Local presence has become a selling point in Australian AI deals, and it is doing a lot of unexamined work. A vendor says it has an Australian team and a buyer hears data sovereignty, local accountability and someone who answers the phone at 3pm on a Tuesday. Those are three different things and only some of them follow from an office.
This is a practical guide to testing the claim. It applies whether you are looking at Claude, ChatGPT or anything else. Where the two platforms are genuinely comparable we say so, because a comparison that manufactures a gap is worse than no comparison at all.
What does local presence actually mean for an AU business?
It means five separate things that vendors tend to bundle into one word. Where your prompts and outputs are stored, where the model actually runs, which legal entity you sign with, who supports you inside Australian business hours, and how the vendor handles Australian Privacy Principle obligations when personal information is involved. Each one is independently verifiable, each one can be true while the others are false, and only the last two are affected much by whether the vendor has staff in Sydney.
Once you split them apart, most procurement arguments about local presence resolve quickly, because you stop debating a feeling and start asking for documents.
The checkable list
Take this to any AI vendor. The questions are deliberately blunt and the answers should be in writing.
| What to test | The question to ask | What a good answer looks like |
|---|---|---|
| Data residency | Where are prompts, outputs and conversation history stored for this product? | A named region, in writing, tied to the specific product we are buying |
| Inference residency | Where does the model actually run when we send a request? | A named region, stated separately from storage, not assumed to match it |
| Contracting entity | Which entity do we contract with, and under which governing law? | A named entity and jurisdiction, plus the DPA that sits under it |
| Support and escalation | What are the support hours in AEST and what is the escalation path? | Published hours, a severity matrix, and a named path for a P1 incident |
| Privacy Act posture | How do you handle APP 8 cross-border disclosure for our data? | A documented position, not a verbal assurance from a salesperson |
The second row catches more people than any other. Storage and processing are separate controls, and a vendor can honestly say your data is stored in Australia while requests are processed elsewhere. Ask about both, separately, and get both in the contract.
Where Claude stands, on the record
Anthropic announced on 10 March 2026 that it would open an office in Sydney, its fourth in Asia-Pacific after Tokyo, Bengaluru and Seoul, covering both Australia and New Zealand. In the same announcement it named Canva, Quantium and Commonwealth Bank of Australia as existing customers, and said Australia and New Zealand rank fourth and eighth globally in Claude usage relative to population.
Two lines in that announcement matter more than the office itself. Anthropic said it is exploring adding local compute capacity through third-party partners in Australia using infrastructure already in place, and that this is among the most consistent requests it hears from Australian enterprises and government agencies with data residency requirements. That is a company describing a gap it intends to close, not a company claiming to have closed it. We covered the ground-level implications when the office opened in our piece on what the Sydney office means.
On residency itself, Anthropic's published position is that regional data residency is available across Asia-Pacific including Australia, delivered through AWS Bedrock, Google Cloud Vertex and Microsoft Foundry. The important qualifier is that this is a property of the deployment path, not of the brand.
The residency mistake almost everyone makes
Residency attaches to how you access the model, not to the logo on the login screen. The same vendor can offer you an onshore deployment through a cloud provider and a consumer subscription that routes offshore, and both are honestly described as that vendor's product.
Ask per product and per surface. A Claude deployment running through a cloud provider's Australian region is a different arrangement from a seat on the consumer app, and the answer to a residency question can differ between them.
Ask which certifications attach to which path. An assurance that exists via a cloud provider's Australian region is not the same as an assurance the model vendor holds directly. This distinction gets blurred constantly in sales decks.
Ask what happens to logs, abuse monitoring and support access, not just to your prompts. These are the pathways that most often sit outside the region you contracted for.
Get the answer for the product you are actually buying this quarter, not the product line. Feature and region coverage move, so a claim verified in June is not a claim verified in September.
If onshore processing is a hard requirement rather than a preference, that requirement mostly steers you toward a cloud deployment path rather than toward a particular vendor. We have written up what that path buys and does not buy in our note on onshore data residency for Claude.
Where Claude and ChatGPT are comparable
Being straight about this matters more than winning a point. Both vendors are US-headquartered, both offer enterprise agreements with data processing addenda, both publish security certifications, both make regional deployment available through major cloud platforms, and neither is an Australian company. If your procurement team is looking for a clean sovereignty distinction between the two, it is unlikely to find one at the level of corporate structure.
We could not load OpenAI's own residency and enterprise policy pages directly while researching this piece, so we are not going to characterise its Australian position for you. That is not a criticism of OpenAI, it is a limit on what we verified. Put the five questions in the table to both vendors and compare the written answers you get back. A vendor that will not answer them in writing has told you something useful either way.
What local presence does genuinely change
An office in Sydney does not move your data. What it does change is real but narrower: support inside Australian business hours rather than overnight, a partner and solutions team that has seen Australian regulatory conversations before, engagement with Australian policy and procurement processes, and a local escalation path when something breaks during your working day.
For a business running a production workload, that last one has a dollar value. If an incident costs you a day of a ten-person team, that is roughly $8,000 in loaded time before you count the client consequences. Whether a vendor can start working the problem at 9am AEST or at 2am the next morning is a commercial question, not a sovereignty one, and it is fair to weigh it.
How to decide
Write your five answers down for each vendor before anyone builds a business case. If residency is a hard requirement, the deployment path decides more than the brand does. If it is a preference, support hours and local engagement are legitimate tiebreakers. If neither applies to your workload, choose on model quality and how well the tooling fits the work you actually do, which is where most of the value sits anyway.
Our AI readiness assessment runs exactly this exercise for Australian buyers, and you can see how we scope platform selection work on our services page.
If you are midway through a vendor comparison and want someone to pressure-test the answers you have been given, book a time with us. We will tell you plainly where the difference is real and where it is marketing.
