Could AI Data Centres Support New Zealand's Grid Instead of Straining It?

The artificial intelligence boom is filling the world with data centres, and the argument over the power, water and land they consume has now reached Aotearoa New Zealand. Writing in The Conversation, energy systems researchers Alan Brent of Te Herenga Waka Victoria University of Wellington and Andrew Crossland of Durham University pose a question that cuts against the usual framing. Rather than asking whether a data centre is good or bad for the grid, they ask how one might be built and run so that it helps.

The occasion is concrete. Singapore-based Datagrid has secured approval for a NZ$3.5 billion, 280 megawatt AI data centre near Invercargill, scheduled to open in 2028. When it does, the authors note, it will become the country’s second-largest electricity user after the Tiwai Point aluminium smelter, and is expected to more than double New Zealand’s data processing capacity. The project has drawn public unease in Southland, and the Green Party has proposed a one-year pause on new AI data centres while stronger rules are written.

Brent and Crossland do not dismiss those concerns. Their point is that the outcome depends on design. “If they can adjust when and how they use power,” they write, data centres “could become increasingly valuable as New Zealand adds more wind and solar generation to its electricity system.” Much of the computing work involved, they explain, is not time-critical: AI model training and large-scale data processing can often be scheduled for periods when renewable electricity is plentiful, wholesale prices are low, or the grid is under less strain. A centre that shifts its load that way absorbs surplus wind and solar that might otherwise be wasted, rather than adding to evening demand peaks.

The financial case rests on figures the authors draw from New Zealand’s own system operators. Transpower estimates that every gigawatt reduction in peak demand could avoid around NZ$1.5 billion in system costs, and the Energy Efficiency and Conservation Authority has identified flexible electricity demand as one of the cheapest ways to cut future network investment. On the generation side, a data centre’s large and predictable appetite for power can underwrite new wind and solar through long-term purchase agreements. The authors point to Datagrid’s own arrangement: its power purchase agreement with Mercury covers 140 megawatts of the 280 megawatt demand, which they suggest leaves the remaining half open to underpin community-owned renewable projects across Southland.

There is a further role beyond consumption. Data centres already contain batteries, uninterruptible power supply systems and backup generation. With the right market rules, the authors argue, those assets could provide fast frequency response and reserve capacity, and the centres could temporarily reduce demand during network congestion. The caveat is honest: many existing systems are technically capable of this, but market arrangements would need to evolve before they could take part.

The significance is that the same infrastructure can go either way. Poorly planned, the authors write, these projects could increase peak demand, force costly transmission upgrades and compete with other users for renewable electricity. Well designed, they could help integrate renewables, support community energy and improve grid resilience. The relevant question for regulators, Brent and Crossland conclude, is not whether data centres are good or bad, but how to set the planning rules, market incentives and operating arrangements that decide which of the two they become. With five more South Island data centres reported to be testing the grid, that choice is not hypothetical.

The full piece, including the authors’ disclosure statements and the sources behind each figure, is worth reading in The Conversation.