scx.ai logo

Strategy Blog

Your Data Lives in Sydney. That Doesn't Make Your AI Sovereign.

Hyperscalers are selling sovereign cloud regions on Australian soil. SCX.ai CEO David Keane explains why residency is not sovereignty, and the five questions to ask before you sign.

By David Keane, Founder and CEO, SCX.ai6 min read

Open any hyperscaler pitch deck from the last few months and you will find the word sovereign on the cover. A new cloud region in Sydney. Your data never leaves Australia. Sovereignty achieved.

Except it isn't.

There is a difference between where data sits and who controls it. Australian organisations are about to spend enormous sums on AI infrastructure, and too many of them are going to buy the first thing while believing they bought the second.

Data residency is the easy part. It means the physical bytes are stored in Australia. A Sydney availability zone gets you residency, and it is worth having.

Sovereignty asks harder questions. Who owns the infrastructure? Who operates it? Under whose law does the operator answer when a government comes asking? Can someone be compelled to hand over access by a court in another country, without anyone in Australia ever finding out?

Here is the uncomfortable part. The United States CLOUD Act allows US authorities to compel data from US-headquartered providers regardless of where the data physically resides. If the company operating your AI platform answers to American law, jurisdiction follows the corporate chart, not the postcode. A data centre in Sydney does not override that.

Australian lawyers have been making this point quietly for a while. Gilbert + Tobin put it plainly: digital sovereignty is about control and governance, not geography. The team at NCS say it even more bluntly. Sovereign capability is not a data centre.

2026 turned this from a policy debate into a board problem

Three things happened this year that changed the stakes.

First, regulation got teeth. The EU AI Act's high-risk obligations became enforceable in August. Australia moved the same direction, with the National AI Plan shifting from voluntary ethics guidance toward mandatory standards covering onshore capability, data protection and AI supply chain independence. Government agencies are being directed to keep AI processing onshore under strict security controls.

Second, the compliance stack got heavier. Regulated Australian organisations now carry obligations under the SOCI Act, APRA's CPS 230 and IRAP assessment requirements. None of those can be discharged by a region sticker on a marketing page.

Third, agentic AI changed what processing means. Twelve months ago, enterprise AI meant storing documents and sending the occasional prompt. Now autonomous agents read everything, act everywhere, and hold the most sensitive context your organisation owns in their working memory. The sensitive surface is no longer the database. It is the context window. And it gets processed continuously, wherever your provider decides to run it.

That is why residency is no longer enough. The question is not where your data sleeps. It is where your data gets processed, by whom, under whose law, every single day.

Five questions worth asking before you sign

If the word sovereign appears in a provider's brochure, here is what I would ask.

One. Who owns the infrastructure? Not who resells it, not who brands it. Who owns it.

Two. Under whose law does the operating company answer? If a foreign court issues an order against the provider, does anyone in Australia have to be told?

Three. Who are the people with privileged access to your workloads, and what jurisdiction are they accountable to?

Four. Where does inference happen? Not storage. Processing. Every token your organisation generates is a fresh act of processing, and each one happens somewhere.

Five. Can the provider give you independent assurance through Australian-governed assessment programs, as an Australian-governed entity?

If the honest answers are "our Sydney region, their Delaware parent, their offshore operations team", you have residency. You do not have sovereignty. That is fine for some workloads. It is not fine for the workloads that carry customer data, health records, or anything a regulator would call critical.

Sovereignty also has to be affordable

The standard objection is that sovereign AI costs more and runs slower. For years that was true, because sovereignty was bolted onto hardware built for a different job.

Purpose-built silicon changes the equation. We run inference on SambaNova's RDU architecture because inference is the workload Australia actually needs, and it is the workload where efficiency matters most. Intelligence per watt is not a slogan for us. It is the difference between sovereign AI being a premium niche and sovereign AI being the rational default.

That is the whole reason we built SCX the way we did. Australian owned. Operated under Australian law. Onshore inference with economics that make the sovereign choice a business decision rather than a patriotic one.

The takeaway

Hyperscalers are not wrong to build here. Their Australian regions are genuinely useful for plenty of workloads. But a region is a location. Sovereignty is control. Conflating the two is comfortable right now, and it will be expensive the first time it matters, because the first time it matters will be the worst possible moment.

Australia will spend more on AI infrastructure this decade than it has spent on anything digital before it. The question that should keep boards awake is not how much compute we host. It is who controls it.

Residency is a line on a map. Sovereignty is who answers when it counts.

Buy accordingly.

David Keane is the Founder and CEO of SCX.ai, Australia's sovereign AI infrastructure company.

Sources

Related Topics

data sovereigntysovereign AIdata residencyCLOUD Acthyperscalersagentic AIAustraliaSCX.aiNational AI Planinference
Your Data Lives in Sydney. That Doesn't Make Your AI Sovereign.