Australia’s AI Power Play

Why the nation that owns the water, the electrons, and the space owns the compute — and why Australia, with the MMC corridor as its backbone, is positioned to become the sovereign AI infrastructure host of the Indo-Pacific.

1. The Three Constraints — Water, Electrons, and Space

Every large AI data centre campus needs three things before it needs anything else: power to run the GPUs, water to cool them, and land to build on. Everywhere these three converge at scale is already being fought over. Everywhere they don’t converge, the hyperscalers are stuck.

Power. A single modern GPU cluster — the kind used for frontier AI training — consumes 50–500 MW continuously. A hyperscale campus of ten to twenty such clusters runs at 500 MW to 5 GW. That is the output of a large coal power station, drawn 24 hours a day, 365 days a year. The grid in most jurisdictions cannot provide that reliably and cleanly. In Virginia — the largest data centre market in the world — new campus approvals are being delayed years because the grid cannot keep up. In Singapore, new data centres were effectively banned for three years due to power and land constraints. In Ireland, data centres now consume 21% of national electricity and regulators are alarmed.

Water. Cooling a large data centre campus requires millions of litres of water per day. Evaporative cooling towers — the most common method — consume water at 1–3 litres per kWh of compute delivered. A 1 GW campus uses 1–3 billion litres of water per year. In water-stressed regions — which includes most of the places where AI demand is highest — this is becoming a political and regulatory constraint that no amount of corporate sustainability reporting can resolve. Phoenix, Arizona has told data centre developers there is no more water. Several European cities are considering similar restrictions.

Space. A multi-gigawatt AI campus requires hundreds of hectares of flat, clear land with ground-bearing capacity for heavy infrastructure, exclusion zones, and room to expand. Urban and peri-urban land near existing power grids is expensive, contested, and increasingly unavailable. The hyperscalers are being pushed further and further from their preferred locations — and still cannot find sites that solve all three constraints at once.

Australia’s inland corridor zones solve all three constraints simultaneously and at a scale that no other location in the Indo-Pacific can offer. This is not a coincidence. It is the direct result of the MMC programme design.

2. The Scale of AI Energy Demand

The numbers are large and moving fast. Global data centre electricity consumption reached approximately 460 TWh in 2022. The International Energy Agency projects this will exceed 1,000 TWh by 2026 — more than Australia’s entire national electricity consumption today — and continue growing steeply through 2030 and beyond as AI model sizes, training frequency, and inference demand all compound simultaneously.

A single large AI training run — the kind used to produce a frontier language model or a large multimodal system — now consumes in the range of 50–500 GWh. The largest known training runs in 2025–2026 are approaching the annual electricity consumption of a small Australian regional city. And training runs are not the dominant load — inference is. Every time a user queries an AI system, compute runs. At billions of queries per day across the major AI platforms, inference load dwarfs training load and runs continuously.

The critical insight is that inference load is always-on. It cannot be time-shifted to match renewable generation peaks. It cannot be curtailed when the wind drops. It requires what the energy industry calls firm, despatchable, 24/7 power — exactly the kind of power that solar and wind alone cannot reliably provide, and exactly the kind of power that Alice Hub PHES at 40 GW and 32-day storage can.

3. Why Location Decides the AI Race

The hyperscalers — Microsoft, Google, Amazon, Meta, and the major Asian cloud providers — are not building data centres where they want to. They are building where they can. Location selection for a multi-gigawatt AI campus now turns on four hard requirements:

No single location in the Indo-Pacific currently satisfies all four requirements at the scale AI demands. Singapore is sovereign and stable but has no power or water headroom. Japan is sovereign and stable but power is expensive and grid carbon intensity is high. India has land and growing power but is not yet a trusted Five-Eyes data jurisdiction. Indonesia and Vietnam have land but political and grid uncertainty.

Australia, with the MMC programme operating, satisfies all four — at continental scale.

4. Australia’s Strategic Position

Australia’s advantages for AI infrastructure are not marginal. They are structural and large.

The best solar resource on Earth. The Australian interior receives 2,200–2,800 kWh per square metre per year of solar irradiance — among the highest sustained values on the planet. The MMC agrivoltaic programme covers 13.4 million hectares of corridor land with dual-use solar panels. The generation potential across the six corridors exceeds 1,000 GW nameplate. Even at 30% capacity factor, that is 300 GW of average output. AI campuses along the corridor receive power directly from this resource via the HVDC spine at 4–7¢/kWh — three to five times cheaper than power costs in Singapore or Japan.

Alice Hub PHES — the despatchable battery. Solar generation is intermittent. AI inference is not. The Alice Hub pumped hydro energy storage system at 40 GW output and approximately 30 TWh of storage capacity bridges this gap. Excess solar generation pumps water uphill during the day. The turbines generate on demand through the night and through cloud events. The result is firm, despatchable, 24/7 renewable power at corridor scale — the specific product AI campuses need and cannot get anywhere else in the region at this price.

The MMC HVDC backbone. Cheap power is only useful if it can be delivered. The MMC HVDC transmission spine runs the length of each corridor — 17,600 km total across the six-corridor system. AI campuses at corridor intersection cities receive direct HVDC feed. The transmission losses are low. The capacity is large. And the backbone is being built for energy and transport purposes regardless — the AI delivery function is incremental.

Water from the aqueduct. The Alice Hub aqueduct system delivers up to 25,000 GL of water annually to southern Australia — sourced from northern flood harvesting and stored in the MacDonnell Ranges reservoir. Corridor towns and campus zones along the route have access to aqueduct water for industrial cooling. This is the constraint that is shutting down data centre approvals in Phoenix, Singapore, and Dublin. Along the MMC corridor, it is a design feature.

Unlimited inland space. The Phase 1, 2, and 3 MMC corridors run through some of the least densely populated terrain on Earth. Crown land is available at negligible cost. Exclusion zones for large campuses do not displace communities. There is no planning congestion, no heritage constraint per hectare of desert, and no neighbours to complain about cooling tower noise.

5. The MMC as AI Power Infrastructure

The MMC was designed as a transport and energy corridor. Its architecture happens to be precisely what AI infrastructure requires.

The five-track viaduct carries freight and maglev passengers — but it also carries the HVDC transmission cables, the fibre optic data spine, the aqueduct water pipe, and the service conduits for every utility. Every MMC pylon is a combined transport structure, transmission tower, data conduit, and water pipe support. The infrastructure is already there for an AI campus to plug into — power, data, and water in a single connection point at each corridor town and intersection city.

For a hyperscale AI operator, the value proposition is extraordinary. Instead of spending years negotiating grid connections, water licenses, and fibre routes separately — then assembling a bespoke campus infrastructure from scratch — the MMC corridor delivers all three services to a ready-zoned campus site in a single package. The corridor is the infrastructure. The campus plugs in.

6. The Inland AI Hub Model — Phase 1, 2, and 3 Corridors

The Phase 0 corridor — Melbourne to Brisbane, 2,300 km — is the platform’s proving build. It delivers passenger maglev, electric freight, and the east-coast HVDC backbone on the same elevated structure, solving the east-coast passenger, freight, and transmission problem. AI compute campuses are not part of Phase 0 — they need continental-scale power (sub-10c/kWh desert solar), continental-scale water (Alice Hub aqueduct), and the fibre spine carried on the continental corridors. The transformational AI compute opportunity arrives with Phase 1 and the continental corridors.

Phase 1 through to the full six-corridor system opens up the Australian interior: the Pilbara, the Great Sandy Desert, the Gulf Country, the Nullarbor approaches, the central corridors connecting Port Hedland to Mackay and Mount Isa to Perth. These are the corridors where the solar resource is highest, the land is cheapest, the water is available from the aqueduct, and the new towns being built along the corridor spines are purpose-designed rather than retrofitted.

A new MMC corridor town is not a suburb of an existing city. It is a planned settlement built to a modern specification — with power, water, data, transport, and community infrastructure designed in from day one. An AI campus co-located with a corridor town is not an industrial intrusion into an existing community. It is part of the town’s economic foundation — providing employment, tax revenue, and anchor demand for the town’s power and water systems.

The most compelling near-term AI hub location is the Pilbara — specifically the inland corridor zone on MMC Corridor #4, east of Port Hedland. This is where the solar resource, the spaceport, the deep-water port logistics, and the subsea cable export route all converge. It is examined in detail in sections 8 and 9 below.

7. Data Along the Corridor — The MMC as a National Data Spine

The MMC corridor is not just a power and water delivery system. It is a data network.

Running the length of every corridor, integrated into the viaduct structure and the pylon conduit system, is a fibre optic data spine. Every corridor town, every intersection city, every freight hub, every solar field substation, and every maglev station is a node on this network. The data spine carries the real-time operational data of the corridor itself — freight tracking, energy dispatch signals, structural health monitoring from the 4.6 million embedded IoT sensors in the Phase 0 viaduct alone — but its capacity far exceeds operational needs.

The excess fibre capacity is a national data asset. For AI operators, proximity to a high-bandwidth, low-latency fibre spine is a significant site selection criterion. Training data moves. Model weights move. Inference results move. Connectivity between campuses — for distributed training across multiple sites — requires high-bandwidth low-latency links. The MMC data spine provides all of this, connecting every campus on the corridor network to every other, and connecting them all to the coastal subsea cable landing points.

The data spine also connects to the 200 corridor towns and 11 intersection cities as a national digital infrastructure asset in its own right — delivering gigabit-class connectivity to regional Australia as a byproduct of the corridor build, without requiring a separate telecommunications programme.

8. The Spaceport and Satellite Dimension

MMC Corridor #4 runs from Port Hedland inland through the Pilbara and across to Mackay. Approximately 450 km east of Port Hedland, on the corridor route through the eastern Pilbara and the Great Sandy Desert, sits one of the strongest spaceport locations in Australia — and arguably in the Southern Hemisphere.

8.1 The Pilbara Inland Spaceport Site

The site at approximately 20.73°S, 123.04°E combines a set of advantages that no coastal or southern Australian spaceport candidate matches:

• Downrange corridor. East and northeast launch trajectories cross 285,000 km² of the Great Sandy Desert — near-zero population density. Booster separation at 60–120 km downrange falls on uninhabited desert. The safety case for high-cadence launch operations is stronger here than at any alternative Australian site.
• Fuel supply. Liquid methane — the propellant of choice for next-generation heavy launch vehicles including Starship — is produced in abundance at Port Hedland, Dampier, and Karratha. Rail connection from the Pilbara’s existing 2,800 km heavy-haul mining railway network delivers propellant to the inland site at minimal cost.
• Port Hedland recovery and launch pairing. The inland site functions as the heavy assembly, propellant production, and booster recovery/refurbishment hub. The coastal Port Hedland facility handles primary orbital launches, with proven Indian Ocean splashdown zones for upper-stage recovery already demonstrated by recent orbital flight programmes. Deep-water Port Hedland’s port infrastructure handles ship-to-shore component movement; the heavy-haul rail network moves components inland for refurbishment.
• Latitude. At 20°S, the site is usable for a wide range of orbital inclinations with only a modest delta-v penalty compared to equatorial sites. Sun-synchronous, equatorial, and highly inclined orbits are all accessible.
• MMC Corridor #4 connection. The spaceport site sits directly on the Phase 1 corridor route. Power from the MMC HVDC spine, water from the aqueduct, and fibre connectivity from the data spine are all delivered as part of the corridor build — not as separate infrastructure projects.

8.2 Satellite Connectivity as AI Infrastructure

The spaceport’s value to the AI compute strategy extends beyond launch operations. A launch facility of this scale generates and requires a ground station network — high-bandwidth satellite uplink and downlink infrastructure along the corridor spine. That same infrastructure serves AI campus connectivity.

Low Earth Orbit (LEO) constellation connectivity — Starlink and its successors — provides low-latency broadband coverage across the corridor and across Australia’s marine economic zone. For AI campus operations, LEO connectivity provides:

• Redundancy. A second high-bandwidth path alongside fibre for mission-critical AI inference services, providing resilience against fibre cuts or corridor disruptions.
• Maritime and remote connectivity. Inference services delivered to ships, remote mining operations, and agricultural operations across the continent and the region — extending the AI compute market beyond fixed campuses.
• Ground station revenue. The corridor’s 2,300–17,600 km of unobstructed flat terrain is ideal for distributed ground station arrays. Operating ground stations for LEO constellations is a revenue stream for corridor towns and a negotiating instrument for preferential LEO service terms.

The Pilbara spaceport, the MMC data spine, and the LEO ground station network together form a layered connectivity architecture: fibre for high-bandwidth low-latency intra-corridor traffic, LEO for wide-area coverage and redundancy, and the subsea cable corridors (described in the next section) for high-volume inter-continental export traffic.

9. AI Compute Export — The Subsea Corridor

The most underappreciated element of Australia’s AI infrastructure position is its geography relative to Southeast Asia and the Pacific.

Australia’s northwest coast — Port Hedland, Dampier, Exmouth — is the shortest submarine cable route between a high-capacity renewable energy zone and the major Asian compute demand centres of Singapore, Jakarta, Kuala Lumpur, Manila, and beyond. The undersea distance from Port Hedland to Singapore is approximately 4,000 km — comparable to the length of a US transcontinental fibre route.

9.1 Why Subsea Beats Satellite for Bulk Compute

Satellite connectivity — even LEO — has fundamental physical limits that make it unsuitable for bulk AI compute export:

For bulk AI compute export — delivering inference results, training outputs, and hosted AI services to Asian customers — subsea fibre is the only economically viable transport medium at scale. Satellite is the redundancy and last-mile layer. Fibre is the trunk.

9.2 The MMC Subsea Cable Corridors

The MMC programme’s coastal termination points — Port Hedland, Darwin, Cairns, Brisbane, Sydney, Melbourne, Adelaide, and Esperance — are the natural landing points for subsea cable corridors connecting Australia to Asia and the Pacific. These are not new cable routes. Australia already has subsea cables to Singapore, Japan, Indonesia, and the Pacific. The MMC changes the equation by:

• Creating inland generation capacity that justifies dramatically higher cable bandwidth than Australia’s current export needs. Today’s cables are sized for Australia’s internet traffic. Tomorrow’s cables need to be sized for Australia’s AI compute export.
• Connecting the inland AI campuses to coastal cable landing stations via the MMC data spine — eliminating the need for separate terrestrial fibre routes from inland campuses to the coast.
• Providing the sovereign routing advantage. Data transiting Australian-owned subsea cables under Australian jurisdiction is subject to Australian law. This matters to Asian governments and corporations who want compute services that are not routed through US or Chinese-controlled cable infrastructure.

9.3 Compute as an Export Commodity

Australia has spent two centuries exporting raw materials — iron ore, coal, LNG, wool, wheat. The value-added product has always been manufactured elsewhere. AI compute reverses this completely.

When an Australian-owned data centre on the MMC corridor trains a model for a Southeast Asian client, or runs inference for a Japanese enterprise, or hosts a Pacific island nation’s government AI system, Australia is exporting a high-value manufactured product. The raw material — sunlight — is converted into electricity, into compute, into intelligence, and exported at a price per teraflop-hour rather than a price per tonne.

The market is large and growing. Southeast Asia’s digital economy is expanding at 15–20% per year. AI adoption across ASEAN, Japan, South Korea, India, and the Pacific is accelerating. The demand for trustworthy, affordable, sovereign-adjacent compute in the region is unmet. Australia, with the MMC delivering cheap clean power and the subsea cables delivering the output, is the natural supplier.

10. The Sovereign Compute Argument

There is a version of this story where Australia simply leases corridor land and power to Microsoft, Google, and Amazon, collects a modest rental, and calls it a win. That version is better than nothing. It is not the right version.

AI model weights — the trained parameters of a large AI system — are among the most strategically valuable assets that will exist in the coming decade. The nation, corporation, or alliance that controls the models controls the productivity of every sector that uses them: healthcare, defence, agriculture, manufacturing, education, law, finance. A country whose AI runs entirely on foreign-owned infrastructure, trained on foreign servers, subject to foreign terms of service and foreign law, has outsourced a critical layer of its sovereign capability.

The MMC compute programme should be built with sovereign ownership as a non-negotiable design requirement:

• Sovereign Compute Fund. A government-backed facility that finances Australian-owned data centre capacity on the MMC corridor — not as a subsidy but as a strategic infrastructure investment, structured similarly to the Sovereign Build Corporation itself. The fund owns the hardware. It leases compute capacity to government agencies, research institutions, and Australian enterprises at cost, and sells surplus to international clients at market rates.
• Australian model training. Critical government AI systems — Defence, intelligence, health, agriculture, transport — must be trained and run on sovereign infrastructure. This is not optional. The Five Eyes framework provides a trusted data-sharing environment, but it does not substitute for national control of the underlying compute.
• Technology transfer mandate. Any international hyperscaler granted access to MMC corridor power and land at preferential rates must commit to technology transfer — Australian engineers trained in hyperscale operations, Australian manufacturing of server components, and Australian participation in model governance.

11. What Needs to Happen

The opportunity is structural. The infrastructure is being built. Three policy decisions lock it in.

Decision 1: MMC energy mandate for AI campuses. Reserve power allocation at each Phase 1–3 corridor intersection city for AI campus use. Define the connection specification — HVDC tap-off voltage, water allocation per campus, fibre capacity guarantee — so that hyperscalers can model the infrastructure in their site selection processes. Make the offer concrete and public.

Decision 2: Sovereign Compute Fund. Establish the fund in the same legislative package as the Sovereign Build Corporation. Initial capitalisation of $10–20B — a rounding error relative to the SBC programme — acquires the first generation of sovereign GPU capacity and data centre infrastructure. The fund is self-funding within five to ten years from compute revenue.

Decision 3: Inland AI campus zoning in Phase 1–3 new towns. Designate AI campus precincts in the town plans for Phase 1–3 corridor new towns before they are built. A campus precinct has its power, water, fibre, and exclusion zone designed in from the start — not retrofitted after the town exists. This is the planning decision that makes the inland AI hub model real rather than aspirational.

Australia is not competing to host the world’s AI data centres. It is building the infrastructure that makes it the only viable choice at scale in the Indo-Pacific. That is a different and stronger position. The corridor is the offer. The offer is already being built.