AI Data Centre Power: NEXTDC's Revenue Test

AI data centre power becomes a commercial problem when a signed contract gets ahead of the electricity needed to fulfil it. NEXTDC's latest Melbourne update puts that gap in plain view. On 21 October, the company removed the affected M3 customer contracts from its FY27 guidance because it could not reliably estimate their contribution. Its previous guidance included approximately A$140 million of both net revenue and underlying EBITDA from those contracts.[1]

That is more useful than another announcement about future megawatts. It shows how an unfinished network upgrade can reach into an operating campus and change the assumptions behind its revenue. For anyone buying compute, the question is whether the promised capacity can support the load when the customer needs it.

The Melbourne delay reaches beyond unfinished halls

NEXTDC says Jemena suspended parts of the 66kV network augmentation works in September and advised that they were not expected to restart before February 2027. Based on Jemena's indicative estimate, NEXTDC now expects the next increase in M3 grid supply during FY28, with delivery and billing of the capacity dependent on it following shortly afterwards.[1]

The less obvious detail is what NEXTDC excluded from its forecast. The affected contracts include capacity already billing. The company says existing grid supply cannot be relied on to support the full contracted power ramp in halls already delivered while parts of the works remain suspended. Actual FY27 results will include whatever those contracts contribute.[1]

Do not read that as a declaration that all M3 revenue has vanished or the entire campus has stopped operating. It is an exclusion from guidance because the contribution remains uncertain. NEXTDC says construction inside the existing buildings continues and it is working with the customer on a revised power ramp.[1] That distinction matters when reading the numbers.

The release does not identify the affected customer or establish that its workloads are AI. This is a data-centre delivery case with a direct lesson for AI procurement: a delivered hall is not evidence that the full contracted electrical load can be supported throughout the customer's expansion schedule.

Contracted capacity carries delivery obligations

NEXTDC says the deferred capacity remains under contract, subject to the customer's rights. Those contracts provide for service level credits when capacity is late or not fully powered and, in some circumstances, termination rights. The company is assessing whether force majeure or other contractual relief is available.[1]

None of that establishes that a customer has terminated or that relief will be granted. It means the operating delay has contractual consequences still being assessed. NEXTDC also warns that service level credits could take underlying EBITDA below its revised guidance.[1] A forecast that excludes an uncertain contribution does not necessarily exclude the associated downside.

If you are procuring compute, ask for a delivery schedule that separates building handover from usable electrical capacity. Put the customer load expected at each stage beside the supply available at that stage. An initial deployment may fit comfortably while a later expansion depends on works outside the operator's control. Both dates belong in the procurement decision.

Then examine what happens when the later stage slips. Have advisers check the actual agreement for credits and termination conditions, rather than assuming a headline capacity commitment settles the issue. A contractual remedy may compensate for part of a failure without supplying the compute your business planned to use that month.

A power ramp needs more than a restart date

The M3 release describes several separate dependencies. The next supply increase relies on Jemena's works. Later increases depend on upstream transformer projects at Brooklyn Terminal Station. NEXTDC says those upstream delays do not separately affect the revised FY27 guidance, but do affect capacity expected to convert to billing after FY28.[1]

Treat those as different milestones. Restarting construction does not establish when an upgrade will be energised. Energising the first upgrade does not establish that later expansion has power. A useful delivery plan names the organisation responsible for each dependency and explains what must happen before the next stage can proceed.

Community engagement belongs in that plan. According to NEXTDC, Jemena attributed the suspension to safety concerns and access problems arising from community opposition, alongside the absence of local council consent pending further consultation. NEXTDC says it supports that consultation and intends to build on its own engagement through a board-overseen community engagement plan.[1]

Jemena's own 30 September update makes an important distinction: routine maintenance and critical substation works were resuming, while the separate program involving larger replacement poles remained paused.[2] That earlier statement is not a fresh restart notice for the works NEXTDC needs. A general announcement that crews are back cannot establish that a particular project's dependency has cleared.

This extends the argument in our earlier piece on AI infrastructure's power procurement bottleneck. Procurement should follow the supply dependency through to usable capacity. A general power agreement or progress elsewhere on the network cannot answer a site-specific delivery question.

Buy compute against deliverable supply

The response should be more disciplined procurement, not a blanket claim that Australian data centres cannot deliver. NEXTDC is dealing with identified works and customer contracts. Its release does not establish the same exposure at every campus or provider. Start with the facility you intend to use and the load profile you actually need.

For each proposed deployment, request evidence of the supply supporting the initial load. Identify which part of the expansion schedule depends on future network works, and ask what documentation supports the dates. If the provider offers temporary or demand-managed supply, establish the operating limits before treating that arrangement as interchangeable with the original commitment.

Test the contingency against the workload. In a hypothetical training deployment, delaying a run might be tolerable if the commercial timetable allows it. A service with a fixed customer launch date may have much less flexibility. Price the alternative that would actually work for that deployment, rather than accepting a fallback described only as available capacity somewhere else.

The operating consequences can also change the operator's spending priorities. NEXTDC says it has paused new development projects and land acquisitions to prioritise near-term contracted capacity, and will report the outcome of its capital expenditure review by its 20 November AGM.[1] That is a dated checkpoint, not proof that the power issue will be resolved by then.

I would judge AI data centre power claims by the evidence connecting the customer's load schedule to deliverable supply. Ask the provider to identify the unfinished work and the party responsible for it. Then put the effect of a delay against your own deployment timetable. NEXTDC's update gives those questions a concrete setting. A signed capacity contract still needs an electrical system capable of supporting the promised ramp.