Claude will run from Queensland. Does it matter to you?
Anthropic's Queensland lease starts in 2027 and promises no Australian data residency. Today Claude's API runs global or US only, so ask every AI vendor six questions in writing.
In September 2026 Anthropic signed a lease to run Claude from the Western Downs Digital Park, a data centre campus still to be built near Dalby, about 250 kilometres north-west of Brisbane. Reuters broke the story on 16 September, and the ABC reported on 18 September that Anthropic plans to start using the first building in 2027, with the lease still subject to Foreign Investment Review Board approval. On 13 October, Bloomberg reported Anthropic's envoy Jeff Bleich saying the company is looking at more Australian sites.
For a 50 to 250 person Australian business, it changes nothing about where your data goes today. At the time of writing, 16 October 2026, Anthropic's own documentation lists two choices for where Claude API requests run: anywhere ("global") or the United States only. The only place it stores API data at rest is the US. Nothing in the announcement promises Australian data residency, and nobody at Anthropic has said the Queensland site will offer it.
What it should change is the questions you ask. If you use Claude, ChatGPT, Copilot or anything built on them, you should be able to answer six things in writing: where your data is stored, where it is processed, whether it trains a model, how long it is kept, who else touches it, and which regional options exist right now. The table below gives you those questions and what a good answer sounds like.
What was actually announced, and what was not
We are a Brisbane business, so this one is close to home. Here is what the reporting supports, and nothing more:
- The site. The ABC describes a 725.5 hectare site at Kogan, 37 kilometres north-west of Dalby, developed by Singapore based Zerra DC. Full construction is expected to take four to six years.
- The tenant. Anthropic has leased Building One and plans to start using it in 2027. The tenant of each building arranges its own grid connection, and Anthropic plans to seek long-term renewable power agreements.
- The work it will do. Reuters, citing an unnamed source, reported the facility will be used for inference, which means answering requests with models that are already trained. Anthropic and Zerra DC declined to comment to Reuters.
- The condition. The lease is subject to Foreign Investment Review Board approval.
- More sites. Speaking at the Australian Financial Review's energy and climate summit in Sydney, Bleich said Anthropic prefers regional or former industrial sites running on renewable power, and that he did not think any state would be left out.
This follows Anthropic opening a Sydney office, announced on 10 March 2026, where it said it was "exploring opportunities to expand our compute capacity in Australia".
What none of it says: that Australian customers' prompts will be processed in Queensland, that data will be stored here, or that a new Australian region option is coming. Those may follow. Until Anthropic publishes them, treat them as unannounced.
- 10 March 2026Anthropic announces its Sydney office and says it is exploring more compute capacity in Australia.
- 16 September 2026Reuters reports Anthropic has signed a lease at the Western Downs Digital Park near Dalby, for inference.
- 18 September 2026The ABC reports Anthropic plans to start using Building One in 2027, subject to Foreign Investment Review Board approval.
- 13 October 2026Jeff Bleich says Anthropic is looking at more Australian sites.
- PendingForeign Investment Review Board approval of the lease.
- 2027, plannedAnthropic starts using Building One.
- Not announcedAn Australian option for where Claude API requests run or where data is stored.
Sources: Anthropic, 10 March 2026; Reuters, 16 September 2026; ABC News, 18 September 2026; Bloomberg, 13 October 2026. Dashed steps have not happened.
Does a data centre in Queensland mean data residency in Australia?
No. Data residency is a promise about your data, written into a contract or a product setting. A building is only where the computers sit.
Claude's API documentation splits the question in two. "Inference geo" controls where the model runs for each request. "Workspace geo" controls where data is stored at rest and where things like file processing happen. As of today:
- Inference geo offers "global" (the default, which can run anywhere) or "us". There is no Australian option.
- Workspace geo offers "us" only, and it cannot be changed after a workspace is created.
- US-only inference costs 1.1 times the standard token price on Claude 4.6 and later models.
A Queensland data centre could one day sit behind a new option on that list. It could equally serve global traffic with no setting you can choose. For your privacy obligations, the setting and the contract are what count.
Can you keep Claude processing in Australia today?
Partly, through Amazon Bedrock, Amazon's AI service. AWS offers Australian "geographic cross-Region inference" profiles for some Claude models. AWS launched the first ones in October 2025 for Claude Sonnet 4.5 and Haiku 4.5, routing requests between Sydney (ap-southeast-2) and Melbourne (ap-southeast-4), and said processing stays within Australia for the whole request.
Three catches worth knowing before you plan around it:
- Model coverage varies. Not every Claude model has an Australian profile. Check the model's page in the Bedrock documentation before you commit.
- Check the actual routes. AWS's own guidance says not to rely on the "au." prefix alone, and to confirm the destination regions with a GetInferenceProfile request from each region you call from.
- Different processor, different paperwork. On Bedrock, Anthropic's documentation says the cloud provider is the data processor. Your contract and retention terms are with AWS, which suits businesses that already run on AWS.
Our post on where your data actually goes when Claude reads your CRM compares the direct API, Bedrock and Google Cloud in more detail, so we will not repeat it here.
Sources: Claude Platform documentation on data residency, and AWS, 31 October 2025. As at 16 October 2026.
Not sure where your AI tools send your data?
We will go through your current setup with you and tell you what to ask each vendor.
Six questions to ask any AI vendor this week
These work for Anthropic, OpenAI, Microsoft, Google, HubSpot's AI features, or a smaller tool built on any of them. Ask for answers in writing, and ask for the document they come from.
| Question | Why it matters | A good answer sounds like |
|---|---|---|
| Where is our data stored at rest? | Stored copies are what a breach exposes and what a regulator asks about | A named country or region, written into the contract or a setting you control |
| Where is it processed? | Processing can happen somewhere other than storage, and "global" can mean anywhere | A named region, a setting to pin it, and a way to check where each request ran |
| Is our data used to train models? | Training copies can outlive your contract | "No, unless you opt in", in the commercial terms, covering every plan your staff use |
| How long is it kept? | Retention is how long the risk lasts | A number of days, the exceptions (abuse review, legal holds) listed, and a zero retention option if you need it |
| Who are your sub-processors? | Each one is another place your data can go | A published list with locations, and notice before it changes |
| Which region options exist today? | Roadmaps are not settings | Options you can switch on now, with their price, as against plans for next year |
Here is how Claude answers some of these today, from Anthropic's own pages. For API customers, Anthropic says retained data "is never used for model training without your express permission", and its commercial policy says API inputs and outputs are automatically deleted within 30 days, with exceptions such as Usage Policy enforcement and law. Zero data retention is available on request through Anthropic's sales team. Anthropic publishes its sub-processor list in its trust centre.
The bigger gap is often staff on personal accounts. On consumer Claude plans, Anthropic says chats may be kept for up to five years in de-identified form if the person has allowed their data to be used to improve the models. A business plan with its own settings beats a dozen personal logins. Our ChatGPT, Claude or Copilot comparison covers which business plan fits which team.
What does APP 8 mean when your AI runs overseas?
Australian Privacy Principle 8 covers sending personal information overseas. Before you disclose personal information to an overseas recipient, you have to take reasonable steps to make sure that recipient does not breach the Australian Privacy Principles. The OAIC says this is usually done with an enforceable contract.
The part most businesses miss is accountability. Under section 16C of the Privacy Act, if the overseas recipient mishandles the information, the OAIC treats it as your breach, even if you took reasonable steps. Customer details pasted into a US hosted AI tool fall squarely in that territory if the vendor counts as an overseas recipient.
The OAIC also says that in limited cases, giving data to an overseas cloud provider can be a "use" instead of a disclosure. That needs a binding contract limiting what the provider does with it, the same terms on its subcontractors, and you keeping effective control, including the right to access, change and delete it. Even then, you still hold the information and the other principles still apply.
In practice: read the vendor's data processing terms against those conditions, record which tools handle personal information, and update your privacy policy to say personal information may go overseas and to which countries. Our post on where your data goes when you use AI covers what the OAIC has said about AI tools specifically, and the Privacy Act change on 10 December 2026 covers the next deadline.
What to do this week, and what to watch for
- List your AI tools. Every one your staff use, including browser extensions and the AI features inside HubSpot, Microsoft 365 and Google Workspace. Mark which ones see customer or employee personal information.
- Send the six questions. To each vendor on that list. File the answers with your privacy records.
- Move people off personal accounts. If staff use consumer AI plans for work, move them to a business plan your admin controls.
- Decide if you need Australian processing now. If a contract, a regulator or a customer requires it, look at Bedrock's Australian profiles today. Waiting for 2027 is a bet on something Anthropic has not announced.
- Watch three things. An Australian option in Claude's inference geo or workspace geo, newer Claude models getting Australian profiles on Bedrock, and the FIRB decision on the Queensland lease.
We will update this post if Anthropic announces regional processing for Australia.
Frequently asked questions
Is Anthropic building a data centre in Queensland? Anthropic has signed a lease at the Western Downs Digital Park near Dalby, which Zerra DC is developing. The ABC reported Anthropic plans to start using Building One in 2027, subject to Foreign Investment Review Board approval.
Does Claude offer Australian data residency? Not through Anthropic's own API at the time of writing, 16 October 2026. Its documentation lists only "global" and "us" for where requests run, and "us" for where data is stored.
Can I run Claude in Australia today? For some models, yes, through Amazon Bedrock's Australian cross-Region inference profiles, which route between Sydney and Melbourne. Check that the model you want has one and confirm its destination regions with AWS first.
Does Anthropic train on business data? Anthropic says data retained through its API is never used for model training without your express permission. Consumer plans are different: whether chats are used depends on the user's model improvement setting.
What is APP 8? Australian Privacy Principle 8 requires reasonable steps before you disclose personal information overseas, and section 16C of the Privacy Act can make you accountable if the overseas recipient mishandles it.
Sources
- Anthropic is set to run Claude out of a Queensland data centre, here's what you need to know, ABC News, 18 September 2026, for the site, developer, 2027 start, FIRB condition and power arrangements.
- Anthropic signs first Australia data centre agreement, Reuters via The Star, 16 September 2026, for the inference use and Anthropic declining to comment.
- Anthropic busca oportunidades para construir centros de datos en Australia, Bloomberg Línea (Bloomberg syndicated, in Spanish), 13 October 2026, for Jeff Bleich's comments on further sites.
- Sydney office announcement, Anthropic, 10 March 2026, for the Sydney office and compute capacity plans.
- Data residency, Claude Platform documentation, for inference geo, workspace geo and US-only pricing.
- API and data retention, Claude Platform documentation, for training, zero data retention and who the processor is on Bedrock.
- How long do you store my organisation's data?, Anthropic Privacy Center, for the 30 day API retention period.
- How long do you store my data?, Anthropic Privacy Center, for consumer plan retention and training.
- Sub-processors, Anthropic Trust Center, for the published sub-processor list.
- Introducing Amazon Bedrock cross-Region inference for Claude Sonnet 4.5 and Haiku 4.5 in Japan and Australia, AWS, 31 October 2025, for the Sydney and Melbourne routing.
- Supported Regions and models for inference profiles, AWS documentation, for checking destination regions.
- Chapter 8: APP 8 Cross-border disclosure of personal information, OAIC, for reasonable steps, section 16C and the cloud "use" exception.
Want someone to read your AI vendors' data terms against APP 8 with you? See our AI consulting work, or talk to us.
Do you know where your AI tools process your customers' data?
In 15 minutes we will check your setup and tell you which questions to send.