Amazon Bedrock has added a new inference profile that keeps Kimi K3 model calls inside Indian AWS Regions. For teams that must keep health, financial or personal data within national borders, the change removes the need for custom cross‑region networking and opens a path to lower latency and cheaper multi‑turn interactions.
Choosing an Inference Profile
Kimi K3 can be invoked through two profiles:
global.moonshotai.kimi-k3– the default option that routes the request to the lowest‑cost destination region, regardless of geography.in.moonshotai.kimi-k3– the India geographic profile that forces processing to stay in theap-south-1(Mumbai) orap-south-2(Hyderabad) regions.
Both profiles use Bedrock’s cross‑Region inference mechanism, which encrypts data in transit and keeps traffic on the AWS backbone. No public‑internet exposure or multi‑Region VPC setup is required, simplifying compliance engineering.
Sample Architecture for Indic Document Processing
A practical use case is extracting structured data from scanned Hindi documents and answering follow‑up queries in mixed Hindi‑English (Hinglish). The workflow consists of:
- Upload the image to an S3 bucket in the same Indian region.
- Invoke Kimi K3 via the
in.moonshotai.kimi-k3profile, passing the image bytes as part of the prompt. - Apply prompt caching to reuse the model’s context for related follow‑up questions.
- Store the extracted JSON payload back to S3 or a DynamoDB table for downstream processing.
Prerequisites include an AWS account, Python 3.11+, the AWS CLI, and an IAM policy that grants Bedrock invoke permissions for the India profile. A minimal policy fragment looks like:
{
"Version": "2012-10-17",
"Statement": [
{
"Sid": "BedrockCallWithBearerToken",
"Effect": "Allow",
"Action": ["bedrock:CallWithBearerToken"],
"Resource": ["*"]
},
{
"Sid": "BedrockKimiK3India",
"Effect": "Allow",
"Action": ["bedrock:InvokeModel", "bedrock:InvokeModelWithResponseStream"],
"Resource": [
"arn:aws:bedrock:ap-south-1::foundation-model/moonshotai.kimi-k3",
"arn:aws:bedrock:ap-south-2::foundation-model/moonshotai.kimi-k3",
"arn:aws:bedrock:ap-south-1:{ACCOUNT_ID}:inference-profile/in.moonshotai.kimi-k3"
]
}
]
}
Replacing {ACCOUNT_ID} with the actual account identifier completes the setup.
Operational and Security Considerations
Because the India profile confines processing to regional data centers, any health or financial payload remains inside the AWS network and never traverses the public internet. Encryption is applied both at rest (standard S3/DynamoDB encryption) and in transit between the source and destination regions.
From an operations perspective, prompt caching reported by early adopters can cut model‑call spend by roughly 80 % and accelerate response times by 85 %. Teams should monitor Bedrock usage metrics to verify cost savings and to detect any unexpected cross‑region latency spikes that could indicate mis‑configuration of the profile ID.
IAM permissions are scoped to the specific inference profile ARN, limiting the blast radius if credentials are compromised. However, the policy still grants bedrock:InvokeModel on the foundation‑model ARN, so standard secret‑management practices for AWS credentials remain essential.
Related CloudNinjas coverage: AWS.
What This Means For Practitioners
- Adopt the
in.moonshotai.kimi-k3profile for any workload subject to Indian data‑residency rules. - Leverage prompt caching to reduce Bedrock spend and improve multi‑turn latency.
- Validate that IAM policies reference the correct inference‑profile ARN and that no broader
bedrock:*permissions are granted. - Instrument Bedrock request latency and cost metrics to confirm that the regional routing delivers the expected performance benefits.
- Watch for future announcements expanding in‑country inference to additional models or regions, which could affect architecture decisions.
