AWS PrivateLink Tunnel Endpoints Let Vendors Reach an Entire Shared CIDR Range Instead of Listing Resources One by One
AWS detailed tunnel endpoints, a PrivateLink VPC endpoint type launched in September that lets a vendor tunnel into a customer-defined CIDR range rather than one resource at a time.
Overview
AWS has published a detailed technical walkthrough of tunnel endpoints, an AWS PrivateLink capability that launched on Sep 18, 2026 and lets an application in one VPC reach every address inside a shared network segment rather than one resource at a time. The walkthrough, published on Nov 5, 2026 by AWS solutions architect Bruce Pinsky and principal product manager Ashish Kumar, lays out the mechanics, setup steps and constraints behind a feature AWS’s own launch announcement summarized more briefly in September. InfoQ covered the walkthrough on Nov 18, 2026, describing it as a feature that “lets an application in one VPC reach every address in a CIDR range shared from another account, instead of one resource at a time.”
What changed
AWS PrivateLink already let one AWS account privately reach load-balanced services and resources in another account’s VPC. Before this launch, according to AWS’s blog post, “to share resources with a vendor, the customer had to enumerate them individually by creating a resource configuration for each using IP address or DNS name.” AWS says that approach “was limiting for use cases where resources are ephemeral or aren’t knowable up front.”
Tunnel endpoints change the unit of sharing from a single resource to a network segment. A customer — the “provider” in AWS’s terminology — can now create what AWS calls a CIDR-type resource configuration, which “represents a network segment as one or more IPv4 or IPv6 CIDR ranges,” per the blog post. The customer shares that configuration with a vendor account through AWS Resource Access Manager. The vendor, as “consumer,” then creates a tunnel endpoint — a new VPC endpoint type — and uses it to reach anything inside the shared range, including, the post says, resources that did not exist when the share was first created.
AWS frames the use case around vendors and service providers: a security scanner that needs to reach every host in a customer’s subnets, an analytics service querying databases wherever a customer keeps them, or an integration platform that must connect to whatever the customer deploys next. The What’s New announcement says the capability launched across more than two dozen AWS Regions spanning the Americas, Europe, Africa and Asia Pacific, with an hourly charge for the tunnel endpoint plus a per-GB charge for data processed through it.
How the tunnel works
The mechanism relies on GENEVE encapsulation, an IETF tunneling protocol that, according to AWS’s blog post, is defined in RFC 8926. Because a consumer’s VPC cannot natively route to an address inside the provider’s shared range, the consumer’s client wraps its traffic in a GENEVE header and sends it to the tunnel endpoint over UDP port 6081. The endpoint strips the header, checks the inner destination against the shared CIDR ranges, rewrites the source address to a resource gateway address, and forwards a plain packet into the provider’s VPC. AWS’s post says Linux supports GENEVE natively, and its walkthrough shows a consumer creating a GENEVE network device inside its own network namespace so the tunnel’s routing does not interfere with the instance’s other traffic.
Because addressing on the provider side exists only inside the tunnel, AWS says “the consumer and provider VPCs can use overlapping CIDR blocks.” The feature also supports resolving DNS names through the tunnel as if the consumer’s compute were inside the provider’s VPC, by directing queries to a link-local resolver address inside the tunnel.
The security model AWS describes
AWS’s post is explicit that the shared CIDR range functions as an access perimeter: “Resources located outside the perimeter can’t be reached by the vendor.” Connections are consumer-initiated only — AWS writes that “Connections (TCP SYN) can start only from the consumer VPC, not from the provider network” — so a vendor’s tunnel endpoint cannot be used by the provider to reach into the vendor’s network. If a provider later removes a consumer account from the Resource Access Manager share, AWS says that account can no longer create new tunnel endpoints, though connections already established keep working.
The post also flags a risk customers control themselves: sharing the broadest possible range, 0.0.0.0/0 or its IPv6 equivalent ::/0, “grants access to everything the provider VPC can reach,” including networks reachable through VPC peering, AWS Direct Connect or AWS Site-to-Site VPN, and internet destinations through the VPC’s own egress path. AWS’s guidance is to “share the narrowest ranges that cover the workload(s).” On logging, the post says access records capture the outer, encapsulated addresses and the resource gateway’s translated address, but carry no field for the consumer’s inner source address — meaning a provider can attribute tunnel traffic to a particular vendor’s endpoint, but not to a specific workload behind that endpoint.
Other constraints detailed in the walkthrough: only TCP application traffic is supported inside the tunnel, with UDP permitted solely for DNS queries, so protocols like ICMP (and therefore ping) do not work; IP fragments are dropped rather than forwarded; a resource gateway’s DNS resolution setting cannot be changed after the gateway is created; and a tunnel endpoint’s subnets must sit in Availability Zones that overlap with the resource gateway’s, or the association is reported as inaccessible.
What we don’t know
Neither AWS’s walkthrough nor its What’s New announcement states the numeric per-VPC quota for tunnel endpoints, pointing instead to AWS’s general PrivateLink quotas documentation. AWS’s post also does not disclose how many customers have adopted the feature since its September launch, and the cited sources do not describe how tunnel endpoints are priced relative to other PrivateLink endpoint types beyond the stated hourly and per-GB charges.