AI & Compute

Pros & cons

The current path vs the MMA plan.

Cons: the existing system and current trajectory. Pros: the integrated MMA corridor programme.

Cons of the current system & plan

AI data centre approvals blocked or restricted globally

Virginia is at grid capacity. Singapore has banned new campuses. Dublin has imposed a moratorium. Phoenix is on water restriction. The three constraints - power, water, land - are shutting down hyperscaler expansion in the locations Asia is currently served from.

Australian compute industry currently sub-scale

Australia has no inland hyperscale campus on the scale of those being built in the US Midwest or Indonesia. Sovereign compute infrastructure exists at small scale; sovereign data, sovereign AI training, and sovereign inference at hyperscale do not.

Latency disadvantage from coastal data centres

Sydney-based compute to Singapore is approximately 65 milliseconds one-way. Coastal Australian sites cannot compete with the latency that an Asia-Pacific compute hub requires. The infrastructure available today is in the wrong place.

Heat dissipation expensive at tropical and coastal sites

Singapore data centres operate at 27 degree Celsius ambient nights, making cooling expensive and water-intensive. Tropical sites cannot recover heat as electricity. The thermodynamic advantage of a desert site is being left on the table.

Space-based AI compute hits physics walls

GEO orbit carries a 239 ms round-trip latency floor. LEO requires hundreds of thousands of tonnes of battery mass at gigawatt scale to bridge eclipses. Orbital compute is decades away from meaningful contribution to the demand pipeline.

Sovereign capital structure not in place

Foreign hyperscalers building inland Australian capacity today would own the land, the cables, and the power generation. Australian sovereignty over the digital infrastructure of the next era is not assured by default.

Pros of the MMA plan

Inland desert solves the three constraints simultaneously

Power: sub-10c/kWh from corridor solar and PHES. Water: aqueduct from Alice Hub. Land: uncontested Crown land on Phase 1 corridors. The constraints shutting down approvals globally are design features in the Australian desert.

Latency that wins Asia

Alice Springs to Singapore via Darwin and the Inligo ACC-1 subsea cable is approximately 23 ms one-way. Closer on fibre than GEO orbit is to anywhere. Every major Indo-Pacific market within conversational AI latency budgets.

ORC heat recovery - the fourth desert advantage

3.9 degree Celsius desert winter nights enable Organic Rankine Cycle recovery of 76-107 MW per GW of compute. The cold sink is as valuable as the power supply. No tropical or coastal site can match this.

Three-tier sovereignty structure

Sovereign Corridor Trust holds Australian land, fibre, and water (100% Australian). Infrastructure JV pairs Australian engineering with global capital (51/49). Foreign tenants own only their hardware. Foreign capital builds; Australian sovereignty remains intact.

Anchor revenue funds everything else

AI campus take-or-pay agreements are the highest-value, most-bankable revenue stream in the corridor. That revenue underpins construction debt for the water, energy, and freight pillars that follow.

Multi-decade contracted demand

AI workloads run at near-100% load factors continuously, generating predictable revenue over 20-year horizons. Higher and more bankable returns than any other Indo-Pacific infrastructure customer class.

Cost and Return Summary

The dollar case for AI & Compute

CapexAI compute campus capex is anchored by Alice Hub co-location (cooling-water interfaces $0.5–1.5 B inside Alice Hub total of $65–133 B, Memo 5 §11.5). Data centre infrastructure itself is built by hyperscalers under take-or-pay leases — MMA provides power, water, fibre, and physical sites, not the compute hardware.

Tier 1 — Direct MMA revenueAI compute and data services: $15–20 B/yr at maturity (locked working figure, 10–15% regional share; conservative case at 5% share = $5 B/yr is still material). Sub-10c/kWh power cost is 3–10× cheaper than US/EU data centres. See Memo 20 §2.6.

Tier 2 — Enabled outcomes and cascading upliftSovereign AI compute capability is a strategic asset not separately monetised. Capital migration cascade (Tier 3): corporate headquarters, sovereign wealth funds, and investment capital increasingly require renewable-grid jurisdictions — MMA makes Australia the natural destination for energy-intensive compute.

Without MMA — what Australia spends insteadWithout MMA, Australia does not build a sovereign AI compute industry at regional scale. The 23 ms latency advantage to Singapore is not captured. Hyperscaler infrastructure goes elsewhere. The opportunity is not on Commonwealth balance sheet either way — but the foregone revenue is material. Memo 21 §8.1.

Programme-wide ROI summary → · Memo 19 (cost) · Memo 20 (returns) · Memo 21 (counterfactual)