AWS Interconnect Multicloud Goes GA with Google Cloud as Launch Partner, Setting an Open Spec That Azure and OCI Have Agreed to Follow
AWS moved its cross-cloud private networking service to general availability on April 14, turning a preview into production and a bilateral deal into what analysts are calling a de facto standard.
Overview
Amazon Web Services declared its cross-cloud networking product generally available on April 14, graduating AWS Interconnect - multicloud from a five-month preview into a production service with Google Cloud as its first live partner and Microsoft Azure and Oracle Cloud Infrastructure committed to join later in 2026. The announcement on the AWS News Blog covers two launches in one: the multicloud GA, and a new AWS Interconnect - last mile option that simplifies on-premises connectivity through partners such as Lumen Technologies, with AT&T and Megaport in progress.
The move turns what the two hyperscalers had spent years telling regulators was a non-problem into a productised service, and it establishes an open API specification that Azure has publicly agreed to adopt, as previously reported in the context of the UK Competition and Markets Authority probe into Microsoft.
What We Know
AWS Interconnect - multicloud is a managed Layer 3 service that provisions dedicated private connections between Amazon VPCs and VPCs in a partner cloud. Traffic flows entirely over the two providers’ private backbones and never traverses the public internet, and every link between AWS routers and partner-cloud routers uses IEEE 802.1AE MACsec encryption at the physical layer, according to the AWS News Blog.
Five region pairs are live at GA, all connecting AWS Regions to nearby Google Cloud regions: US East (N. Virginia) to Google Cloud N. Virginia, US West (N. California) to Google Cloud Los Angeles, US West (Oregon) to Google Cloud Oregon, Europe (London) to Google Cloud London, and Europe (Frankfurt) to Google Cloud Frankfurt, per the AWS News Blog. Pricing is a flat hourly rate billed prorata on requested capacity, and that same post notes that Microsoft Azure and Oracle Cloud Infrastructure will be added later in 2026.
The architecture is built on pre-provisioned capacity pools at AWS Direct Connect and Google Cloud Interconnect points of presence, so customers no longer arrange physical cross-connects themselves. An AWS networking blog describes the control plane in detail: the Direct Connect Gateway acts as the AWS attachment point and a BGP route reflector, the Google Cloud Router handles the partner side, and BGP dynamically exchanges routes between the two providers. The same post says the infrastructure spans at least two physical facilities with independent power and networking, and that provisioning, modification, and deletion are all driven through APIs or console calls.
On the Google side, Google Cloud’s announcement confirms that preview connections start at 1 Gbps and scale to 100 Gbps at GA, with four redundant connections between each Google Cloud and AWS region pair, always-on MACsec at line rate, and setup times measured in minutes rather than days. Google published the joint design as an open specification on GitHub under AWS’s Interconnect standard and said more than half of the Fortune 500 already use its Cross-Cloud Network product.
The original preview landed at AWS re:Invent on December 1, 2025. The Register noted the awkward timing: both AWS and Google had told the UK Competition and Markets Authority the previous year that cross-cloud networking barriers were minimal, yet were now announcing a joint product to eliminate what AWS VP of network services Robert Kennedy called the “heavy lifting” of multicloud deployment. In that December announcement Kennedy said activation would take “minutes with a simple point and click.”
InfoQ’s preview coverage reported that the specification uses OpenAPI 3.0 for symmetric API coordination between providers, that connections are encrypted by default with MACsec between edge routers, and that both providers monitor links continuously. During preview, AWS offered one 1 Gbps connection per AWS account at no cost across the five region pairs. InfoQ also quoted Corey Quinn, chief cloud economist at The Duckbill Group, warning that “this is either transformative or a waste of everyone’s time, and it’s impossible to tell which because they aren’t disclosing [pricing] yet” - a gap AWS has now partially closed by disclosing the GA pricing model.
What We Don’t Know
AWS has not published per-region-pair GA pricing tables in the blog post, and Google Cloud’s own announcement declined to disclose pricing when it covered the preview in December. Without public rate cards, it is difficult to judge how the flat-fee model compares to the egress charges that have historically made multicloud designs expensive - the question Quinn flagged in InfoQ remains partly open.
The timeline for Azure and OCI integration is also vague. AWS’s GA post says both will arrive “later in 2026” without naming regions or a more specific date, and Microsoft has not published its own implementation plan for the open specification.
Analysis
For more than a decade AWS and Google positioned themselves as destinations, not waypoints, and multicloud was treated as a problem for customers to solve on their own with cross-connect brokers and VPN tunnels. The new service inverts that posture. By pre-building capacity at shared points of presence and exposing provisioning through a single API on either side, AWS and Google have turned a networking integration project into a menu selection in a console - the kind of change that, if it works at scale, reshapes where enterprises feel comfortable placing workloads.
The open specification matters more than the bilateral plumbing. Because Azure has already committed to adopt it and OCI is being pulled in as the third launch partner, the AWS-Google design is on track to become the default interface for private cross-cloud networking among the four biggest providers. That is the opposite of how cloud interoperability has usually played out, where each hyperscaler’s networking primitives were deliberately incompatible to discourage migration. Whether regulators treat the open spec as a genuine reduction in switching costs or as a convenient moat that still excludes smaller providers is likely to depend on how pricing and capacity pools evolve over the rest of 2026.