Giant data centre loads are whipsawing ordinary energy consumers in the US, driving retail prices upwards amid a once-in-a-generation land-grab to exploit the artificial intelligence boom.

Customers — who feel they are being forced to bear an unfair share of the financial burden of electricity generation and network expansions driven by data centres — are complaining ever more loudly to regulators.

Australian policymakers and poles and wires networks are striving to get ahead of the problem and prevent similar outcomes shaking confidence in the energy market here at a time when rising costs are already a sore point.

The Australian Energy Regulator has cut Jemena’s revenue claim for 2026-2031 sharply back to size in a draft ruling — rejecting $261 million (13.1 per cent) of the northern and western Melbourne distribution network’s $1.989 billion claim.

The AER said Jemena hadn’t justified a projected doubling in demand and related expenditures based on growth in data centre connections and other large loads. Jemena says it only recently received the draft ruling and is yet to respond.

Fatima Bazzi, Ausgrid’s head of customer connections, pictured with a two-way electric vehicle charger at Ausgrid’s Artarmon, NSW depot. - Supplied

Fatima Bazzi, head of customer connections at Ausgrid, says the network for the greater Sydney area through Newcastle and northwest adopted an approach in its tariff plan for 2024-2029 that ensured data centre connection costs were borne by the proponents.

“Say they’re a 200 megawatt data centre — they’ll be paying for that, even if they do not use it,” Bazzi says. “We’ve done it very consciously to minimise or eliminate the impact onto mums and dads.”

It shows how mechanisms for setting retail power prices in Australia differ in key respects to those in the US, where most costs are socialised by utilities at periodic “rate setting” hearings. The rollout of data centres here also lags the US, which is vying for “AI dominance” with China.

“You shouldn’t see the from our regulatory system the kind of increase in network costs and generation costs that appears to be happening in parts of the US,” says Tennant Reed, director of climate and energy policy at the Australian Industry Group.

Still, this is the right time for Australian regulators to ensure their price-setting mechanisms will remain fit-for-purpose as data centres grow more rapidly and consume more and more power, energy expert and entrepreneur Ezra Beeman says.

Timely

Beeman was in charge of connecting large “cost-reflective” customers to Ausgrid’s networks in NSW before he co-founded Energeia, where he remains managing consultant.

“We haven’t added load like that since the smelters. So anything that is not quite right in the way that connections are managed [and] the cost of those connections are managed — it’s all going to blow up with data centres, because they’re massive,” he says.

“I’m hopeful that we can better plan the system so that we’re not doing those negotiations with your back against the wall.”

Data centres in Australia currently consume 3.9 terawatt hours of electricity, or about 2 per cent of grid power. But this is set to grow by a quarter each year, according to modelling by Oxford Economics for the Australian Energy Market Operator (AEMO), which is now studying data centres as a category on its own.

Ausgrid has requests for 10GW of data centre connections, of which 2GW was sufficiently credible to feed into AEMO’s demand projection estimates.

AEMO has received 18GW of “large load connection” requests in its role as Victoria’s grid planner (which will pass to VicGrid on 1 November), but does not expect them all to eventuate. It included 1GW for data centres in Victoria by 2035 in its latest Electricity Statement of Opportunities (ESOO).

AusNet — Victoria’s transmission company — has requests “in the gigawatts”.

The ESOO projects that that aggregate demand from data centres will reach 12TWh by 2030 and 34TWh by 2050, accounting for 12 per cent of National Electricity Market supply. The NEM covers the eastern states and South Australia, where most of Australia’s data centres will be located.

Greater Sydney is leading the charge, and data centres already consume 4 per cent of NSW grid supply, headed for 11 per cent by 2030 and 17.2TWh or 18 per cent of NSW supply by 2050.

That puts Sydney and NSW in the front lines of the battle to ensure surging data centre connections don’t trigger surging retail electricity prices like those hitting US households. Melbourne and Victoria will play catch up after 2030.

‘Data centre alley’

Customers of US utilities where large data centres are concentrated are already complaining loudly to regulators about massive hikes in their power bill, according to a report in Bloomberg Technology.

One customer of Dominion Power in Baltimore — an hour’s drive from Northern Virginia’s ‘Data centre alley’, said her bill had increased 50 per cent in a year; another said he was slugged with an 80 per cent increase over three years. Analysis by Bloomberg found bills catapulted 267 per cent — almost fourfold — over five years in areas nearest to major data centre concentrations. Similar problems have been encountered as far away as Wisconsin and Oregon.

“Doesn’t seem fair, does it?” says Beeman.

In many US states monopoly power system operators purchase generation and distribution assets on behalf of their customers — known as “ratepayers” — and are reimbursed via regulated tariffs.

When many small customers are gradually added to the grid, that system mostly works. But when a large load such as a data centre demands several hundred megawatts to set up in a locality, the system struggles to manage the timing of expenditure and revenue recovery so as to fairly apportion the costs, Beeman says.

In the NEM, transmission and distribution costs are spread among energy customers in a similar way via tariff plans set every five years by the Australian Energy Regulator according to formulae set by the Australian Energy Markets Commission.

Generation assets sit outside these systems, and in theory their investors must take their chances in the NEM’s “spot” wholesale market or strike power purchase agreements with retailers or large energy users to manage their risk.

In practice, though state and federal governments have intervened in the market by setting price floors for a number of years, for example via the 40GW federal Capacity Investment Scheme (CIS) and similar state and territory schemes prior to the CIS.

Only a fraction of the costs of these schemes come back to large customers such as data centres, says Beeman.

The introduction of large new loads to networks should in theory reduce the cost for everyone, because the cost of the assets gets spread over a larger customer load. But in the US the timing mismatch between capital outlays and cost recovery is playing havoc with that equation.

“At the end of the day major new loads should reduce everyone’s bills,” Beeman says. “If not, something is wrong, it’s not working, and we need to look at it.”

Other problems can arise. Data centre developers may demand capacity but not show up to use it, leaving other customers to pay for capacity expansions and new assets such as substations built in anticipation. Or their connection demands may crowd out connections for vital infrastructure such as new housing — a desperate need in Australia.

**Australian way **

The AER has attempted to get ahead of the problem — and succeeded, in Bazzi’s view — by insisting on data centres being charged “cost-reflective tariffs” and bearing the costs of new assets built specially for their grid connection. Its Jemena draft ruling shows it is taking a tough line.

Bazzi gives the example of four new data centres seeking to connect to Ausgrid’s Macquarie Park substation which — unlike much of Ausgrid’s network — doesn’t have spare capacity.

The data centre developers are paying for Ausgrid to build a new substation to accommodate them. They will also pay for the maximum capacity they’ve requested — regardless of how much or how little they actually use as they ramp up — based on a cost-reflective tariff.

They may have to meet “capacity commitment” milestones — such as buying the land or a transformer — as well to keep their place in the connections queue. If they miss these milestones they risk surrendering their place to the next developer in line.

This is a departure from the conventional process for adding new housing estates, where the cost of new substations is gradually spread over the population via tariffs.

“Our regulatory framework doesn’t allow us to pass costs onto other customers,” Bazzi says.

“I’m not saying it’s impossible, but it’s not something within our current rules that allows us to replicate that scenario that happened in the US.”

These practices suggest the risk of ordinary energy customers being left holding the bag for connection and network augmentation costs of abandoned data centre developments is very limited, Reed says.

But we can’t be definitive because we don’t know the scale of investment that’s coming, and the data centre developers may not be as creditworthy as the “hyper scalers” — Google, Microsoft, Amazon and so on — that use them. “Networks will have to look very carefully at the credibility of their counterparties,” Reed says.

And we will have to see what else happens. For example, similarly bubbly projections for demand from green hydrogen production and electric vehicle adoption have proven premature at best, perhaps illusory.

Stock market nerves about heroic valuations for generative AI-fuelled companies such as Nvidia betray scepticism about the AI bubble. Reed says if all the data centre demand projections in NSW materialised, they would absorb between one third and two thirds of current grid capacity. This amount is “oddly similar to the scale back of green hydrogen demand” but “the numbers are bouncing around”.

Data centre developers can also make themselves an asset for the grid rather than a burden by offering flexibility for part of their load. If they insist the grid guarantee the “five nines” (99.999 per cent) reliability they promise their own customers they’ll require major investment from networks, Reed says. But if they can lower their demand when supply is tight they can save networks capital expenditures and help prevent blackouts.

“I think there’s a fair amount of flexible versus inflexible loads in the computing space,” Beeman says. “We have to tap into it better. And I think starting to tell them ‘no’ is probably going to be part of that.”