Build Your Campus
Each hall is ~25 MW of IT load. Grow the campus, tune its efficiency (PUE) and cooling type, and watch the footprint numbers move. Drag to orbit.
drag to rotate
—
GWh / year
—
≈ homes' power use
—
ML water / year
—
permanent jobs
—
construction jobs (peak)
—
est. capex
The uncomfortable ratios
| Metric | Rule of thumb | Why it matters |
|---|---|---|
| Jobs per 100 MW | ~30–80 permanent (ops, security, maintenance) | Construction employs thousands for 2–3 years; operations employ few. The "jobs boom" is mostly temporary. |
| Power per 100 MW | ~1,100 GWh/yr at PUE 1.3 (≈ 170,000 homes) | Australia's grid is already strained; data-centre demand competes with electrification of homes and transport. |
| Water (evaporative cooling) | ~1–2 GL/yr per 100 MW in hot climates | Meaningful in drought-prone regions; dry cooling trades water for ~2–4% more electricity. |
| Capex | ~A$10–15M per MW for AI-grade builds | Big headline numbers — but much of the spend goes to imported chips and equipment, not local supply chains. |
How a country actually captures the value
- Sovereignty, not just sheds: hosting servers ≠ owning capability. The value ladder runs: land & power → operations → cloud services → models & IP. Most host countries stay on rung one.
- Condition the approvals: jurisdictions increasingly require new-build renewable PPAs, heat reuse, or grid-support commitments before connection (Ireland and Singapore both paused approvals to force this).
- Count net, not gross: honest evaluation subtracts the electricity, water, land, and transmission upgrades from the headline investment figure.
- The upside is real too: local low-latency compute enables domestic AI industries, research access, and data-residency-sensitive sectors (health, defence, banking) — benefits that never appear in the construction press release.