Google Cloud Took the First Seat on AWS Interconnect

Google Cloud is the only multi-region production partner on AWS Interconnect, the private pipe AWS and Google Cloud unveiled on November 30, 2025. Oracle Cloud Infrastructure later reached general availability in a single AWS Region. Microsoft Azure is still in public preview.

The customer pitch was minutes, not weeks. The industry result is a new map of who collects the toll, because AWS wrote the spec and Google sat down first.

Google Cloud Sat Down First

On November 30, 2025, Rob Enns, VP and GM of Cloud Networking at Google Cloud, and Robert Kennedy, VP of Network Services at AWS, published a jointly engineered multicloud networking solution that pairs AWS Interconnect-multicloud with Google Cloud’s Cross-Cloud Interconnect. They also pointed other providers at a public API spec.

That post is the chapter when AWS and Google first described the joint pipe as a managed alternative to colo cages, LOA paperwork, and carrier tickets. Jim Ostrognai, SVP of Software Engineering at Salesforce, said AWS Interconnect-multicloud “allows us to establish these critical bridges to Google Cloud with the same ease as deploying internal AWS resources, utilizing pre-built capacity pools and the tools our teams already know and love.”

Google Cloud already sold a one-sided version of this idea. Cross-Cloud Interconnect, launched in 2023, let a customer order a Google-managed private path toward AWS, Azure, or Oracle. The 2025 joint product is different because AWS now treats the far end as a first-class attach in its own console, and because both companies agreed to a shared provisioning API rather than a Google-only workflow.

AWS moved the Google pair to general availability on April 14, 2026, and named Google Cloud the first launch partner. The launch footprint was five AWS Regions: US East (N. Virginia) to Google Cloud in N. Virginia, US West (N. California) to Los Angeles, US West (Oregon) to Oregon, Europe (London) to London, and Europe (Frankfurt) to Frankfurt.

THE RACE TO A PRODUCTION PIPE

1. November 30, 2025: AWS and Google Cloud announce the joint interconnect and publish an open API spec.
2. April 14, 2026: AWS Interconnect-multicloud reaches general availability with Google Cloud in five AWS Regions.
3. May 29, 2026: AWS adds one free local 500 Mbps interconnect per Region for each generally available provider, on the AWS side only.
4. July 29, 2026: Oracle Cloud Infrastructure reaches general availability in us-east-1 (N. Virginia).
5. August 31, 2026: Microsoft Azure joins in public preview in four AWS Regions.
6. September 2, 2026: Equinix says Fabric One will use the same open spec, with general availability planned for 2027.

From April 14 until July 29, a span of 106 days, the only generally available far end was Google Cloud. That window is the seat Google actually won.

An Apache Spec Lives in AWS’s GitHub

AWS posted the OpenAPI 3.0 Interconnect specification on GitHub under Apache 2.0. The repo describes a Connection Coordinator API, a symmetric REST contract two providers use to request, accept, and monitor a managed Layer 3 path. Google Cloud’s joint post links the same package and invites other providers to adopt it.

Open in the license is not open in the club. AWS says a cloud provider that wants to become an Interconnect partner must implement the spec and also meet AWS operational requirements, including resiliency standards, support commitments, and service level agreements. The document lives in a public AWS repository. The production on-ramp still runs through AWS’s bar.

When they need multicloud connectivity, it’s ready to activate in minutes with a simple point and click.

Robert Kennedy, VP of Network Services, AWS, in the joint Google Cloud post

Kennedy’s line is the product claim. The spec is the power claim. Once Azure, Oracle, and Equinix implement the same coordinator API, AWS does not have to design a new attach for each brand. Other clouds have to look like AWS Interconnect to reach the largest installed base of virtual private clouds.

Google Cloud is not a passenger on that point. Enns has kept repeating that Cross-Cloud Network was already Google’s answer to DIY circuits, and the 2025 deal folds AWS into that story instead of forcing Google to keep provisioning AWS as a remote afterthought. Both companies get what they wanted: AWS sets the interface, Google gets the first production seat on it.

Azure’s Preview Still Caps the Triangle

Most large shops that split clouds split them across some mix of AWS, Azure, and Google Cloud. Until Azure is generally available on the same attach, the triangle is unfinished, and the unfinished side is the one that still sends architects back to ExpressRoute plus a broker.

AWS opened a public preview with Microsoft Azure on August 31, 2026, in four AWS Regions: US East (N. Virginia), US West (N. California), Asia Pacific (Sydney), and Europe (Frankfurt). Preview interconnects are capped at 1 Gbps, limited to one interconnect per customer per supported Region, and billed at no AWS charge for the preview period. AWS also warns that those preview 1 Gbps connections are removed from accounts as that provider approaches general availability.

WHO IS ACTUALLY GENERALLY AVAILABLE

Production teams already sort the grid that way. Google Cloud is the path you can put in a design review. Oracle is the path you can put in a design review if the workload lives in N. Virginia. Azure is the path you can lab. That gap, not the original press language about openness, is how the four major clouds now sit on the same product.

Equinix Is Writing Itself Onto the Spec

The quiet casualty was supposed to be the interconnection brokers. For years the way you joined an Amazon VPC to a Google VPC without the public internet was a cage, a cross-connect, or a software fabric from Equinix, Megaport, PacketFabric, or Console Connect. Google’s 2023 Cross-Cloud Interconnect already tried to pull that job in-house. The 2025 joint spec goes further by making the cloud consoles the place you click.

Equinix did not file a protest. On September 2, 2026, it unveiled Equinix Fabric One on the open spec at its Horizon event, with AWS and Google Cloud named as lead integration partners. Fabric One is an intent-driven any-to-any service: the customer names where, what, and how much, and Equinix composes routing, encryption, and failover. Equinix said the product is expected to enter beta later in 2026, with general availability planned for 2027, initially in North America.

For decades, enterprise networks have been built one connection at a time for each partner and provider they depend on. That approach doesn’t scale in a world of distributed AI that demands dynamic, flexible and real-time connectivity.

Chris Audie, Chief Product Officer, Equinix, September 2, 2026 press release

Kennedy and Enns both showed up in that release too. Kennedy said customers “should just define what they need and have it work.” Enns called cross-cloud interoperability “the backbone of the AI era.” Equinix, which lists more than 10,500 interconnected businesses and about 3,000 cloud and IT providers on its exchange, is betting it can remain the place you stitch clouds the spec does not yet pair, plus AI trainers and neoclouds the hyperscalers will not onboard as Interconnect partners.

Megaport is taking a different door. When AWS launched Interconnect last-mile on April 14, 2026, Lumen was the first partner in US East (N. Virginia), with AT&T and Megaport listed as in progress. The old circuit landlord becomes a last-mile agent of AWS in that model, not the person who owns the AWS-to-Google hop.

What the Hourly Bill Looks Like

AWS replaced port-plus-transfer folklore with a single-fee hourly pricing model based on bandwidth and geographic scope. There is no separate AWS charge for data moved across the interconnect itself. The other cloud still sets its own price for its half of the infrastructure, and AWS tells customers to read that page before they click create.

Tiers run from local (Tier 1) to long-haul (Tier 4 and Tier 5). Attach the interconnect to AWS Cloud WAN and a far-away core network edge, and AWS automatically bills the highest tier that covers every path on that Direct Connect gateway. A cheap local pipe can become a continental one without anyone ordering a new circuit, which is the fine print that will surprise finance teams.

HOW THE METER ACTUALLY RUNS

The free 500 Mbps path, which arrived on May 29, 2026, is the on-ramp AWS wants teams to try. AWS says a 500 Mbps interconnect can move about 160 TB a month, and each free path includes Amazon CloudWatch Network Synthetic Monitor at no extra cost. It cannot be converted into a paid interconnect. Google Cloud, Oracle, or Azure can still invoice their side.

That split is why Oracle’s later seat is not a copy of Google’s. Oracle said its Interconnect for AWS carries no egress fees, with general availability starting in OCI Ashburn and AWS US East (N. Virginia). Google Cloud has not matched that waiver on the AWS pair. The pipe can look identical in the AWS console and still cost a different amount once traffic leaves.

Console Clicks Meet an MTU Trap

The April 2026 general availability walkthrough is the honest picture of what “minutes” means. On AWS, you pick Google Cloud, choose a Region pair, set bandwidth, attach a Direct Connect gateway, and paste a Google Cloud project ID. AWS hands back an activation key. On the Google side, the same walkthrough still used the gcloud CLI to create a transport and a VPC peering, because no Google Cloud web console for that step was available in the AWS write-up.

Traffic stays on the AWS backbone and the partner’s private network and does not ride the public internet. Physical links between the providers’ edge routers use IEEE 802.1AE MACsec by default. Each connection spreads logical links across at least two physical facilities. Google Cloud’s joint post describes that design as quad-redundancy across redundant interconnect facilities and routers, with both providers watching the path.

The gotchas are ordinary networking, which is why they will still page people.

WHAT STILL BREAKS A FIVE-MINUTE PIPE

• Overlapping CIDR: The AWS VPC and the Google Cloud VPC cannot share address space, and the walkthrough only worked because the defaults (172.31.0.0/16 on AWS, 10.156.0.0/20 on Google Cloud) did not collide.
• Mismatched MTU: AWS and Google Cloud VPCs do not share a default maximum transmission unit, and a mismatch drops or fragments packets across the interconnect.
• IP version: IPv4, IPv6, or both must be selected the same way on both ends.
• Last AWS route: After the attach is up, the AWS VPC route table still needs a route for the Google Cloud range through the virtual private gateway.
• Tier creep: Adding a remote Cloud WAN edge can silently lift a local interconnect into a higher geographic tier.

For a single Region pair, a virtual private gateway is enough. Several VPCs in one Region go through AWS Transit Gateway. Global estates go through AWS Cloud WAN, which can reach any interconnect attached to the same Direct Connect gateway. Those extra hops are how the product scales, and how the bill leaves Tier 1.

Oracle Took a One-Region Seat

Oracle had already rehearsed this move with Google Cloud. In 2024 the two companies opened Oracle Interconnect for Google Cloud in 11 commercial regions and waived cross-cloud data-transfer charges, then layered Oracle Database@Google Cloud on top. The AWS Interconnect seat is the same strategy aimed at a larger hypervisor estate: make it cheap and private to keep an Oracle database next to an AWS application.

The July 29, 2026 GA is narrower than Google’s April launch. AWS limited the OCI pair to us-east-1. Oracle’s own note starts the same way, Ashburn to N. Virginia, with more pairs promised later. A shop that needs Oracle in London or Tokyo is not on this product yet. A shop that needs Google Cloud in those metros has been on it since April 14, 2026.

Azure’s delay is the leftover political fact. Microsoft is the other hyperscaler with an enterprise directory, an on-prem estate, and a habit of charging for egress. It adopted the spec last among the three, and only in preview. Until that preview becomes general availability, the AWS-Google path is the one with a real SLA story, a free 500 Mbps trial, and five live Region pairs.

Equinix’s 2027 date is the other leftover. The brokers are not gone. They are rewriting themselves as implementers of a coordinator API that now lives in an AWS GitHub repo, while Google Cloud keeps the only multi-region production chair at the table those brokers used to own.