Is Lead-Acid Really Dead in AI Data Centres? Reading the 2026 Backup Power Debate

Scroll through this year's data centre coverage and you will run into two headlines that flatly contradict each other.

One says lithium won, cleanly and completely: AI server builders now treat battery backup units as standard kit, Australian data centres have more or less stopped installing lead-acid, and GPU clusters north of 100 kW per rack have made the lead-acid footprint impossible to defend.

The other, in exactly the opposite tone: EnerSys said out loud in April 2026 that plenty of data centres still specify lead-acid UPS batteries, and listed why - recyclability, predictable capital cost, maintenance practice that everyone already knows how to do.

Our first reaction, honestly, was: which of these two groups is making things up?

Neither, as it turns out. They are quoting real evidence from real sites. They just are not talking about the same purchase at all.

What actually moved is the architecture, not the chemistry

The most honest thing you can say about 2026 is that backup power is relocating. It is not a straight swap of one chemistry for another.

Classically, backup sat downstream: one big centralised UPS, a dedicated battery room, feeding the whole hall, sized for ten to fifteen minutes of ride-through until the generators picked up the load. Three things broke that model in AI halls:

• Rack density. Old halls ran 6–8 kW a rack. AI halls casually run 30–40 kW, and GPU clusters get quoted above 100 kW. At that density, pushing power from a central room stops being a battery question and becomes a cable-and-copper question.
• How often you actually discharge. AI loads produce sharp, repetitive transients. A battery that needs twelve to twenty-four hours to recover after a discharge is a bad match for a site where the grid misbehaves every other week.
• What a square metre is worth. When every square metre of floor could be hosting revenue-earning racks instead, the battery room has to justify the rent.

The industry's answer is distributed BBU: smaller lithium packs in or beside the rack, supercapacitors covering the sub-second gap, diesel for anything long. That is a real architectural move - and lithium fits it better, because of energy density and because monitoring comes built in.

Here is the bit the headlines leave out, though: most of the world's installed data centre estate is not an AI hall. Somewhere around the ordinary enterprise machine room - the one with a ten-year-old UPS and a maintenance contract - "lead-acid is dead" stops being useful advice and starts being expensive advice.

That 55% number is about visibility, not chemistry

You have seen this statistic. Power interruptions carry the largest share of data centre outages, around 37% per Ponemon research, and roughly 55% of unexpected failures involve ordinary lead-acid batteries.

Real number. Routinely misread.

A VRLA string ships without any management system. Unless somebody instruments it, nobody has a live view of individual block health. You have a string voltage - and a string voltage looks perfectly healthy right up until the afternoon it isn't.

Now compare. Every lithium pack arrives with a BMS reporting state of charge, state of health and cell-level deviation, standard, no extra charge. Lithium's advantage here is not better electrochemistry. It is that monitoring was never optional. It came in the box.

That distinction is worth money, because instrumenting a lead-acid string costs a small fraction of the price gap between the two systems. A facility that fights lead-acid reliability problems and has no per-block monitoring is almost always running a blind battery - and then blaming the chemistry for what is really a specification gap.

Put it bluntly: if you cannot see it, you cannot maintain it. And you will blame whatever is cheapest to blame.

Temperature is what actually kills lead-acid

The rule is old, and it does not negotiate: run a battery 10 °C above 25 °C on a sustained basis and you roughly halve its service life. That holds across chemistries. What differs sharply is the tolerance window.

Nearly every "our lead-acid died early" story we have traced ended in the same place: a battery room running above 30 °C, and maintainers recording the average rather than the worst case at the top of a rack. That gap between average and hottest point is where three years of life quietly evaporates.

Before you write off the chemistry, put a logger on the hottest spot and leave it for two weeks. It is a twenty-dollar experiment that regularly saves a five-figure decision.

The sizing mistake that costs the most money

This one is our favourite, mostly because it is completely invisible in a spreadsheet.

Batteries are sold on Ah at a stated hour rate - C10 or C20. A UPS pulling ten minutes of runtime does not get anything like the nameplate capacity. At high discharge rates usable capacity falls off a cliff: for a short 10-minute discharge, available energy can be roughly half to two-thirds of the C10 figure, depending on design.

So when someone sizes a string off the Ah number, they are buying a battery that looks right on paper and under-delivers the one time it matters.

Ask for the constant-power discharge table, not the Ah rating. Every serious manufacturer publishes one - watts per cell or per block, plotted against minutes to end voltage. Size from that table, then add an ageing margin so the string still delivers its minutes at end of life, not just the week it was commissioned.

Here is the pleasant surprise: a string sized properly from the 10-minute power table often comes out smaller and cheaper than one guessed from Ah - and it will still be keeping its promise in year four. Correct maths is cheaper. We see this almost every time.

Run this ten-year comparison yourself

Below is illustrative, for a hypothetical ~1 MW IT load with about ten minutes of ride-through. Treat it as scaffolding and drop your own quotations in.

• Lead-acid: cheapest to buy, no contest. But at a typical 3–5 year service life you are replacing it two or three times across a decade. Every cycle brings labour, disposal, and a live-work risk during change-out nobody enjoys signing off on. It also wants tight temperature control, which is its own cooling load.
• Lithium: materially more expensive up front - commonly quoted around two to two-and-a-half times the battery capital - but generally no mid-life replacement, dramatically less floor space, a wider temperature window that lets you relax cooling, and recharge in one to two hours.

Where does it cross over? Pretty predictably, actually. Lithium takes the ten-year total when floor space is expensive, ambient control is poor, maintenance access is awkward, or the site sees frequent discharge events. Lead-acid takes it when capital is tight, the room is properly air-conditioned, maintenance is disciplined and instrumented, and the duty is genuinely standby - a battery that may never be called upon in its whole life.

The five questions we actually ask

Skip "which chemistry is better". Answer these in order instead:

1. What is your realistic worst-case battery room temperature? If you cannot hold it under about 27 °C all year, lean hard toward lithium or derate the lead-acid explicitly on paper.
2. How often do you really discharge? Unstable grids punish slow recharge, and they punish it repeatedly.
3. Is your floor space earning money? At AI density, the answer is nearly always yes.
4. Do you have people doing quarterly inspections - and per-block monitoring? If not, buy the chemistry that reports on itself.
5. Is the constraint capital or operating budget? They point in opposite directions. Pretending they point the same way is exactly how bad specifications get written.

And when the answers genuinely split - high density, tight capital - a hybrid is a legitimate answer, not a cop-out: lithium at the rack for the highest-value AI load, conventional VRLA strings still guarding the general-purpose halls.

Two regulations that will change your number

Specify lithium and you increasingly inherit dedicated battery-room requirements. In China, DL/T 5892-2024 calls for separate battery rooms and automatic fire suppression for lithium installations, and the GB 50174 revision is widely expected to carry comparable expectations into data centre practice. Build that civil cost into your comparison before you decide - it is not a rounding error, and it can flip the answer.

Lead-acid's constraint is a different animal entirely: lead is one of the most successfully recycled industrial materials we have, with collection and recovery above 95% in mature markets and new batteries typically carrying over 80% recycled content. If your organisation reports on circularity, that is not a marketing line - it is a real procurement advantage that shows up in the ESG section.

Where we land

We build both lines, which means we have no reason whatsoever to steer you toward either one. What we do have a reason to do is get the specification right, because a lead-acid string sold into a hot, unmonitored room becomes a warranty conversation that nobody on either side wants to have in year three. We would rather lose the order than have that call.

So send us four things: your load profile, the runtime you actually need, worst-case ambient temperature, and whatever space constraints you are up against. We will size both chemistries from the constant-power tables and put ten-year costs next to each other - not just a unit price, which on its own tells you almost nothing.