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

The rapid expansion of AI-driven data centres presents a significant challenge to electricity grids, not primarily due to the volume of power consumed but because of the unpredictable and highly volatile nature of their demand. Unlike traditional industrial users that draw power steadily, AI data centres experience sudden surges in compute activity, causing abrupt spikes and drops in electricity use. This erratic consumption pattern can destabilise grid frequency and voltage, as data centres may disconnect instantaneously during disturbances, creating ripple effects across the system. This volatility necessitates a robust energy buffering solution capable of absorbing excess renewable energy and releasing it swiftly to meet sudden demand peaks. While lithium-ion batteries have been the dominant technology for grid storage due to their fast response times and commercial maturity, they are ill-suited for the intense, high-frequency cycling demanded by AI data centres. Their performance degrades quickly under such conditions, and costs escalate sharply for storage durations beyond four to six hours, making them economically and technically inadequate for this emerging application. Alternative storage technologies, such as flow batteries, offer promising advantages by decoupling energy capacity from power capacity, allowing for scalable and durable operation without the rapid degradation seen in lithium-ion systems. These batteries can handle relentless cycling over extended periods, making them a more suitable buffer for the spiky load profiles of AI data centres. As renewable energy deployment accelerates and data centre capacities grow into the hundreds of megawatts, the need for multi-hour, resilient storage solutions becomes increasingly critical. The choice facing energy planners and policymakers is clear: either continue relying on fossil fuel generation to firm up the grid in the face of AI-driven volatility or invest in advanced storage technologies designed for long-term, high-cycle performance. The latter approach would enable AI data centres to coexist with a clean energy future, leveraging renewable power more effectively rather than undermining it. Properly matched storage infrastructure could transform AI data centres from a grid liability into a catalyst for accelerating the transition to sustainable energy systems.