Published on the 24/07/2025 | Written by Heather Wright

When the chips are down…

A London Economics International report has poo-pooed projected data centre growth – and by extension panic about forecasts of aggregated electricity demand growth reported by the US power industry.

The report, which London Economics says is a ‘sanity check’, says high level forecasts for data centre growth, and the accompanying electricity demands, are ‘not credible’. Its independent research suggests the projections touted about US data centre growth (yes, the report is US based) and the electricity which will be required to power AI, are most likely overinflated, with uncertainty and ‘upward bias’ inherent in data centre electricity demand projections.

“This is simply unlikely to occur.”

Key to the issue is that there simply aren’t enough chips globally to support the projected growth of US data centres.

The power ramifications of data centre build outs particularly in the face of AI demand, and ‘power congestion’ from data centres clustering together, have been a big concern for many power companies globally, leading to a rush to add energy resources – impacting power pricing for many as those rising levels of investment in the network bite.

Exacerbating the issue are AI models, which are typically much more energy-intensive than the more traditional data retrieval, streaming and communications apps that drove data centre growth over the past 20 years. The Electric Power Research Institute said earlier this year that while AI applications are estimated to use 10-20 percent of data centre electricity, that percentage is growing rapidly.

LEI says while typical average data centres in the past had a load of around 5-10 megawatts, hyperscale data centres have loads of 100 megawatts or more, with even a 200 megawatt facility now considered ‘typical’.

In Australia data centres account for around six percent of overall electricity usage, with that figure expected to double by 2030,  and some forecasts tipping it to hit 15 percent under high-demand scenarios.

Two new builds in Western Sydney alone could add more than 2GW of new load – roughly equivalent to two large aluminium smelters.

After more than 15 years of subdued power demand increases, post 2008, thanks to more efficient devices, power usage in Australia has surged back, driven in part by demand from data centres. The Australian Energy Market Operator has forecast up to 1.5 gigawatts of new demand by 2035 in an ‘accelerated data centre scenario’. Industry analysts have forecast much higher increases – up to five gigawatts. (To put that in perspective, a large coal-fired power plant usually has around a gigawatt of capacity.)

In New Zealand, data centre builds are booming as big players look to capitalise on the country’s renewable energy (around 85 percent of New Zealand’s electricity is from renewable sources) to lower their carbon footprint. Government estimates suggest by 2030, data centres will be consuming more than seven percent of the country’s future supply – equal to the amount Tiwai Point aluminium smelter – New Zealand’s largest power user – consumes.

The Australian and New Zealand power needs pale in comparison to the US, where the Department of Energy has estimated an additional 100 gigawatt of new peak capacity will be required by 2030, with half of that required for data centres.

But reports also suggest growth is beginning to outpace capacity, at least in traditional data centre hubs.

The London Economics International report says results of its research – admittedly only into the United States market – indicate that growth in data centre electric load in the US will fall short of the predictions because there will not be enough AI chip capacity to meet demand.

LEI performed an ‘independent outlook’ for global expansion of semiconductor chip supply incorporating ‘rapid but realistic’ growth in chip capacity, as well as increases in chip energy efficiency and computing capacity.

It says AI chip manufacturing capacity has grown at around 6.1 percent over the past decade. If that were to rise ‘significantly’ to an average rate of 10.7 percent per year, it would still only accommodate an incremental 63 gigawatts of data centre related demand globally over the next six years.

LEI found that if demand outlooks from the electricity sector are taken at face value, the US would need 90 percent of incremental global supplies from 2025-2030.

“This is simply unlikely to occur,” LEI says, noting that the United States currently buys less than 50 percent of global semiconductor chips annually, and other countries are also seeing strong growth in data centre development.

“Even under optimistic scenarios for global chip manufacturing, LEI finds that projected data centre demand from just 77 percent of the US power market would require 90 percent of global chip supply through 2030.”

LEI says the potential for overestimating electricity demand from new data centres has important implications, with demand projections driving expansion plans – but if projected data centre loads fail to fully materialise, the cost of the assets are borne by other customers.

Earlier this year Auckland electricity distributor Vector flagged ‘significant uncertainty’ around increasing power demands from hyperscale data centres. It has forecast a 60 percent rise in demand from data centres in the coming decade and says high demand for data centres in Auckland is creating unique challenges for the region.

But in echoes of the LEI report, Vector raised concerns about the uncertainty about the rate at which demand will grow to match the capacity requested.

LEI says data centre developers ‘have incentives’ to duplicate requests for electric service ‘and evidence shows they are doing so’.

“This reflects the variety of locations where a data centre can choose to locate, in the United States and across the world; the incentive to duplicate requests and evidence of duplication of requests; and recent attrition in data centre load based on announcements from such customers.”