Australia Says Its Datacentre Backlash Is a US Import as Buildout Accelerates

A Guardian Australia report published on 26 September laid out the argument. It described a national debate that has moved from scattered complaints to calls for a moratorium, community protests and three government inquiries. State governments, keen for economic growth, have in some cases overridden local council approval processes.

The industry's case rests on scale. Belinda Dennett, head of the lobby group Data Centres Australia, said in a March speech that the US has 5,400 datacentres with 53.7 gigawatts of total capacity, against 252 facilities and 1.4 gigawatts in Australia. The Guardian noted that measurements of the number of US datacentres vary. It also noted that Dennett's comparison omitted a per-capita calculation: Australia, with a population of 28 million, has just under one datacentre per 100,000 people, against 1.5 per 100,000 in the US, population 342.7 million.

Small base, fast build

The base is small, but the pipeline is not. Several large datacentres of 1 gigawatt or more are in planning, according to the Guardian. Trending videos on TikTok and Instagram have highlighted a 2.4 gigawatt centre planned for outer Melbourne and a 2 gigawatt gas-powered centre in the Northern Territory. Rob Nicholls, a professor and senior research associate in digital policy at the University of Sydney, told the paper the NT project is equivalent to the annual electricity consumption of 2 million households, in a territory with a population just over 265,000.

Canberra has responded by planning to override the states and ban gas being used to power datacentres. The rules would require operators to bring as much new renewable energy into the system as the centre would otherwise take. Victoria's then-minister for economic growth, Steve Dimopoulos, said in July that his state was second only to New South Wales in datacentres and was outstripping it in growth. He blamed US social media for shaping local opinion. "We are not the United States," he said. "Our planning, energy and water systems are much more carefully monitored, managed and regulated."

"The social media input from the US is very vocal and rationally based: 'It's going to increase the cost of electricity and it's going to increase the cost of water and I don't get a benefit from it.' And that's a rational position," Nicholls said.

The Australian argument lands in a week when the chip supply chain produced its own reminders of how physical the AI buildout has become. SemiEngineering's weekly review, published on 25 September, noted that Alibaba unveiled its Zhenwu V900 AI accelerator with 216GB of memory and 1,200GB/s inter-chip bandwidth, claiming three times the performance of its M890 predecessor, with mass production planned for the first quarter of 2027. The same roundup reported that CXMT put its fifth-generation G5 DRAM into mass production, claiming an 11.95nm active-area half-pitch and at least 50% more dies per wafer than the previous generation, though it did not disclose actual manufacturing yields.

Nor is the Australian complaint the only sign of export controls reshaping who gets compute and where. Tom's Hardware reported on 23 September that Nscale, a UK neocloud, drew $24 million of its $33 million 2025 revenue from Spring, a Singaporean ByteDance subsidiary, under a contract that gave Spring access to 2,304 Nvidia B200 chips at Nscale's Glomfjord data centre in Norway. The details surfaced in Nscale's S-1 filing for a US initial public offering, and in a supporting document, a $105 million loan agreement with Macquarie, rather than in the filing itself. The Financial Times reported the arrangement was legal and exploited loopholes in US export controls, according to Tom's Hardware.

Design tools, and the limits of analogy

Reuters, cited in the same SemiEngineering roundup, reported that CXMT is preparing an R&D production line for NAND flash in Beijing, and that YMTC could begin LPDDR5 trial production as early as year-end, as the two Chinese memory makers move into each other's traditional markets. The roundup also carried ASML EVP Frank Heemskerk's remark that the company is currently selling no chipmaking machines in Europe, citing insufficient investment and a lack of new fabs.

Meanwhile the tooling that fills those fabs is itself being rebuilt around AI agents. SemiEngineering reported on 24 September that Cadence is pushing "super agents" that assemble skills dynamically, while Synopsys' Prith Banerjee compared multi-agent workflows to human teams, and Mo Faisal of Movellus argued that most companies will have to specialise rather than run large language models for everything. None of that settles the Australian question, which is about power, water and who pays. It just explains why the hardware keeps arriving.