AI Data Centres Need Big Batteries But Lithium Isn’t Fit-For-Purpose

Support CleanTechnica’s work through a Substack subscription or on Stripe.

By Dr. Thomas Nann, CEO and Co-Founder of Allegro

The biggest constraint facing AI data centre expansion is not generation. It is storage, and not only storage, but the unique way that AI data centre’s use power.

As AI-driven data centres scale, the grid challenge is no longer simply how much electricity they consume. It is how unpredictably they draw it, and whether the system has sufficient buffering to absorb sudden power surges and dips reliably when required.

This context matters when BlackRock chair Larry Fink warns that data centres cannot realistically run on renewables alone in the near term, and that coal and nuclear will remain necessary. At a high level, he is correct. AI data centres will draw unprecedented amounts of power and on the surface it might seem attractive, even necessary, to return to more conventional generation to meet this need.

But the stress point is not renewable energy suitability per se. It is the volatility of the way AI data centres use energy and the absence of sufficient “buffering” capacity capable of reconciling intermittent supply with increasingly intermittent demand.

Traditional large industrial loads, such as aluminium smelters, draw power in continuous, predictable bands. AI data centres do not. Their demand profile is burst-driven. Compute intensity can surge within minutes as millions of users simultaneously query AI platforms or as automated systems across finance, logistics and defence ramp computation in response to events.

Grid operators are already observing a new phenomenon. When major data centres disconnect during disturbances, they do not ramp down gradually like traditional industrial loads. According to Hitachi Energy, they can disconnect abruptly, triggering voltage and frequency deviations that ripple across the system.

In other words, data centres are becoming one of the most volatile loads on the grid.

Dispatchable generation alone does not solve this. What data centres require is buffering — the capacity to absorb surplus renewable energy when it is available and release it instantly and repeatedly when loads spike.

Lithium-ion batteries have dominated early grid deployments for good reasons. They respond quickly and are commercially mature. But lithium-ion systems degrade rapidly under constant high-frequency cycling. Their economics also become progressively weaker as duration extends beyond approximately four to six hours, where capital cost per stored kilowatt-hour could escalate rapidly.

For infrastructure that underpins financial systems, healthcare delivery and national security, lifetime and cycling resilience are not marginal considerations. They are core design constraints.

Other storage technologies, such as flow batteries, address this challenge differently. In a flow battery architecture, energy capacity is determined by electrolyte volume while power capacity is determined by stack size. This allows energy and power to scale independently and enables relentless cycling without the degradation profile typical of lithium-ion systems.

The result is a fundamentally different cost curve over lifetime and duration.

The world’s renewable build-out is accelerating. Large-scale batteries are being installed. Yet most remain short-duration systems. If AI-driven data centre capacity expands as projected — with individual campuses now measured in hundreds of megawatts — storage requirements will increasingly shift from short bursts to multi-hour and repeat cycling scenarios.

This is not an argument against lithium-ion. Lithium-ion has its place in the energy landscape. But lithium-ion alone will not be sufficient for the next phase of (digital) infrastructure.

If storage architecture is underbuilt or mismatched to load volatility, fossil firming becomes the default fallback. Not because renewables are inadequate, but because the buffer layer was incomplete.

We face a structural choice. We can treat AI data centre growth as justification to prolong fossil generation. Or we can invest in storage technologies designed for multi-decade, high-cycle operation that will allow us to harness renewable energy and store it in systems that will be purpose-built to meet AI’s spiky and volatile load profile. With the right technology, AI data centres can help fast-track, not derail, a clean energy future.

Dr. Thomas Nann is the CEO and Co-Founder of Allegro Energy