MotherDuck’s BYOC keeps queries in your AWS account—but not every byte of operational context
The private-region launch moves compute, storage and encryption keys into a customer account while leaving identity, billing, telemetry and optional AI processing across the control-plane boundary.
MotherDuck’s Bring Your Own Compute option addresses a narrow but important enterprise requirement: keeping database compute and stored data inside a customer-controlled AWS account. It does not make the service fully self-contained, and MotherDuck’s own documentation gives architects a useful checklist for the remaining boundary.
What moves into the customer account
MotherDuck says BYOC deploys a dedicated private region into a member account in the customer’s AWS Organization. Compute, storage, caching and encryption keys run there; native MotherDuck storage and DuckLake data live in customer S3 buckets. Queries execute on the customer’s compute, and AWS administrators can inspect resources and audit MotherDuck access with CloudTrail.
MotherDuck still operates and upgrades the private region. BYOC is an Enterprise-plan feature, with AWS infrastructure charges billed to the customer account and MotherDuck’s platform billed separately under the enterprise agreement.
What still crosses the boundary
Identity and operations remain split. Sign-in, authentication and MotherDuck billing stay in MotherDuck’s control plane through app.motherduck.com. The control plane also receives organization, user and database metadata for access control and billing, plus logs, metrics, query history and usage records.
There is a second boundary for AI. MotherDuck’s docs say that when AI functions are enabled, inputs passed to those functions are processed outside the customer AWS account. Administrators can change organization settings for AI features, but teams evaluating BYOC for residency or regulated workloads should treat that setting as a separate data-flow decision—not assume the private region covers it automatically.
The deployment review to run
Before approval, security teams should classify four flows independently: stored data, query-time processing, operational telemetry and AI-function inputs. They should also review the dedicated-account access granted to MotherDuck, CloudTrail coverage, region immobility and the limitation that shares remain inside the private region rather than crossing to organizations in other MotherDuck regions.
The practical conclusion is not that BYOC fails at residency. It is that “data stays in your account” describes stored data and query execution, while control-plane metadata, operational records and optional AI inputs retain separate paths that must appear in the architecture review.
sources
- MotherDuck Docs — Bring Your Own Computemotherduck.com
comments · 0