OpenAI models land on Bedrock in India, and the data never leaves

Amazon Bedrock has started serving OpenAI's GPT-5.6 Terra and Luna models in India with an in-country inferencing option. According to the AWS announcement, requests route exclusively between the Asia Pacific (Mumbai) and Asia Pacific (Hyderabad) regions; data does not leave India's borders during inference.

Both models offer a 1-million-token context window and accept image input alongside text. Routing is handled through the in.openai.gpt-5.6-terra and in.openai.gpt-5.6-luna inference profiles. Bedrock's default is zero data retention (ZDR): inputs and outputs are not stored, with the sole exception of content flagged by abuse-detection classifiers.

AWS names its audience plainly: customers with local data processing requirements, including financial services, healthcare and the public sector. In this phase of the AI race, where data is processed has become as much of a purchasing decision as how strong the model is, and India is the first big stage for this pattern.

There is a competitive layer too: OpenAI's models being sold on Amazon's cloud, on Amazon's commercial terms, shows the "every model, everywhere" era of distribution is now the norm. For other countries with similar data-localization expectations, the template is set: twin regional hubs, in-country routing, ZDR by default.