AI's Next Macro Shock: Bloomberg Economics Warns Australia's Boom Could Push Capex Past 2% of GDP — and Rates Back Up

For most of the past two years, the AI infrastructure story has been told in gigawatts and square feet — an American story about deserts in Texas and cornfields in Iowa. On September 7, Bloomberg Economics turned the lens on a country few expected to be at the center of it: Australia. Its warning is disarmingly simple. The AI boom now gathering pace Down Under could push capital expenditure above 2% of GDP in 2026-27, according to Bloomberg Economics economist James McIntyre — a demand surge big enough to matter to the Reserve Bank of Australia at exactly the moment policymakers hoped inflation was behind them.

It is the first time the AI buildout has been framed, in a major sell-side macro note, as a live monetary policy problem for a developed economy outside the US. And it lands on top of a physical reality that Australian communities are already living with: a data centre pipeline worth more than A$155 billion (£80bn), 162 existing facilities with 90 more planned, and neighborhoods discovering what a one-gigawatt building sounds like at 2 a.m.

The numbers behind the warning

The Bloomberg Economics analysis, published September 7, argues that the surge in AI-related spending — data centres, GPUs, energy infrastructure, and the machinery that surrounds them — could push Australian capital expenditure above 2% of GDP in the 2026-27 fiscal year. To put that in context: Australia’s celebrated mining investment boom of the early 2010s peaked with capex approaching similar shares of GDP, and it kept the currency elevated, wages hot, and the central bank hawkish for years.

The building blocks are already visible in the national accounts. Private investment climbed by almost 4% in the first quarter of 2026, led by data centre outlays, with a 16% jump in machinery and equipment spending — much of it imported servers and networking gear. Westpac estimated in May that the data centre investment pipeline would exceed A$155 billion, equal to about 5.6% of one year’s GDP, while CommBank puts the pipeline at roughly 6 gigawatts of capacity and expects data centre investment to contribute around 0.2 percentage points to real GDP growth in both 2026 and 2027. Bloomberg’s own reporting from February noted Australia has become the world’s third-largest AI investment destination behind the US and China.

The RBA has noticed. In July, the Australian Financial Review reported that the central bank flagged the data centre boom as a new threat in its inflation fight, warning the investment was running higher than expected and could create short-term “capacity pressures” — central-bank language for demand arriving faster than the economy’s ability to supply it. Then in September, GDP data showed the economy growing faster than expected, keeping rate hikes firmly on the table. The Bloomberg Economics note connects the dots: if AI capex adds demand on that scale while the economy is already running near its speed limit, the RBA’s next move is more likely up than down.

The physical economy behind the macro story

What makes Australia’s version of the AI boom distinctive is how concentrated it is — in a handful of Sydney suburbs, on a grid that was never designed for it, in a country that is simultaneously drought-prone and resource-hungry.

The BBC’s September investigation, published just days before the Bloomberg note, mapped the collision. In Lane Cove, on Sydney’s lower north shore, residents describe a constant humming from a data centre 350 meters from their front doors, with four more planned in the local industrial park — one of them 16 meters from the nearest home and 160 meters from a primary school. In western Sydney’s Marsden Park, the biggest data centre in the Southern Hemisphere is under construction 100 meters from a community. If approved, another western Sydney project would become one of the world’s largest single data centres — and at one gigawatt, Australia’s largest single energy user.

The resource math is sobering. The Australian Energy Market Operator says data centre energy demands could triple by 2030. The Climate Council estimates that without significant new renewable generation and storage, data centres could push power prices 26% higher in New South Wales by 2035. On water, some facilities consume up to 40 million litres a day — the equivalent of 80,000 households — for cooling, and Sydney Water warns data centres could use up to 25% of Sydney’s drinking water by 2035. Because data centres need uninterrupted power at a scale wind and solar cannot yet match alone, there are fears the boom could trigger a new generation of gas-fired plants; one operator, Cloud Carrier, plans three of them 90 minutes from Sydney’s CBD.

The political response arrived in July, when Prime Minister Anthony Albanese announced a legal obligation for large-scale data centres to underwrite new power supply, limit water use, and pay for extra water infrastructure — with the legislation expected in 2027, and applying only to new proposals, not projects already built or under construction. NSW Treasurer Daniel Mookhey insists the boom and the energy transition are complementary: “We see data centres allowing us to replace a lot of our old coal-fired power with clean, green renewable energy with the private sector picking up a large amount of the tab.” Community groups, led by figures like Lane Cove deputy mayor Rochelle Flood, are demanding a pause instead.

Why a small economy’s AI boom matters globally

Three things make this story bigger than Australia’s 2% of GDP.

First, it’s the template for the second wave of AI countries. The US buildout is so large it distorts global capital flows on its own. But Australia — third in AI investment, rich in land, renewables, and political stability, inside the Five Eyes alliance — is what AI infrastructure adoption looks like when it arrives at mid-sized developed economies. The same tension between investment-led growth and capacity-constrained supply will play out in Canada, the Nordics, Malaysia, and Japan. If AI capex can flip the RBA from easing to hiking, no mid-sized central bank is immune.

Second, it hardens the link between AI and the cost of money. The dominant macro debate about AI has been about productivity — whether it lifts potential growth (disinflationary) or just burns capital (neutral). Bloomberg Economics’ warning introduces a third channel: an investment shock large enough to raise rates before any productivity dividend arrives. That’s the classic mining-boom sequence, and Australia has lived it once already. Households, already squeezed, would bear it through mortgage rates; the Productivity Commission’s hoped-for AI dividend is years away, while the capex hit to demand is now.

Third, it forces the question of who pays for the boom. Australia’s answer — increasingly explicit in the Albanese framework — is that the industry must underwrite its own power and water. If that model spreads, it raises the effective cost of AI infrastructure everywhere and tests whether the economics of training-and-serving frontier models still clear at higher input costs. Lane Cove’s residents, asking what the humming is for while a reported share of AI compute goes to casual consumer use, are asking the question that every AI-hosting community will eventually ask.

The bottom line

Bloomberg Economics has crystallized something the AI industry has been slow to internalize: the buildout is now a macroeconomic event, not just a corporate capex line. In Australia, an AI investment surge worth more than 2% of GDP arrives in an economy with no spare capacity, a central bank openly discussing hikes, and a population only beginning to price the water, power, and neighborhood costs of hosting the world’s compute. The productivity story may still win in the end — CBA argues the boom could revive Australia’s “dismal” productivity growth, and economists credit the data centre boom with keeping the country out of recession. But between now and then, the AI era’s first monetary policy casualty may be an Australian mortgage holder listening to a data centre hum through the bedroom wall.