01 · The build
Australia’s data centre build already outweighs Snowy 2.0
The stages already under construction or commissioned represent about $16 billion to $25 billion of committed capital, more than Snowy 2.0’s revised cost.
A$ billion
Hover the capital range for its basis.Tap the capital range for its basis.
Published project costs and committed capacity
CSVValues
| Committed data centre capital, low | 16.2558 |
|---|---|
| Committed data centre capital, high | 25.2868 |
| Snowy 2.0, revised cost | 12 |
| Committed capacity, gigawatts | 1.8062 |
| Cost per megawatt, low | 9 |
| Cost per megawatt, high | 14 |
SourceRegister of planning filings; AusNet via PwC; Snowy 2.0 Updated Business Case, May 2024.
ScopeCommitted stages only. The announced pipeline is larger and is not counted here.
02 · The build
A generation of lost competitiveness
Australia sells the world fewer, and simpler, things than it did a generation ago. Its economic complexity rank fell from 33rd in 1995 to 73rd in 2024.
complexity rank by year
Economic complexity and export-mix measures
CSVValues
| Year | Complexity rank |
|---|---|
| 1995 | 33 |
| 2000 | 42 |
| 2005 | 48 |
| 2010 | 59 |
| 2015 | 63 |
| 2020 | 63 |
| 2024 | 73 |
SourceHarvard Atlas of Economic Complexity; World Bank; ABS.
Why hereThe case for building sits before any result, as evidence rather than as a statement of position.
03 · The build
Ongoing headcount is the wrong ruler for infrastructure
An operating data centre employs few people. So does a substation, and nobody measures a substation by its headcount. The value is in what the asset enables.
Built capital standing behind each ongoing job
Hover a bar for its exact value.Tap a bar for its exact value.
Capital standing behind each ongoing job
CSVValues
| Industry | Capital per ongoing job |
|---|---|
| Electricity, gas & water | 4.41546 |
| Data centres (illustrative) | 1.82813 |
| Transport & warehousing | 1.06847 |
| All industries | 0.578366 |
| Manufacturing | 0.258912 |
SourceABS national accounts and Labour Account; attribution at A15.
NoteThe data centre figure is illustrative and is derived, not published. It is marked as such on the chart.
04 · The build
Three generational buildouts, one timeline
The data centre build lands directly on top of two other generational buildouts, the energy grid and the Housing Accord.
concurrent electrician FTE by quarter index
The model's central path, quarter by quarter
CSVValues
| Quarter | Homes | Energy grid | Data centres | All three |
|---|---|---|---|---|
| 2026Q1 | 26587.5 | 901.362 | 129.761 | 27618.6 |
| 2026Q2 | 26587.5 | 901.362 | 129.761 | 27618.6 |
| 2026Q3 | 29558.1 | 4369.56 | 353.783 | 34281.5 |
| 2026Q4 | 29558.1 | 4369.56 | 401.505 | 34329.2 |
| 2027Q1 | 29558.1 | 4369.56 | 590.885 | 34518.6 |
| 2027Q2 | 29558.1 | 4369.56 | 669.855 | 34597.6 |
| 2027Q3 | 31637.6 | 5089.7 | 1293.95 | 38021.3 |
| 2027Q4 | 31637.6 | 5089.7 | 1455.14 | 38182.5 |
| 2028Q1 | 31637.6 | 5089.7 | 1644.99 | 38372.3 |
| 2028Q2 | 31637.6 | 5089.7 | 2169.6 | 38896.9 |
| 2028Q3 | 31934.7 | 7765.73 | 2216.9 | 41917.3 |
| 2028Q4 | 31934.7 | 7765.73 | 2069.55 | 41770 |
| 2029Q1 | 31934.7 | 7765.73 | 2004.38 | 41704.8 |
| 2029Q2 | 31934.7 | 7765.73 | 2171.72 | 41872.1 |
| 2029Q3 | 32528.8 | 6975.9 | 2213.03 | 41717.7 |
| 2029Q4 | 32528.8 | 6975.9 | 2005.97 | 41510.7 |
| 2030Q1 | 32528.8 | 6975.9 | 1867.93 | 41372.6 |
| 2030Q2 | 32528.8 | 6975.9 | 1651.39 | 41156.1 |
| 2030Q3 | 32528.8 | 2218.44 | 1038.36 | 35785.6 |
| 2030Q4 | 32528.8 | 2218.44 | 964.028 | 35711.3 |
| 2031Q1 | 32528.8 | 2218.44 | 824.995 | 35572.2 |
| 2031Q2 | 32528.8 | 2218.44 | 75.7377 | 34823 |
| 2031Q3 | 32528.8 | 2252.93 | 75.7377 | 34857.5 |
| 2031Q4 | 32528.8 | 2252.93 | 0 | 34781.7 |
SourceNational Housing Supply and Affordability Council; ABS; AEMO Integrated System Plan; the register of planning filings.
05 · The build
Every build depends on the same four qualification-dependent trades
All three builds draw on the same four trade workforces. The relevant licences, registrations and competencies are not interchangeable.
schematic · no value axis, not to scale
The licensing and registration structure of the four trades
CSVSourceState and territory licensing law for electricians, plumbers and HVAC&R mechanics; the national cabling registration requirement administered by the Australian Communications and Media Authority.
06 · How we modelled it
A workforce is a bathtub
Recruiting more apprentices delivers no qualified electricians before about 2030, because the qualification takes four years and nothing shortens it.
schematic · no value axis, not to scale
Hover a flow for its exact value.Tap a flow for its exact value.
Who joins the workforce and who leaves it
CSVValues
| Electricians employed at the model start | 197511 |
|---|---|
| Apprentice completions a year | 8892.12 |
| Field completions a year, all trades | 11899 |
| Migrant arrivals a year | 621 |
| Retirements and exits, per quarter | 0.0035 |
| Retirements and exits, a year | 0.0139267 |
SourceABS; NCVER; Home Affairs; Jobs and Skills Australia; the supply engine at A3.
NoteFlows are not drawn to scale. Time is the scaled element: the apprenticeship pipe’s four year-segments carry the lag, and the visible figures are rounded teaching forms with the exact values on hover.
07 · What we found
A shortfall in nine futures out of ten
In 90.6 per cent of the 10,000 futures the model treats as plausible, there are not enough electricians at the 2027–29 peak.
In the remaining one in ten, supply holds through the peak. At the single likeliest value of every input it holds as well, with about 2 per cent to spare.
concurrent workers
Hover the distribution for its bins.Tap the distribution for its bins.
Each of 10,000 futures is a complete run of the model under one draw of the parameters. The tallest bar represents the model runs in which the peak gap is small or nil.
5,433 concurrent electricians short is the median: as many futures land above it as below, and four in five land inside the shaded band.
Re-run the same 10,000 futures with every new data centre removed and the gap does not close: the median falls, and a shortfall persists in two in three of futures.
Each of 10,000 futures is a complete run of the model under one draw of the parameters. The tallest bar represents the model runs in which the peak gap is small or nil.
5,433 concurrent electricians short is the median: as many futures land above it as below, and four in five land inside the shaded band.
Re-run the same 10,000 futures with every new data centre removed and the gap does not close: the median falls, and a shortfall persists in two in three of futures.
Ten thousand runs of the model
Deletion CSVCSVValues
| tenth percentile | 109.785 |
|---|---|
| median | 5432.72 |
| ninetieth percentile | 11565.7 |
| modelled futures | 10000 |
Remove every new data centre, and the shortfall persists
Re-run the same 10,000 futures with zero new data centres and a shortfall still occurs in 67.7 per cent of them, with a median gap of 1,662 concurrent electricians at the peak. On the difference-of-medians reading, the pipeline accounts for 69 per cent of the median gap; the other defensible readings sit lower and are set out in the report’s appendix.
SourceEvery figure computed from the model at build time; method at A4.
MergeThe briefing draws this distribution twice, as a histogram and as a strip. Here they share one axis, so the counterfactual sits directly under the result.
DomainAxis ends at the 99.5th percentile. Fifty draws sit beyond it and are counted in the readout, not drawn.
08 · What we found
Not just electricians
Plumbers, HVAC&R mechanics and cablers gap at their own scale.
concurrent workers
Hover a band for its range.Tap a band for its range.
The model’s ten thousand runs, across the four qualification-dependent trades
CSVValues
| trade | tenth percentile | median | ninetieth percentile |
|---|---|---|---|
| Electricians | 109.785 | 5432.72 | 11565.7 |
| Plumbers | 0 | 2354.35 | 6292.5 |
| HVAC&R mechanics | 210.95 | 1681.25 | 3319.05 |
| Telecom trades | 658.254 | 1156.2 | 1903.91 |
Each label records how closely that trade’s model reproduced the official employment series before it was used to project anything. Electricians reproduce it best, which is why the headline result rests on them. HVAC&R mechanics sit just outside the benchmark across the full period, and inside it once a single 2025 survey reading is set aside, so the result is kept and qualified rather than dropped. Telecommunications is the weakest series in the model, being smaller, noisier and structurally declining, and the national registration count that would measure it was last published in 2019. It is carried as an early warning rather than a headline, and it is excluded from the composite.
SourceABS Labour Force detailed; NCVER; reconciliation at A9.
CareThe electrician row is the headline model. The other three come from the stack model and are never the same object.
09 · What we found
Where the collision concentrates
The data centres are being built where the tradespeople are not.
share · workers
The western Sydney corridor at its tightest quarter
CSVValues
| Corridor share of the staged pipeline | 0.358051 |
|---|---|
| Corridor share of the trades workforce | 0.0960322 |
| Corridor demand at the peak | 2859.38 |
| Corridor workforce available | 2073.9 |
| Corridor shortfall, low supply basis | 0 |
| Corridor shortfall, central supply basis | 785.486 |
SourceABS Census; the register of planning filings.
10 · What would change it
No single measure is enough
Together the plausible package closes about 96 per cent of the median gap, and about 45 per cent of the gap in the hardest futures.
Recruit more apprentices now and 0 arrive inside the window. The qualification takes four years, so the first completions from a surge that starts now land in 2030, after the peak has passed.
in-window concurrent electrician FTE
The response package, lever by lever
CSVValues
| family | low | high |
|---|---|---|
| Recruit more apprentices now | 0 | 0 |
| Lift completion rates | 651 | 1503 |
| Re-activate licensed non-practising | 206 | 2534 |
| Retain more (cut exits) | 800 | 1640 |
| Raise migrant licensing throughput | 512 | 1816 |
| Double the female-entry trajectory | 200 | 700 |
| Sequence the discretionary pipeline | 574 | 1298 |
SourceLever accounting at A12; parameter tiers at A16.
RuleThe axis ends at the median gap, and the terminus is the finding. Positions clamp there.
11 · What to do
Act inside the window
Every response below is real, and none of them closes the gap alone. Together the plausible package closes about 96 per cent of the median gap.
Contributions are not additive. Variants within a family are alternatives rather than cumulative, and the package figure is the model’s result rather than the sum of the measures below.
Inside the window · supply
Keep the electricians Australia already has
1,200 · 800 to 1,640 · In-window FTE
Evidence · Strong on size, absent on cause
Employers hold this one, through pay and conditions. It costs government nothing, and it is not government’s to pull. No registered source says why electricians leave.
Bring back licensed electricians who are not practising
1,029 · 206 to 2,534 · In-window FTE
Evidence · Weak
State licensing regulators hold this one, because the registers are theirs. The figure is an upper bound on an upper bound, multiplied by an unsourced take-up band.
Raise apprentice completion rates
701 · 651 to 1,503 · In-window FTE
Evidence · Bounded
The Commonwealth pays for completions, while employers and providers hold the conditions. No cohort in a decade has improved by five points, and the ten-point variant sits outside the observed range.
Expand recognition and licensing capacity before expanding visas
600 · 512 to 1,816 · In-window FTE
Evidence · Saturation observed
The Commonwealth holds visas and skills assessment, and the states hold provisional licences and supervised hours. Expanding one side alone saturates the pathway: saturation is observed near three times current intake, where five times and ten times deliver the same number. Supervised hours consume licensed electricians, so a careless expansion subtracts from the workforce it means to add.
Inside the window · demand
Sequence the peak deliberately
1,298 · 574 to 1,298 · In-window FTE
Evidence · Priced on the discretionary tranche
States and territories hold approvals and connections, and the Commonwealth convenes. The mechanism is a published, prospective framework that separates committed projects from reschedulable ones.
It is a coordination instrument that moves work in time rather than a response on the supply side, and which programme moves is a choice for governments on grounds this work does not model. The figure is what an eight-quarter deferral of a block of the discretionary tranche’s size buys.
After the window
The build after this one
0 · this window · In-window FTE
Evidence · Arithmetic
Apprentice recruitment, the female-entry trajectory and industry-funded places are quarantined here. Recruiting more apprentices delivers no qualified electricians before about 2030, because the qualification takes four years and nothing shortens it. Trades capacity is national economic infrastructure and is not currently treated as such.
What remains after the package is about 205 concurrent electrician FTE at the median, about 4 per cent of the gap. In the hardest futures about 55 per cent of the gap remains.
RankingOrdered as the model ranks them, largest in-window contribution first.
EvidenceEvery response carries its own grade, including the ones where the evidence is thin.
12 · What to do
Restore the measurements
This work reconstructed, from 290 registered public documents, numbers that no instrument publishes. The megawatts are measured quarterly, while the workforce instruments published their final editions in the year the build begins.
There are seven instruments below, each with an owner, a frequency and something it would reveal. The largest supply lever in the model is retention, and the reason nobody can say how to pull it is that the series that would have told us stopped. Restoring these is the cheapest response on this page and the only one that outlasts the window.
The seven
These instruments would serve every build. A quarterly count of the qualification-dependent trades by region tells a housing developer, a network operator and a data centre operator the same thing, and none of them can plan a workforce without it.
How this was modelled
The figures come from a cohort stock-and-flow model of the qualification-dependent workforce, run quarterly from 2026 to 2031 and sampled ten thousand times across the parameters that are genuinely uncertain: apprenticeship completions, migration throughput, exit rates and the labour content of each programme’s published schedule. Demand comes from the three programmes’ own announced schedules, discounted for the share of staged capacity that historically reaches construction. Nothing is forecast that the schedules do not already assert. The model tests the workforce demand implied by published schedules against evidence-based assumptions about workforce supply. Every figure on this page traces to one of 290 registered public documents, and every chart offers its values and its underlying series. The full method, the parameter tiers, the sensitivity tests and the corrections made during the work are in the report’s appendix.
No published source decomposes electrician employment by type of work. The model therefore measures what can be measured, being the electricians the three programmes required in 2025 at their observed activity levels, and treats the rest of the trade as ongoing work carried forward at that level. The pool that remains is roughly a fifth of the trade. Within it, the model treats one electrician as able to work on any of the three builds, which licence classes do not always allow, so the real constraint is tighter than the one modelled.
The worded share is generated from two registered claims, 42,812 concurrent electrician FTE free at the tightest quarter over the model’s starting stock of 197,511: a 2028Q3 numerator over a 2025Q4 denominator, the same construction as the published gap-to-stock ratio.