The case for sovereign AI infrastructure in Australia.

Most Australian businesses using AI today are sending their prompts, their data, and their context to servers in Virginia. They didn't decide to do that. They just used the default.

The default is a decision

Every API call to OpenAI, Anthropic via api.anthropic.com, or most managed AI platforms ends up routed through US infrastructure. For consumer use that's fine. For a law firm, a consultancy, or a government contractor, it's a decision that deserves more thought than it usually gets.

What sovereign actually means

Sovereign AI doesn't mean building your own model. It means choosing infrastructure where the data path, the storage, and the inference all happen in a jurisdiction you control. In Australia, that mostly means AWS Bedrock in ap-southeast-2 or Azure OpenAI in Australia East.

The models are the same. The capability is the same. What changes is where the bytes live and which laws apply when something goes wrong.

When it matters

• Anything covered by the Privacy Act with personal information at scale
• Commercially sensitive deal data, M&A work, legal strategy
• Government and defence-adjacent work
• Anything where a client contract specifies data location

When it doesn't

Marketing copy. Code generation. Internal experimentation. The cost of using sovereign infrastructure is real — fewer model options, slightly higher latency, more setup — and you should only pay it where it matters.

The practical answer

For most clients I work with, the right pattern is split: sensitive workloads on Bedrock in Sydney, everything else on whatever's fastest and cheapest. The architecture has to support both, and the team has to know which is which. That's the work.