OpenAI brings API inference residency to India for banks and enterprises

OpenAI will now run AI requests from Indian business customers on servers inside India. The company calls the arrangement API inference residency: both the data a company submits and the computation that produces the answer stay within the country's borders instead of being routed to data centres overseas.

That detail matters most to banks and other financial firms, which have been cautious about putting AI on serious work. Rules governing customer information often require it to be stored and handled domestically, and that ruled out tools that processed requests abroad.

Nitin Bawankule, who heads OpenAI's enterprise business in India, said some companies held back from using AI for important tasks precisely because the information had to be processed in-country. In his view, local inference gives firms more control over where supported workloads run, and should give Indian developers and startups more room to build products for domestic and enterprise needs.

For now the feature covers a single model, GPT-6 Luna, and focuses on text work: condensing long documents, drafting replies and analysing internal material. The scope is a defined set of text scenarios rather than localisation of every model and modality.

Using it is a deliberate setup rather than a toggle. A company creates a new API project on OpenAI's platform, picks India as the region for both storage and processing, and connects through a dedicated India endpoint; text moving in and out of GPT-6 Luna is supported through the Responses and Chat Completions APIs.

The launch sits inside OpenAI's OpenAI for India programme, announced earlier this year to strengthen AI infrastructure and business adoption in the country. To build local capacity, the company earlier teamed up with the Tata Group on domestic AI data centres.

India now joins Australia and Japan among the markets where OpenAI has widened in-country processing tools, and the company recently offered a comparable residency option through AWS Bedrock. Read together, those moves turn domestic-only processing from a one-off concession into a repeatable regional deployment pattern.

The open question is coverage. If residency extends from one model to more models and modalities, adoption across Indian finance, healthcare and government-adjacent sectors could accelerate quickly; a permanently text-only window would cap which workloads it can unlock.