The register is a census of the public record and the model a seeded simulation of every tranche in it; both are rebuilt from their source files, so every printed figure can be traced and re-run
The register is a census of the public record, 230 campuses in 272 tranches, and the model a seeded simulation of every tranche in it: 20,000 runs, with one shared shock for the conditions that move all projects together. AEMO's energy forecasts are converted to IT capacity and kept as a comparison, and the backtest scores the method as it stood in 2021.
The register
The register. Source: inashanu register (21 Oct 2026); inashanu delivery record (21 Oct 2026). As at 21 October 2026.
- Scope
- 230 campuses in 272 tranches: every campus a planning record or an operator, ASX or government statement supports, operating or proposed, with no lower size limit. An announcement needs a named site.
- Searched
- In full: NSW Major Projects, the ACT's DA Finder, Victoria's Development Facilitation Program, the EPBC portal, thirteen operators' disclosures and the Hosting Certification Framework list. In part: council and state registers elsewhere, by project.
- Status
- One of ten classes, from operating to cancelled, each with its date and a quoted source; refused, withdrawn and cancelled tranches stay on the record.
- Capacity
- As each source states it. The model works in IT load: total power is divided by 1.43, the median of 21 campuses that state both.
- Checks
- 1,512 quotes confirmed at their source (8 by hand); campuses screened for duplicates by distance; totals reconciled to AEMO's connection data.
The delivery record
The delivery record. Source: inashanu register (21 Oct 2026); inashanu delivery record (21 Oct 2026). As at 21 October 2026.
- Milestones
- Announcement, lodgement, approval, construction start and operation for each phase the public record describes, with every withdrawal, cancellation and refusal.
- Censoring
- Phases still in progress are censored at 21 Oct 2026, the model's reference date; 21 of the 74 phases followed from entry are open.
- Estimates
- Completion incidence with withdrawal as a competing outcome (Aalen-Johansen), and net durations with that outcome censored (Kaplan-Meier); pointwise bands by bootstrap.
- Lateness
- 15 first openings with a stated target period, one per campus.
- Build rates
- Installed-capacity snapshots from operator reports: built or operational quantities only, never targets.
How a printed figure traces back
- A source an operator release, planning record or AEMO file, with its date
- A quoted row the register or the demand inputs: the figure, its basis and the quote
- A model input an input the simulation reads, fingerprinted in its outputs
- A workbook cell the sheet and cell that carry it, with its source ID
- A printed figure drawn from the same cell, so page and workbook agree
Each of the 20,000 runs draws every tranche's completion and timing from the evidence, with one shared shock for the conditions that move all projects together
One run, tranche by tranche
- Completion odds. From Oxford Economics' median of 28% for a proposed project, conditioned on AEMO's 2025 to 2026 project progression: 50% for approved and 94% for building tranches, the last from the delivery record's construction starts. Earlier stages are capped by the record's incidence at their evidence age; a paused tranche takes the odds of an announced one.
- Deliverability. A score adjusts the odds for distance to a 220 kV line, substation voltage, planning status, a pending EPBC decision and any disclosed power delay. The weights are judgements, each with its sensitivity.
- Timing. First operation follows the record's durations. Started campuses with no disclosed schedule fill at operator build rates; NEXTDC's book, CDC's 555 MW and SYD3's works date follow their disclosed timing, with lateness drawn from past openings. Where an operator states the period of its next grid supply, as NEXTDC does for M3, no new IT there enters service before it.
- Shared conditions. One normal shock per run raises or lowers every tranche's odds and shortens or lengthens its lag, so good and bad years move projects together.
- What a run records. Each tranche's IT in operation at every year end to 2030. Across the runs these give the distribution, the annual fan and the contribution of each stage.
The judgements that move the answer most
| Judgement | Settings | Median, GW |
|---|---|---|
| Campus build-out | observed rates / full build-out | 4.36 / 7.56 |
| Completion odds anchor | population record / delivery record alone | 4.36 / 5.52 |
| Campus rollout duration | half as long / half as long again | 5.10 / 4.05 |
| Completion odds | halved / doubled | 3.91 / 4.84 |
| Operators' schedules | ignored / followed | 3.50 / 4.36 |
| Time to first operation | a quarter shorter / a quarter longer | 4.60 / 4.17 |
| Odds for building projects | AEMO category / construction record | 3.95 / 4.36 |
What it reads
What it reads. Source: inashanu market model (21 Oct 2026); Oxford Economics for AEMO (25 Aug 2026); AEMO 2026 ESOO; NEXTDC (21 Oct 2026). As at 21 October 2026.
- Register
- 272 tranches, their events and the delivery record
- Demand
- 396 dated inputs: AEMO energy, load factors, PUE and contracted books
- Population
- AEMO's 2025 to 2026 progression and Oxford Economics' completion odds
Reproducible by construction
Same seed and inputs reproduce the distribution to the last digit: tranche streams are keyed by register ID and the workbook records the input and code fingerprints. The central case is 4.36 GW.
The comparisons are kept as comparisons: AEMO's energy forecasts are converted, not used as a cap, and the backtest scores the method as it stood in 2021
From energy to IT capacity
From energy to IT capacity. Source: inashanu market model (21 Oct 2026); AEMO 2026 ESOO and Oxford Economics for AEMO (25 Aug 2026); NEXTDC (21 Oct 2026). As at 21 October 2026.
- Average draw
- Each scenario's financial-year energy over the hours in that year.
- Load factor
- Average draw over connection capacity: 27% in 2026 (AEMO), moving to 48% over 6 years; AEMO expects 45% to 51% at maturity.
- IT capacity
- Connection capacity over 1.43, the register's ratio of total power to IT.
- Perimeter
- The NEM and the WEM; the Northern Territory and the rest of WA are outside both.
The backtest
The backtest. Source: inashanu market model (21 Oct 2026); AEMO 2026 ESOO and Oxford Economics for AEMO (25 Aug 2026); NEXTDC (21 Oct 2026). As at 21 October 2026.
- Cutoff
- Only sources dated on or before 30 Jun 2021; the method as it stood then.
- Outcomes
- Each project read at 31 Dec 2025 from a dated record; 15 unknown and left unscored.
- Score
- Brier score 0.2211 against 0.2215 for a flat guess: no demonstrated skill yet, and an error on the low side.
- National error
- Not computed: no public source gives the national operating stock at mid-2021.
What the model does not do
What the model does not do. Source: inashanu market model (21 Oct 2026); AEMO 2026 ESOO and Oxford Economics for AEMO (25 Aug 2026); NEXTDC (21 Oct 2026). As at 21 October 2026.
- Cap supply
- The central case is deliverable supply; demand caps are screens (page 39).
- See headroom
- Distance to a line is a screen; spare network capacity is not public.
- Place contracts
- Portfolio books are spread over construction tranches by judgement.
- Price capacity
- Revenue scenarios are set on services revenue, separately from capacity.
Open judgements, and the evidence that would settle each
| Judgement | Central choice | What would settle it |
|---|---|---|
| Completion odds by planning stage | Oxford Economics' anchor, conditioned on AEMO's progression | A record of Australian planning-stage completion |
| Build rates on large campuses | Operator panels (NEXTDC, CDC); others take NEXTDC's rates | Installed-capacity histories for hyperscale campuses |
| The fleet's load factor | 27% in 2026 rising to 48% over 6 years | AEMO's measured load factor, year by year |
| Contract timing by building | Portfolio books spread over construction tranches | Tenant activation dates by building |
| M3's power delay | No new IT before a date within FY28, NEXTDC's stated period; a completion penalty and past lateness | Jemena's restart of its 66 kV works, not expected before Feb 2027 |
| A national backtest | Not computed | A dated national operating stock at mid-2021 |
Sources
- inashanu register, as at 21 Oct 2026
- inashanu delivery record, as at 21 Oct 2026
- inashanu market model, as at 21 Oct 2026
- Oxford Economics for AEMO, as at 25 Aug 2026
- AEMO 2026 ESOO
- NEXTDC, as at 21 Oct 2026
- AEMO 2026 ESOO and Oxford Economics for AEMO, as at 25 Aug 2026
Related
- How much data centre capacity will Australia have in 2030?
- How fast do Australian data centre projects reach operation?
- How should these Australian data centre figures be read?
- What data is missing on the Australian data centre market?
The full primer is published as a 103-page PDF (download). The market model and the register behind it are licensed to firms on request; the study page lists what is published and what is licensed.
This page is one of 34 in Australian data centres: the research, indexed by the question each one answers, with the four pages where the published sources disagree set first.
inashanu publishes original research for complex decisions in markets and public policy. This primer is its study of the Australian data centre market, and commissioned work follows a published catalogue and fee schedule.
A different question? If this leaves your question open, or you would put it another way, write to research@inashanu.com. Questions that keep coming up shape what we study next.
Need this updated? Every figure here carries its date. If a decision needs it brought up to date, write to research@inashanu.com.