A 1GW Outback Deadline Just Vanished — and $586.6B of Hyperscaler Capex Still Assumes Permitting Is Never the Bottleneck

Project Ares is Energy North's proposed hyperscale campus at Murranji cattle station in the Barkly Region: 1GW across 191 square kilometres, with a gas turbine system, a solar farm, an airstrip and accommodation for 4,300 construction workers, drawing on as much as four billion litres of groundwater a year. Its environmental referral carried a first-power date of June 28, 2028. After the regulator suspended the seven-week deadline it had set itself to agree environmental parameters, Energy North told Reuters it was no longer working to that June 2028 date and would confirm timing only once the regulator sets the terms of an environmental impact statement.

The standard framing of Australian AI data centres, power scarcity and water limits, is already out of date. AEMO's August 2026 Electricity Statement of Opportunities forecasts data centre consumption in the National Electricity Market rising from about 5 TWh in 2025-26 to 34 TWh by 2035-36, and from roughly 3% to about 13% of grid supply. The same document reports no forecast reliability gap before 2030, 9.1GW of new generation and storage connected in 2025-26, and a 40GW pipeline of committed and anticipated projects into the early 2030s.

The decisive detail is methodological. AEMO's 2026 forecast now accounts for project attrition during the connection process and for the gap between connection capacity and actual electricity use. The grid operator has conceded inside its own forecast that nameplate megawatts are not deliverable megawatts. Australia is engineering around its power shortage and has not engineered around its permission shortage, which is why a 1GW campus with a solar farm and a gas turbine system is now waiting on a regulator rather than on a switchboard.

October 2026 produced two guidance statements that look contradictory and are not. NextDC cut FY27 net revenue guidance to A$475-500 million from A$615-640 million and underlying EBITDA to A$245-265 million from A$385-410 million, after Jemena suspended the network upgrade works its Melbourne M3 campus needs following community opposition. Around A$140 million of contracted customer revenue slipped, the next power delivery moved to 2028, and the works will not restart before February 2027. The shares fell 41.7%, the worst single day since the company listed in 2010.

In the same month, Vertiv reported third-quarter net sales of $4,313 million, up 61% year over year, with organic orders up more than 100% and full-year guidance raised. The reason vendors are not punished is not indifference to permits. It is that power and cooling equipment is bought against a build plan, not against a completed approval. Vendors are paid when gear ships; capacity operators are paid when a building energizes. A permitting slip therefore re-times revenue instead of cancelling it, and moves the loss onto whoever carries contracted but unenergized capacity.

This is why the outback is not an Australian story. On October 28 Microsoft told investors that September-quarter capital expenditures of $45.2 billion came in below plan because some third-party capacity deliveries it expected in the quarter were delayed, and lowered its calendar-2026 spending forecast to roughly $170 billion from about $175 billion. Reuters put calendar-2026 capital spending at $118.1 billion through September. Amazon held 2026 cash capital expenditure guidance at roughly $220 billion after spending $77.3 billion in the September quarter alone, and said 2027 would be meaningfully higher. Alphabet guided to $195-205 billion for 2026 after spending $132.0 billion in nine months, against $63.6 billion a year earlier. Oracle carried $664 billion of remaining performance obligations, up $209 billion year over year, against a fiscal 2027 capital plan of $90-95 billion.

Those four companies spent $586.6 billion on property and equipment in the trailing twelve months. That is the number the read-through has to work on, because a hyperscaler guidance update is now as much a statement about approval and connection calendars as it is a statement about demand. One regulator in the Northern Territory will not change Amazon's income statement. A hundred of them, on the same schedule, will.

Between the June quarter and November 2 the market did not de-rate AI infrastructure as a single trade. It sorted it by who has to deliver before they can collect, and balance sheets, not demand curves, separated the winners.

Two patterns sit inside that table. First, the balance sheet did the separating: Microsoft converted $168.3 billion of operating cash flow into $60.0 billion of free cash flow after spending $135.9 billion on property and equipment, and its equity added about $1.1 trillion of value in the same window in which Amazon's free cash flow turned negative at $46.9 billion. Second, CoreWeave and IREN, the two names whose revenue literally begins when a building energizes, fell 44.8% and 41.5%, the sharpest moves in the group, on 11.5x and 56.0x net debt to EBITDA, and the market has repriced exactly that fragility.

This is inference, not proof. The same window contained Alphabet's earnings miss and the collapse of the Firmus listing, either of which could move these prices on its own. What the data does support is narrower: the approval layer has become visible enough to move the equity value of companies that had been valued as pure expressions of AI demand, which means the approval layer is now in the price.

The next five months contain a stack of dated tests, and each one either confirms the thesis or quietly punctures it.

Over one to three years, the delay does not shrink the pipeline; it changes who owns it. Australia's move to a single national rulebook, plus a Senate report in March 2027, creates one approval regime for a market where AEMO expects data centre demand to rise roughly sevenfold by 2035-36. One rulebook cuts both ways. It removes nine jurisdictions' worth of bespoke risk, which lowers the cost of the pipeline, and it creates one gate that every project must pass, a gate with a demonstrated habit of suspending its own deadlines. Expect capital to concentrate in developers that can fund an approval cycle without a financing window open: IREN's $3.6 billion of investment-grade GPU financing at 6.0% is the template, and even its 800MW Australian campus explicitly ties early works and procurement to regulatory approvals.

The outback ruling is small in dollars and large in signal. A 1GW campus with an airstrip, a solar farm, a gas turbine system and a village for 4,300 workers lost its date because a regulator needed more time to set environmental parameters, not because there was no power, no turbine, no transformer and no money. The binding constraint on AI capacity has moved from engineering supply to granting permission, and it moved while four US companies wrote $586.6 billion of cheques against a schedule that treats permission as a formality.

The investable conclusion is not that the buildout is slowing. It is that the buildout has entered a phase where the winners are paid on order and the losers are paid on energization. That favours the equipment chain, Vertiv and GE Vernova, over the capacity operators carrying contracts against buildings that have not been permitted, and it means every hyperscaler guidance update through 2027 should be read first as an approval-calendar disclosure and only second as a demand signal.