Is AI Driving Up Electricity Prices? | Dirt Report

Why AI requires so much electricity

When someone uses a generative AI service, the processing generally takes place in a data centre rather than on their laptop or phone. These facilities contain servers, graphics processing units, cooling systems and other supporting infrastructure.

The video cites an International Energy Agency estimate that data centres worldwide already consume about 460 terawatt-hours of electricity a year. That is approximately 2% of global electricity use. The estimate discussed in the video suggests this could more than double by 2030, largely because of AI growth.

A typical AI-focused data centre can draw as much electricity as 100,000 households, according to the video. The largest facilities being planned could use considerably more. The transcript also gives an example of a proposed 650-megawatt data centre at Eastern Creek in Sydney, which, at full capacity, could have electricity demand comparable to about 780,000 homes.

Australia’s growing and concentrated demand

The video states that Australian data centres currently consume about 4 terawatt-hours of electricity annually, representing roughly 2% of the main grid’s electricity. Planning expectations cited in the video put this at about 6% by 2030 and potentially 12% by 2050 if data-centre and AI growth continues.

The issue is not only the total amount of electricity. Data centres tend to be concentrated in locations such as Sydney and Melbourne, placing substantial new demand on specific parts of the network. This can require upgraded substations, higher-capacity power lines and other infrastructure to deliver electricity reliably.

Those upgrades form part of network costs. The video says network charges make up almost half of a typical Australian household electricity bill. These charges cover infrastructure such as poles, wires, underground cables, substations, transformers, maintenance and some grid upgrades. If networks expand to meet large new loads, some of those costs can eventually be reflected in electricity bills.

What overseas electricity markets show

The video points to parts of the United States where data-centre expansion is already being linked with higher household costs. It cites analysis suggesting that households in some regions are paying an additional 10 to 27 US dollars per month as utilities invest in new generation and larger transmission networks to serve data centres.

It also refers to analysis finding that wholesale electricity prices in some areas near AI data centres have risen by more than 200% compared with five years earlier. These examples do not establish that the same increases will occur in Australia, but they illustrate how concentrated demand can affect local electricity markets and infrastructure costs.

The water and energy connection

Electricity is only part of a data centre’s resource use. Cooling the equipment can require substantial quantities of water. The video cites an estimate that a 1-megawatt data centre may use up to 25.5 million litres of water each year for cooling, excluding drinking water.

It also refers to projections that planned data centres could make Sydney’s water demand exceed Canberra’s total drinking-water use by 2035. A separate global estimate cited in the video suggests AI-related workloads could consume between 4.2 and 6.6 billion cubic metres of water annually by 2027.

Water use can also have an energy cost. Desalination, pumping and treatment require electricity, creating an indirect connection between data-centre water demand and power demand. The scale of these impacts depends on the cooling systems, water sources and infrastructure used, as well as the location of each facility.

Is AI the main cause of high power bills?

The video distinguishes between AI’s current contribution and its possible future impact. It identifies older fossil-fuel generation, gas setting marginal prices, network upgrades, replacement of ageing infrastructure, retail margins and the slow construction of renewable generation and storage as larger contributors to Australian bills today.

AI data centres add demand to this existing system. If new demand is met with additional coal or gas generation because renewable energy and storage are not available quickly enough, the result could be higher or more volatile prices. If new data centres are supported by additional solar, wind and batteries, the effect may be more manageable.

The transcript also notes that individual prompts are not what places significant pressure on the grid. The larger issue is the global construction and operation of very large AI data centres. Household tools that use AI to analyse smart-meter data, manage batteries or shift consumption to off-peak periods may help individual users save money, but the video says those savings would have only a limited effect on overall grid demand.