The Same Bet, No Rocket Required

Firstly, congrats to the entire SpaceX team — listing SPCX.. incredible result and important for the future of humanity.

And as part of that goal SpaceX wants to put AI compute in orbit.

At first glance, it sounds crazy.

But the more you look at the problem, the more you realise the diagnosis is right.

The Power Problem

AI is no longer just constrained by chips. It is constrained by power, cooling, water, grid access and jurisdiction. The numbers behind that are sobering. The International Energy Agency expects data-centre electricity demand to more than double by 2030 — from around 415 terawatt-hours today to roughly 945 TWh, which is about what all of Japan consumes in a year. IDC expects annual spending on AI infrastructure to pass US$200 billion by 2028. And every one of those facilities has to be powered, cooled and connected somewhere real, on a grid that is already under strain.

That is why people are looking to space.

Solar is abundant and almost constant in the right orbit. There are no local planning objections. No multi-year grid connection queue. No cooling-tower water draw in a country that knows what drought looks like. No fight over whether a data centre should get power before a city, a mine, or a factory.

It is a bold idea — and the instinct behind it is correct.

But Orbit Also Creates Its Own Problems

You need an enormous, sustained launch program — Starship flying again and again, for years, just to put the mass up there. You need to solve heat rejection in a vacuum, where you cannot air-cool or water-cool anything. On Earth, "no water" is a feature. In space, it is the core engineering problem: with no air and no water, the only way to shed heat is to radiate it, and at gigawatt scale that means radiator structures of a size nobody has ever built or operated. It is widely considered the single hardest part of the whole vision. You need to handle radiation hardening, service hardware you cannot physically reach, and build and operate one of the most complex and capital-intensive infrastructure platforms ever attempted.

And then you get to the question nobody really wants to answer: whose laws govern AI infrastructure in orbit? Whose privacy regime, whose security framework, whose courts? There is no clean answer to that — and for regulated data, ambiguity is not a minor detail. It is disqualifying.

At SCX.ai, We Are Making the Same Bet — But on the Ground

We believe the future of AI infrastructure is sovereign, renewable-powered, efficient, and close enough to the markets it serves. The same diagnosis SpaceX has made about power and cooling — solved without leaving the planet.

What We Have Built

Node 1 is live in Sydney now. Inside Equinix SY5 — one of the most secure, interconnected data-centre campuses in the country. Real customers. Real workloads. Real tokens. We have capacity for over a token a month already in place, scaling rack by rack as demand grows. Not a render. Not a 2030 roadmap. Running today.

Node 2 is going into Rural Australia — one of the most renewable-rich grids in the world, where wind and solar regularly meet the entire state's demand. We are targeting 100% renewable energy. We do not need orbit to find abundant clean power. We just need to put compute where that power already is.

Why RDUs, Not GPUs

We are using SambaNova's RDU-based systems purpose-built for inference — not general-purpose GPUs repurposed for AI. The difference matters.

Our SambaNova SN40L racks draw approximately 10 kW per rack. A comparable GPU rack draws two to three times that. In our deployments, we are seeing significantly more tokens per watt than GPU-only infrastructure. That is not a marketing claim — it is a physical consequence of an architecture designed from the ground up to eliminate the data movement that wastes most of the energy in a GPU-based system.

Air-cooled. No liquid cooling loops. No cooling towers. No costly retrofit. We deploy as an overlay inside existing Tier-3 and Tier-4 colocation infrastructure — which means we do not need to build greenfield, and we do not need to wait for new grid capacity to be negotiated and connected.

Sovereignty: The Advantage With No Orbital Equivalent

An orbital data centre — whose laws apply? Whose security framework? Whose courts? There is no clean answer. For regulated data, space does not solve the sovereignty problem. It makes it worse.

Regulated sectors — finance, health, government, defence — are increasingly required to keep data onshore. The US CLOUD Act means that any data held by a US hyperscaler, or in a data centre owned by a US company, and for sure a SpaceX operated satellite can be subject to US law enforcement demands without oversight. That is not a theoretical risk. It is a structural feature of how US cloud infrastructure works.

SCX.ai is an Australian company, governed by Australian law, subject to Australian courts, aligned with APRA CPS 234, IRAP certification, ISO 27001 and SOC 2. For regulated enterprises, that is not a nice-to-have. For many of them, it is the entire reason to choose a sovereign provider.

And Australia is a trusted global partner and we believe that trust and transparency is key for the future of reliable, truly sovereign AI.

On Latency

People sometimes say Australia is too far away from the major AI markets.

But for inference, distance is often less important than people think. The model's own processing time — frequently hundreds of milliseconds, sometimes seconds — can dwarf the few milliseconds light spends in transit. The industry already makes exactly this argument when it talks about compute in orbit. If latency to LEO can be made acceptable, latency to Australia is not the problem people imagine it to be.

For Australian users and Australian data, served from Australian soil, latency is not higher than the alternative. It is lower.

The Macro Context Both Bets Are Riding

IDC estimates annual AI-dedicated infrastructure spending will top US$200 billion by 2028. The IEA puts data-centre electricity consumption at 415 TWh today — roughly 1.5% of global electricity — and projects it will exceed 945 TWh by 2030. AI data-centre water consumption for cooling has become a flashpoint in water-stressed regions. Grid interconnection queues in the US and Europe now run to years.

Both orbit and sovereign-renewable-Australia are escape hatches from the same bottleneck.

The Real Question

The real question is not whether AI compute should go to space. Some of it probably will, eventually.

The real question is whether we need to wait for that future before building better infrastructure today.

We do not.

At SCX.ai, we are building sovereign AI infrastructure on Australian soil — powered by renewables, designed for inference, governed by Australian law, and available now.

Same bet.

No rocket required.

SCX.ai — Southern Cross AI Holdings Limited (ACN 688 872 095). Node 1 live at Equinix SY5, Sydney. Node 2 targeted for 2026. Learn more at scx.ai