Anthropic has opened an office in Singapore and signed an agreement with the government on AI testing, security and adoption. The part that changes daily work arrived in September and gets less attention: organizations in Singapore can now run Claude so that each request is processed inside the country.
The question compliance teams ask first
When a bank or a hospital group wants to use an AI assistant, the capability is rarely the first question. The first question is where the document goes. A claim form, a loan file or a patient letter is sent to a model, and if that model runs in a data center in another country, a compliance officer has to explain that to a regulator.
In-region inference removes that question. The request is served by one region alone, and the prompt and the answer stay in it. It comes with limits. In Singapore the model available this way since September is Claude Sonnet 5, not the larger Opus 5 that Seoul also gets, and capacity is capped by what that one region holds.
Picture a claims team at a mid-sized insurer in Singapore. Each day it gets about 200 claims, each with scanned forms and a doctor's letter, and a clerk spends ten minutes per claim writing up a summary for the assessor. If Claude drafts those summaries and the clerk checks them, that is about 33 hours a day, the working days of four people, moved to the harder cases. Until the data stayed in the country, that workflow was a conversation the compliance officer ended early.
Where the government does the rest
Singapore did not rely on vendors alone. Its government has a programme to help 10,000 businesses adopt AI over three years, with grant support for approved AI tools, and since the end of September a new grant covers up to 70% of the cost of automation and digitalisation projects for small and medium firms. Anthropic's agreement with the government also plans help for small businesses and AI training for public officers.
So the high usage has two causes that work together. Large regulated firms got a lawful way to send real documents to the model, and small firms got part of the bill paid. Grab's savings show how big the result is once both are in place. Anthropic had already let enterprises store their Claude data themselves, and the Singapore step moves the same concern, control over where data lives, from storage to processing.
What other countries will copy
Singapore is small, rich and full of office workers, which helps. But the two causes do not depend on size. Any country with a bank regulator and a finance ministry can offer local processing and share the cost with small firms. I expect the countries that do it to see usage climb toward Singapore's, and the ones that wait to see their banks and hospitals use the same assistants more slowly, on the same models.
For a manager in a regulated company, this changes what to ask for. Do not start with whether the AI is good enough. Ask your vendor whether it can process your requests in your own country, and whether your government pays for any of it. The answers decide how soon your team can use AI on real documents.