environment will not be the ones chasing a fashionable architecture. They will be the ones disciplined enough to classify the problem properly before they build the answer.
If you are looking at a high-load site and want an independent sense-check on the energy strategy, that is exactly where Independent Solar Consultants adds value. We do not need to sell you a product to tell you the truth.
SOURCE LIST
Original article:
FERC:
https://www.ferc.gov/news-events/news/ferc-act-large-load-interconnection-docket-june-2026
LBNL queues:
Reuters - NiSource / Alphabet / Amazon:
Renewable Energy World - Iron Mountain:
Reuters - U.S. power demand:
Found this helpful?
Share this post with others who might benefit.
Related Articles
Commercial Energy Price Cap: Why Businesses Still Need Stronger Protection
Ofgem’s latest guidance on the **commercial energy price cap** makes one thing very clear: businesses, charities and many other organisations are not protected in the same way as domestic energy customers. Independent So...
Commercial Battery Storage for Existing Solar: Why the UK Battery Boom Needs Better Thinking
The BBC has reported on the UK pouring hundreds of millions into battery storage because the grid can no longer cope cleanly with modern demand patterns. Independent Solar Consultants is an independent solar and energy c...
Commercial solar consultants analysing UK energy costs and solar strategy for business
The latest BBC coverage around energy costs and the government’s attempt to weaken the gas-linked electricity price tells an important story, but not the whole one. Independent Solar Consultants exists for exactly this k...