AWS Connected Its Network to Google Cloud With No Egress Fees. Azure and Oracle Are Next.

AWS Interconnect multicloud reached general availability on April 14, 2026, giving enterprises direct private Layer 3 connections between Amazon VPCs and Google Cloud VPCs over the AWS global backbone. Traffic never touches the public internet. The pricing removes per-gigabyte data transfer charges: you pay for the bandwidth tier you reserve and transfer as much data as the pipe supports. At 10 Gbps Tier 1, that is $12.33 per hour. Azure and Oracle Cloud are the announced next partners. Oracle confirmed on April 16, two days after the AWS-Google GA.

How It Works

AWS provisions multiple redundant connections distributed across at least two physically separate interconnection facilities. IEEE 802.1AE MACsec encryption runs on the physical links between AWS and Google’s routers. No BGP configuration or peer IP management is required from the customer. Provisioning takes minutes through the AWS console, CLI, or API, compared to weeks or months for the colocation-based Direct Connect arrangements it replaces in this use case. Bandwidth is adjustable without recreating the connection. AWS has published an open Connection Coordinator API Specification on GitHub that allows other cloud providers, including regional hosters and neoclouds, to integrate on equal terms.

Five region pairs are available at GA across North America and Europe: US East (N. Virginia), US West (N. California), US West (Oregon), Europe (London), and Europe (Frankfurt), each connected to the corresponding Google Cloud region. Google charges separately for its side via its Cross-Cloud Interconnect product. From May 2026, AWS provides one free 500 Mbps local interconnect per region for testing.

The Pricing Model Is the Story

The previous model for moving data between clouds charged per gigabyte leaving the source provider, a mechanism that made multicloud architectures expensive to operate and kept workloads anchored to a single provider. AWS Interconnect multicloud bills by bandwidth tier with no per-GB charge. The five tiers reflect geographic distance between the source VPC region and the local interconnect: 10 Gbps runs $12.33 per hour at Tier 1(same region) and $51.78 per hour at Tier 4 (long-haul). The math for high-volume transfers shifts substantially: at current egress rates, moving a petabyte of data per month between US-based AWS and Google Cloud regions would cost tens of thousands of dollars in transfer fees. On Interconnect multicloud at 10 Gbps Tier 1, the connection costs roughly $9,000 per month with no transfer ceiling.

AWS has publicly framed the flat bandwidth model as a deliberate rethink of cloud pricing. Whether this creates pressure on AWS’s own egress fees for traffic leaving to the public internet is a separate question the company has not addressed.

What Oracle Joining Means

Oracle already runs private interconnects with Google Cloud and Azure. Oracle’s announcement on April 16 means OCI now has direct private connections to all three other major clouds. AWS+Azure remains on the roadmap for later in 2026. For enterprises with OCI in their stack, the friction cost of moving data to any other major cloud is now substantially lower than it was six months ago.

AI Workloads as the Demand Driver

The use case AWS and analysts cite most often is AI architectures that split training and inference across clouds, or that store data in one cloud while running models in another. The cost and latency of moving data between providers has been a constraint on these architectures. theCUBE Research principal analyst Rob Strechay framed the appeal directly: “The future of AI is where your data lives in one place, your models run in another, and the network doesn’t get in the way.” Whether that translates to a meaningful workload migration from single-cloud to multicloud configurations will depend on how quickly enterprises are willing to redesign architectures they spent years building around a single provider.

The Door Just Opened for Regional Hosters in Cloud Networking

Megaport and Equinix Fabric built significant growth narratives on being the neutral broker between cloud providers. That role is now served natively by AWS, and both companies are repositioning toward last-mile services. For most shared hosting and managed VPS providers, the direct relevance is limited: the price point is enterprise and the use case assumes a customer already operating at scale across two major clouds.

The more immediate question is whether enterprise customers who previously chose a single-cloud managed hosting arrangement will now expect their hosting provider to support multi-cloud connectivity as a standard option. A managed hosting provider running customer workloads on AWS that wants to offer Google Cloud as a failover or complementary environment now has a native, cheaper mechanism to do that than Megaport or a VPN.

The longer-term opening is the open API specification. AWS has published the Connection Coordinator API Specification on GitHub and explicitly invited neoclouds and regional providers to integrate. A hosting provider with its own network infrastructure in one of the supported interconnect locations could qualify as a partner and offer private AWS connectivity as a service, effectively entering the cloud networking market that Megaport currently occupies. The requirements around API functionality, security, and physical redundancy are real barriers, but the door is open in a way it was not six months ago.