The Power Imperative

The technical case for Australian desert as the optimal location for Indo-Pacific AI compute — on power, water, latency, land, and the physics of waste heat recovery. Every alternative has failed or hit a physics wall. The data is here.

1. The Power Imperative — What the World Needs and When

The global AI industry is not running short of ideas, talent, or capital. It is running short of electrons. The constraint on AI progress has shifted from algorithms and chips to the infrastructure that powers them — and the gap between what exists and what is needed is widening faster than any grid expansion programme can close it.

The numbers are large and accelerating. Global data centre electricity consumption reached approximately 460 TWh in 2022. By 2026, the International Energy Agency projects this will exceed 1,000 TWh — more than Australia’s entire national electricity consumption. By 2030, credible projections range from 2,000 to 4,000 TWh annually as AI training runs, inference workloads, and the broader digitalisation of industry compound simultaneously.

The pipeline of committed demand is already overwhelming existing supply. As of 2025, there are 160 GW of committed large-load power requests in the US data centre pipeline alone — representing 22% of total US peak electricity demand. Utilities, cities, and communities are pushing back. New campus approvals are being delayed years in Virginia, the largest data centre market in the world. Singapore effectively banned new data centres for three years. Phoenix has told operators there is no more water. Ireland’s data centres now consume 21% of national electricity and regulators are alarmed.

The three constraints are universal:

• Power: enough firm, reliable, clean electricity at the voltages and capacities hyperscale AI requires — not just when the sun shines or the wind blows, but 24 hours a day, 365 days a year
• Water: millions of litres per day for cooling at gigawatt scale — the constraint shutting down approvals from Phoenix to Singapore to Dublin
• Land: hundreds of hectares of flat, clear, politically uncontested ground with ground-bearing capacity, exclusion zones, and room to expand

Every jurisdiction where AI compute demand is highest is failing at least one of these three constraints. The hyperscalers are not building where they want to. They are building where they can — and increasingly, they cannot build at the scale and speed the AI boom demands. This is the power imperative. And it has produced a series of increasingly ambitious attempts to find the answer somewhere else.

2. What the World Has Already Tried — And Why It Failed

The idea that the world’s great deserts could power the world’s great cities is not new. It has been proposed, funded, and attempted multiple times. It keeps failing — not because the physics is wrong, but because the implementation hits political, financial, and sovereignty walls that the physics cannot resolve.

2.1 Desertec — The €400 Billion Collapse

The Desertec Industrial Initiative was launched in 2009 by a consortium of European corporations — Siemens, Deutsche Bank, Munich Re, E.ON, and others — with a vision to power 20% of Europe’s electricity from solar and wind installations across the Middle East and North Africa, connected via HVDC cables across the Mediterranean. The estimated cost was €400 billion. The concept was physically sound. The execution was not.

The consortium collapsed within four years. The problems were not engineering. They were political risk across multiple sovereign jurisdictions, financial uncertainty over which government would guarantee long-term power purchase agreements, sovereignty questions about whether North African nations were selling their strategic resources below value, and the fundamental fragility of infrastructure that crossed six or more national borders and territorial waters. By 2014, most of the founding members had withdrawn. The legacy was a concept proven correct and an implementation proven impossible.

2.2 Xlinks Morocco–UK — Rejected June 2025

Xlinks proposed a more focused version of the Desertec concept: a 10.5 GW solar and wind farm in the Moroccan Sahara, connected to the UK by 3,800 km of subsea HVDC cable at a total cost of £20–22 billion. The project attracted serious investment — Abu Dhabi’s TAQA, TotalEnergies, and Octopus Energy backed it — and was declared a project of national significance by the UK government in 2023.

In June 2025, UK Energy Minister Michael Shanks rejected the project. The stated reasons included delivery risk, operational security risk (a single cable from a foreign nation supplying 8% of UK electricity), and the preference for domestic supply chains. The unstated reason was simpler: the UK government was not willing to sign a 25-year contract-for-difference guaranteeing a price to a private developer of foreign infrastructure. Without that guarantee, the project’s financing did not work.

Xlinks has since pivoted to Sila Atlantik — a new 4,800 km cable to Germany backed by E.ON and Uniper. The same fundamental dependency on foreign government guarantees applies. The concept remains sound. The implementation remains fragile.

2.3 What These Failures Have in Common

Every desert-to-demand power project that has been proposed has shared the same structural weaknesses:

• Multiple sovereign jurisdictions. Desertec needed 27 governments to agree. Xlinks needed two. Every additional border is a veto point, a political risk, and a financing complication.
• Dependency on foreign government guarantees. No private consortium can finance a multi-billion pound intercontinental cable without a sovereign purchaser guaranteeing the revenue. Foreign governments are reluctant to sign such guarantees for infrastructure they do not control.
• Colonial extraction dynamic. Critics of the Sahara projects have consistently identified a structural problem: rich northern nations extracting cheap renewable energy from poorer southern nations, with limited value retention in the source country. This creates political opposition that compounds over time.
• No storage at scale. Xlinks’ storage was a 22.5 GWh battery — enough for a few hours of buffer. Without multi-day storage, the power delivered to the customer is intermittent and weather-dependent.

The MMC programme has none of these weaknesses. It operates within a single sovereign jurisdiction. It does not require a foreign government guarantee — it has domestic customers who fund the build. It is not extraction — it is a sovereign nation building and exporting a manufactured product. And Alice Hub PHES provides 30 TWh of storage — 1,333 times the Xlinks battery capacity.

3. Orbital AI Compute — Technical Considerations

Space-based AI infrastructure is an emerging concept with genuine long-term potential. Several well-resourced actors are pursuing orbital compute and space-based solar power, and satellite infrastructure will have an important role in the broader AI ecosystem — particularly for coverage of remote and maritime areas. However, orbital compute at gigawatt scale faces a set of technical constraints that are worth examining honestly, because they bear directly on the site selection question for near-term AI infrastructure deployment.

3.1 The 24-Hour Sun Problem

The appeal of orbital solar-powered compute is continuous sunlight — no day-night cycle, no weather. This is true at specific orbits. But the orbit that provides near-continuous sunlight and the orbit that provides low latency are not the same orbit. These two requirements are in tension.

LEO has excellent latency but suffers frequent eclipses — it is not always sunny in LEO. Battery storage can bridge the eclipse periods, and this is technically possible at lower orbits. However, battery mass becomes the governing constraint. At gigawatt-scale compute density, the battery capacity required to sustain a 1 GW cluster through a typical 35–45 minute LEO eclipse at 550 km altitude runs into hundreds of thousands of tonnes of battery mass. Even at optimistic future energy densities, the launch cost of that mass — at $100/kg to LEO — is in the tens of trillions of dollars per gigawatt of compute. Battery-sustained LEO compute is physically possible in the same sense that any sufficiently expensive engineering is possible. It is not commercially viable at the scale AI infrastructure demands.

The only orbit that provides near-continuous sunlight without battery dependence and is practically reachable is GEO at 35,786 km.

GEO has a 119 ms one-way latency. The round trip — query up to GEO, result back down — is 239 ms minimum, before a single GPU calculation begins. This is a physics constant. It cannot be engineered away. It is the time it takes light to travel 35,786 km twice.

3.2 What 239 ms Means for AI Inference

Human perception of “instant” response is approximately 100 ms. Conversational AI — the ChatGPT, Claude, Gemini style of interface that hundreds of millions of people use daily — needs total response latency under 500 ms to feel natural. A typical AI inference response involves:

• Query transmission to data centre
• GPU processing time (token generation)
• Response transmission back to user

From GEO orbit, the transmission component alone — before processing begins — consumes 239 ms. That leaves fewer than 261 ms for actual GPU processing and the return transmission. For a large language model generating a multi-paragraph response, that is not achievable. The physics floor makes interactive AI inference from GEO orbit impossible for any application where latency matters — which is most AI applications.

The orbit that provides near-continuous sunlight — GEO at 35,786 km — carries a 239 ms round-trip latency floor. The orbit that provides low latency — LEO at 550 km — requires hundreds of thousands of tonnes of battery mass to survive eclipse periods at gigawatt scale, making commercial deployment prohibitive. Solving one constraint worsens the other. Ground-based infrastructure in the Australian desert solves both simultaneously, at proven technology costs, within this decade.

3.3 The Heat Dissipation Wall

A second engineering consideration for orbital compute is heat dissipation. On Earth, data centres reject waste heat via air cooling, water cooling, and evaporative towers. In the vacuum of space, heat can only leave a structure by infrared radiation, requiring large deployable radiator panels. At gigawatt scale, these structures represent a significant engineering challenge — one that active research programmes are working to address, but which currently limits the practical scale of orbital compute facilities.

Notably, Meta’s April 2026 agreement with Overview Energy for space-based solar power illustrates a pragmatic near-term approach: solar energy collected in GEO is beamed to ground-based receivers, powering terrestrial data centres. The compute itself stays on the ground. Space solar supplements ground-based power; it does not replace ground-based compute infrastructure. This reflects a realistic assessment of where orbital technology is today — and it is entirely compatible with the Australian ground-based model described in this memo.

3.4 The Near-Term Gap

Overview Energy’s space solar system targets demonstration in 2028 and commercial delivery in 2030 — for 1 GW. The global demand pipeline is 160 GW now. The development timelines for orbital compute at meaningful scale are measured in decades, not years — which is not a criticism of the technology, but a statement of where the engineering frontier sits today.

The near-term gap between global AI power demand and available supply is a ground-based infrastructure problem that requires a ground-based infrastructure solution. Australia’s MMC corridor — delivering 40 GW of firm, dispatchable, clean power from Alice Hub PHES and desert solar — is that solution, available within this decade using proven technology. Space-based solar and orbital compute will complement this infrastructure as the technology matures. The Australian desert provides the foundation that bridges the gap.

4. Why Australia Is the Answer

Australia holds a combination of assets that no other nation on Earth can match for the specific requirements of gigawatt-scale AI infrastructure. It is not one advantage. It is five, converging in the same place.

4.1 The Solar Resource

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, comparable to the Atacama Desert in Chile and the Sahara in North Africa. 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 — nearly double Australia’s current total generation capacity, available for export after domestic needs are met.

4.2 The Storage — Alice Hub PHES

Solar generation is intermittent. AI inference is not. The gap between them is filled by the Alice Hub pumped hydro energy storage system at 40 GW output and approximately 30 TWh of storage capacity. To put that in perspective: the Xlinks Morocco proposal included a 22.5 GWh battery. Alice Hub stores 1,333 times more energy. It can supply Australia’s entire current electricity demand for several weeks from storage alone. Every AI campus on the MMC corridor receives firm, despatchable, 24/7 power — not solar when available, but power on demand, backed by the largest energy storage system ever built.

→ See: Alice Hub PHES — MMA Memo 5.

4.3 Water — The Constraint Nobody Else Has Solved

Water is the constraint that is quietly shutting down data centre approvals around the world. Phoenix has told developers there is no more water. Singapore’s effective ban on new data centres was partly water-driven. Dublin is implementing water use restrictions on new facilities.

The Alice Hub aqueduct delivers up to 25,000 GL of water annually along the MMC corridor spine — sourced from northern flood harvesting and stored in the MacDonnell Ranges reservoir. Every AI campus built along the Phase 1–3 corridor routes has access to aqueduct water for industrial cooling. This is not a future plan. It is a designed feature of the MMC programme. No other desert solar zone on Earth has a purpose-built water supply at this scale. The Sahara projects face genuine water scarcity for panel cleaning. The Australian interior, with the aqueduct, does not.

A 1 GW AI campus using evaporative cooling consumes approximately 1–3 billion litres of water per year. At peak aqueduct capacity, the Alice Hub system can supply 50–150 such campuses simultaneously — without competing with agricultural or municipal water users along the route.

4.4 Fibre — 19 ms to Singapore

The MMC corridor carries a fibre optic data spine the length of every corridor, connecting every campus site to every other and to the coastal cable landing stations. The latency from an inland Australian data centre to the major Asian markets on this fibre — and the subsea cables that extend from the coastal terminals — is determined by the speed of light in fibre (approximately 200,000 km/s) and the physical distance.

The MMC data spine runs from the inland corridor sites north to Darwin, where the Inligo ACC-1 subsea cable (under development, landing 2027–28, capacity 256 Tb/s+) provides the onward connection to Singapore and the broader Asian subsea network. The Alice Springs–Darwin fibre leg is approximately 1,500 km — adding approximately 7.5 ms to the one-way latency before the subsea cable begins.

Every major Asian AI market is reachable from Alice Springs via Darwin in under 75 ms RTT — leaving more than 425 ms inside a standard 500 ms conversational AI response budget for actual GPU processing. GEO orbit’s 239 ms RTT physics floor consumes almost half that budget before a single calculation begins. The Darwin fibre spine via Inligo ACC-1 is not a future plan — the cable landing is already under development for 2027–28. The latency advantage is real, quantified, and available on the MMC’s build timeline.

4.5 Sovereign Land — No Colonial Dynamic

The Sahara projects face a structural political problem that has undermined every attempt to implement them: they involve wealthy northern nations extracting cheap energy from less wealthy southern nations, with limited value retention in the source country. Critics have accurately described this as a “familiar colonial scheme” — the same language used historically to describe the extraction of raw materials from the Global South for manufacture in the industrial North.

Australia has no such dynamic. Australia is a sovereign nation building infrastructure on its own Crown land to serve its own domestic market first — and exporting surplus as a manufactured product at commercial prices. The revenue from AI campus tenancy, domestic grid supply, and Asian export flows to Australian workers, Australian businesses, and Australian government revenue. There is no extraction. There is no dependency. There is no colonial dimension. The political sustainability of the programme is structurally different from every Saharan alternative.

5. Why the Desert Specifically — Not Just Australia

Australia is the right country. The Australian desert is the right location within it. These are separate arguments and both matter. A coastal Australian data centre does not deliver the same advantages as an inland desert campus. The desert is not a compromise — it is the optimum.

5.1 The Solar Resource Is Not Uniform

Australian solar irradiance varies substantially by location. Coastal and temperate zones receive 1,600–1,800 kWh/m²/year — excellent by global standards, but meaningfully below the 2,200–2,800 kWh/m²/year of the inland desert. A solar farm in the Pilbara or central Australia generates 20–40% more electricity from the same panel area than a coastal installation. At gigawatt scale, that difference is billions of dollars of additional generation value from the same capital investment. The desert solar advantage is not marginal — it is the difference between a project that is marginally viable and one that is decisively commercial.

5.2 The Latency Advantage Is an Inland Advantage

The fibre latency argument in Section 4.4 applies specifically to inland desert locations connected via the MMC data spine to Darwin and the Inligo ACC-1 subsea cable. A coastal data centre in Sydney or Perth, while in Australia, does not replicate this advantage. Sydney to Singapore via existing cable routes is approximately 65 ms one-way — worse than Alice Springs via Darwin at 23 ms. The short overland distance from the inland desert to Darwin, combined with the direct Darwin–Singapore submarine path, produces latency that a coastal installation cannot match. The inland location on the corridor spine is the source of the latency advantage, not Australian geography generically.

5.3 The Cold Nights Are an Asset

Data centres produce heat as their primary waste product. In tropical and coastal environments, rejecting that heat is expensive — the ambient temperature is high, cooling towers work harder, and water evaporates faster. In the Australian inland desert, the diurnal temperature swing is among the largest on Earth: daytime temperatures exceed 40°C, while desert winter nights drop to 3.9°C average. This swing is not a hardship — it is a thermodynamic asset.

Organic Rankine Cycle (ORC) systems recover waste heat as electricity using the temperature differential between the hot cooling loop (60°C from immersion cooling) and the cold night air. At 3.9°C winter nights, a 1 GW AI campus recovers 76 MW via ORC — rising to 107 MW with solar-thermal boosting of the hot side. This recovery is impossible in Singapore (27°C nights, 33°C delta), marginal in Sydney, and excellent only in the desert. A gigawatt AI campus in the Australian desert recovers the equivalent of a small power station’s output from waste heat that tropical campuses simply expel. → See: The Desert Heat Engine — MMA Memo 9.

5.4 The Land Is Genuinely Free

“Cheap land” is available in many remote locations globally. Genuinely uncontested, large-scale Crown land with no competing agricultural use, no visual amenity concerns, no community opposition, and no planning authority that can veto a project for aesthetic reasons is not available in most of those locations. The Australian desert interior is that land. A 500-hectare AI campus precinct in the Pilbara or central Australia can be zoned, titled, and broken ground on without displacing a farming family, a coastal community, or a planning objection. That is not a trivial advantage — it is the difference between a five-year planning process and a twelve-month land tenure decision.

5.5 The Five Desert Advantages Combined

No other location on Earth combines all five desert-specific advantages simultaneously:

The Australian desert is not the best desert. It is the only desert that solves all five constraints simultaneously. That combination — solar, PHES, water, fibre, and land — is what makes it the optimal location for Indo-Pacific AI compute infrastructure, not just a good one.

6. The Indo-Pacific Demand Picture

The AI power problem is not primarily a Western problem. The largest growth in AI adoption, EV fleet electrification, and industrial decarbonisation over the next decade will occur in Asia — in markets that are energy-import dependent, land-constrained, and facing exactly the same three walls as the US and Europe.

Singapore, Japan, South Korea, and Indonesia are the four markets that define the Indo-Pacific compute demand picture. All four are energy-import dependent. All four are land-constrained. Singapore has already banned new data centres due to power and water limits. Japan and South Korea are actively seeking offshore compute and energy supply agreements. Indonesia has the population and digital economy growth but not the clean power infrastructure.

All four are within 36 ms of Alice Springs on fibre. All four would pay a significant premium for sovereign-adjacent, clean, reliable, cheap compute from a Five-Eyes-aligned jurisdiction with no political risk of supply interruption. No other location in the Indo-Pacific offers this combination. Australia does — once the MMC corridor is built.

7. The Data Centre Trifecta — How the MMC Delivers All Three

The MMC five-track viaduct corridor is not just a transport system. Every pylon carries four services simultaneously: HVDC power cables, the fibre optic data spine, the aqueduct water pipe, and the freight and passenger transport infrastructure. Every AI campus built along the corridor route connects to all four services at a single connection point.

For a hyperscale operator, this is extraordinary. Instead of spending years negotiating separate grid connections, water licences, fibre routes, and road access — building bespoke campus infrastructure from scratch — the MMC corridor delivers power, water, fibre, and transport to a ready-zoned campus site in a single package. The corridor is the infrastructure. The campus plugs in.

Power: 4–7¢/kWh wholesale from desert solar and Alice Hub PHES, delivered via the HVDC spine. Three to five times cheaper than Singapore or Japan. Firm and dispatchable 24/7 — not solar when available but power on demand backed by 30 TWh of storage.

Water: Aqueduct supply from Alice Hub at industrial rates. The constraint that is blocking data centre approvals from Phoenix to Dublin is a design feature of the MMC programme. No other desert solar zone on Earth has this. At 25,000 GL annual aqueduct capacity, the system can supply 50–150 gigawatt-class AI campuses simultaneously.

Fibre: The MMC data spine running the length of every corridor connects every campus to every other campus, to the 211 corridor towns and cities, and to the coastal subsea cable landing stations. Latency to Singapore: 19 ms. To Tokyo: 34 ms. To Jakarta: 16 ms. Better than GEO orbit. Better than any alternative inland location.

The business model: AI campus tenants pay for all three services at commercial rates that are still dramatically cheaper than anywhere else in the Indo-Pacific. A 1 GW campus paying 5¢/kWh generates approximately $440M per year in power revenue alone. Ten campuses along the Phase 1–3 corridors generate $4–5B per year in power, water, and fibre revenue. That revenue stream services the corridor construction bonds directly — the MMC does not wait for government funding. It earns its construction cost from anchor tenants before the export cables are even planned.

The fourth advantage — heat reclamation. Desert AI campuses add a fourth revenue and efficiency stream that no other location offers: ORC (Organic Rankine Cycle) waste heat recovery using the desert diurnal temperature swing as the thermodynamic engine. Australian inland desert winter nights reach 3.9°C — creating a hot-cold differential of 56°C with the 60°C immersion cooling output of a modern AI campus. On winter nights, ORC systems recover 76 MW per GW of compute. With solar-thermal boosting to 90°C on the hot side, this rises to 107 MW per GW — more than 10% of the campus electrical input returned as recovered electricity. No tropical or coastal data centre can do this. The cold night air is as much an asset as the solar resource. → See: The Desert Heat Engine — MMA Memo 9.

8. The Consortium Model — Private First, Government Follows

Every large infrastructure programme in Australian history has been government-led. The phone network. The Snowy. The NBN. Government decides, government funds, government builds. The result is characteristic: slow, expensive, politically contested at every budget cycle, and ultimately delivered late and over budget.

The MMC programme does not follow this model. It follows the SpaceX model: a private consortium builds the first phase, demonstrates the technology and the economics, and government co-investment and regulatory support follow demonstrated progress rather than preceding it.

The logic is simple. Government cannot commit $2.1 trillion to a programme it has not yet seen work. But a private consortium can build Phase 1 — the Port Hedland–Mackay corridor running through the Pilbara and Gulf Country, the zone of highest solar irradiance and the home of the inland AI campus precincts — on private capital, demonstrating the foundation system, the P#7 casting technology, the HVDC transmission performance, and the AI campus economics at real scale. Once Phase 1 is operational and delivering power to AI campuses at 5¢/kWh while the competing grids charge 25¢/kWh, the government case for co-investing in the remaining corridors writes itself.

The $20M prototype programme — validating the foundation caisson system, the P#7 casting die, and the pylon structure — is the first step that makes the full programme investable. Once prototype results are in hand, the consortium formation and construction finance follow demonstrated engineering performance, not a promise.

The full consortium needs four types of participant:

• Engineering and construction partners — Australian and international firms with the steel, precast concrete, civil works, and corridor-scale construction capability required for Phase 1
• Hyperscaler anchor tenants — one or two AI companies committing to inland campus sites in exchange for preferential power and water pricing; their power purchase agreements are the revenue that services the construction bonds
• Sovereign wealth and infrastructure funds — the patient capital that finances 20–30 year infrastructure with stable, contracted revenue streams
• Asian government or utility co-investors — Japan, South Korea, or Singapore taking an equity stake in the subsea cable and export capacity in exchange for long-term power supply agreements

8.1 The Three-Tier Sovereignty Structure

Foreign capital is essential to build at speed. Foreign ownership of critical Australian infrastructure is not acceptable. The three-tier consortium structure separates these two requirements cleanly.

This structure ensures Australian sovereignty over the land and the data pathways at all times, while allowing foreign capital to fund the hardware and tenant fit-out. The Sovereign Corridor Trust holds the strategic assets. Asian tenants own only the equipment they brought. The arrangement mirrors how airport infrastructure works globally: the land and runways are sovereign, the aircraft and cargo are private.

8.2 The Financial Architecture

Traditional infrastructure yields 4–6% annual returns from slow-moving assets like toll roads or water pipelines. AI hyperscalers run continuous, 24/7 workloads at near 100% load factors — the highest utilisation rate of any infrastructure customer. The bundled utility tariff (power + water + fibre) on this demand profile yields an estimated project IRR of 14–18%. At that return rate, the initial construction debt is fully amortised within 7–9 years — leaving Australia with a permanent, debt-free sovereign revenue stream and dominant digital infrastructure in the Indo-Pacific.

The Sovereign Corridor Bonds — issued by the Trust — are offered to Australian superannuation funds with first right of refusal. Australian superannuation holders become financial stakeholders in the corridor, capturing 8–12% returns funded directly by Asian technology revenues. This is the political framing that makes the programme unassailable domestically: everyday working Australians own the bonds that fund the infrastructure that earns rent from Asian hyperscalers. The economic nationalism writes itself.

The revenue model is bankable from day one without government subsidies or contracts-for-difference — the structural failure that killed Xlinks. Ten 1 GW AI campuses generate $450–800M per year in power, water, and fibre revenue. That services the construction bonds at commercial rates. Government co-investment accelerates the full six-corridor build — but it is not required to make Phase 1 work. Funding mechanism options are detailed on the Funding pillar.

The Case in Summary

The world needs gigawatt-scale, firm, clean, low-latency AI compute in the Indo-Pacific. Every alternative has failed or hit a physics wall. The Australian desert solves all four constraints — power, water, land, and latency — simultaneously and at a scale no other location can match.

The solar resource is the best on Earth. The Alice Hub PHES provides 30 TWh of firm despatchable storage — 1,333 times the Xlinks battery. The aqueduct delivers industrial-scale cooling water where every other desert solar zone is dry. The Darwin fibre spine reaches Singapore in 46 ms RTT and Tokyo in 67 ms — well inside the conversational AI response budget, and definitively better than GEO orbit’s 239 ms physics floor. The desert’s 3.9°C winter nights recover 107 MW per GW of compute via ORC — a fourth efficiency advantage that no tropical or coastal site can replicate. And the Crown land is vast, uncontested, and sovereign.

Australian desert AI infrastructure and orbital compute are not competing visions — they are complementary layers of the same system. Satellites provide coverage, redundancy, and remote connectivity. Ground-based desert campuses provide the firm power, the low latency, the water, and the scale that the Indo-Pacific needs now. The MMC corridor is the ground layer. It is ready to be built. The physics and the geography say build it here.