A company runs business applications on AWS.

The design already uses one transit gateway (TGW) per Region and inter-Region TGW peering, which addresses the multi-Region AWS-side routing. The remaining requirement is to extend on-premises connectivity over Direct Connect so that on-premises networks can reach VPCs attached to the TGWs across all three Regions. The most operationally efficient and scalable approach is to use AWS Direct Connect Gateway (DXGW) with a transit virtual interface (transit VIF) and then associate the Regional transit gateways to that DXGW.

Option C is purpose-built for this: a transit VIF is specifically used to connect a Direct Connect connection to a Direct Connect gateway, and a DXGW can then be associated to multiple transit gateways (and can be used across Regions), enabling centralized connectivity from on-premises to TGW-connected VPCs. With TGW attachments and TGW route tables, you can propagate and control routes across many VPCs and accounts, which fits the “50 accounts, thousands of VPCs” scale. Inter-Region TGW peering then allows on-premises routes learned via the DXGW/TGW in one Region to reach workloads in the other Regions through the TGW peering relationships, subject to routing configuration.

Option A is too limited and not scalable because it ties Direct Connect to a virtual private gateway (VGW) associated with a single VPC, which does not meet the multi-VPC, multi-account hub-and-spoke requirement. Option B incorrectly suggests associating a DXGW with a VGW “in each VPC” (VGW is per VPC and would not scale well here, and it doesn’t integrate with the TGW hub design you’ve already built). Option D is not the intended pattern: Site-to-Site VPN and public VIF do not replace the DXGW + transit VIF architecture for large-scale TGW-based private routing.