If you ever connected AWS privately to another cloud, you know the drill. Each provider sells you only its half: a Direct Connect virtual interface on one side, an ExpressRoute circuit, Cloud Interconnect or FastConnect on the other. The middle is yours. And yes, it does feel like the middle finger sometimes. Colo space or a virtual router in a carrier-neutral facility, cross-connects, letters of authorisation, VLANs, point-to-point subnets, BGP with both clouds, and the question if MACsec is worth the extra work. And then, at the worst possible moment, you find out that two teams picked the same 10.0.0.0/16. To be fair, many large enterprises already run their own data centres or colo cages with Direct Connect and ExpressRoute landing right there, so for them the middle is already built (and paid for). For everyone else, it is a project.

Many teams skipped all of this and just ran IPsec over the internet. That works until the throughput ceiling or the egress line on the bill catches up with you. Megaport and Equinix Fabric took away the physical work, but the routing, the monitoring and the blame game stayed with you (Simon Painter).

On 17 November 2026 AWS made AWS Interconnect – multicloud generally available with Microsoft Azure. Google Cloud (GA April 2026) and Oracle Cloud Infrastructure (GA July 2026) were already there. What caught my eye: all three use one control-plane contract, the Connection Coordinator API, an OpenAPI 3.0 spec which AWS and Google published under Apache 2.0 on GitHub. Microsoft also put its own Google Cloud connection into preview on the same spec. So the three biggest clouds can now set up private links to each other, with no colo cage in the middle.

In this post I'll walk you through how it actually works under the hood, why it happened now (spoiler: customers, competition and regulators), what it costs and where the catch is, and a quick checklist so you can decide whether it's worth it for you or whether your existing colo setup already does the job.

The problem it solves

Before vs after
Before: you build and run the middle. After: one managed Interconnect object between the Direct Connect gateway and the partner's attach point.

AWS and each partner pre-build capacity between their edge routers, in the same facilities Direct Connect already uses. You pick a region pair and a bandwidth, AWS gives you an activation key, and you redeem it on the other cloud. VLANs, addressing, BGP and MACsec are then negotiated between the providers, not by you (AWS docs). You still own your VPC/VNet design, routing intent, security policy and recovery plan.

How does it work?

How it works
One Interconnect object, four MACsec-encrypted links across two facilities, BGP and ECMP on top. Example: AWS and Azure.

Under the hood every interconnect is four links between AWS routers and the partner's routers, spread across at least two physical facilities. Traffic is balanced with ECMP, and the routers forward customer traffic only when the MACsec session is up (AWS docs). On the AWS side the interconnect always lands on a Direct Connect gateway, and from there you attach a VGW, Transit Gateway or Cloud WAN. On the Azure side it is an ExpressRoute circuit with the new Multicloud Interconnect port type, and your VNets connect through an ExpressRoute virtual network gateway. The two halves are paired with the activation key, and behind the scenes the providers talk to each other over the Connection Coordinator API (GitHub). Nice detail: either side can start the request.

Why now?

Was it customers, competition, or regulators? I think the honest answer is: all three.

Customers. AWS itself writes that multicloud is often "less about choosing the ideal tool for each job and more about how businesses actually evolve": acquisitions, ISVs deploying where their customers are, regulations requiring more than one cloud (AWS). Salesforce is the only named early adopter I found; it uses the service to reach Google Cloud for Data 360 and to ground "AI and analytics in trusted data, regardless of where it resides". Resilience plays a role too, the launch came not long after the October 2025 us-east-1 outage.

Competition. AWS did not build a proprietary product, it co-authored an open spec with Google. Only AWS and Google are Maintaining Organisations and sit in the steering committee, and the implementers list is still empty. So Microsoft and Oracle use the spec in production, but formally they are not part of its governance (yet). Forrester's Lee Sustar called it AWS "leveraging its market share to muscle the industry into making its approach a de facto standard" (InfoQ). And the order fits: Google first, then OCI, and Microsoft, the closest rival, last. I think the real prize is the attach point: every partner link lands on an AWS Direct Connect gateway. That is my reading, AWS did not say it.

Regulators. In July 2025 the UK CMA found that egress fees and interoperability barriers limit switching and multicloud. The Register pointed out the irony: in 2024 AWS told the CMA "there is no interoperability problem in cloud services", and a year later it shipped a fix with Google. In March 2026 the CMA did not impose binding remedies on cloud. Instead AWS and Microsoft published voluntary, non-binding actions: free egress when switching (at least 180 days), at-cost multicloud egress in UK contracts, and direct interconnects. The CMA kept the option to use its formal powers later (CMA).

In the EU, the Data Act bans switching charges, egress included, from 12 January 2027. But for parallel multicloud use, providers can still charge egress at cost (Article 34). So "free egress" is not the full story. The Commission also reached a preliminary view in June 2026 that AWS and Azure should be DMA gatekeepers, but a final decision has not been published yet.

A flat-rate, no-per-GB interconnect is basically the commercial form of "at-cost multicloud egress". Regulators can point to it, providers can present it as their own idea. The timeline does not prove causation either way.

How the providers stood up

Timeline
Product milestones (blue) and regulatory milestones (amber), May 2023 to January 2027.

  • 30 Nov 2025: preview at re:Invent with Google Cloud, open spec published
  • 14 Apr 2026: GA with Google Cloud, five region pairs
  • 29 May 2026: free 500 Mbps tier on the AWS side
  • 29 Jul 2026: GA with OCI, us-east-1 only
  • 31 Aug 2026: Azure public preview
  • 17 Nov 2026: Azure GA in us-east-1, eu-central-1 and ap-southeast-2

What you get, and what it costs

  • Minutes instead of weeks to provision, from the console, CLI or API.
  • Private path, never the public internet, with MACsec between the provider edge routers by default.
  • Four links across at least two facilities, and a 99.99% SLA for the AWS side (AWS SLA).
  • Flat hourly pricing: bandwidth plus an automatically assigned geographic tier, no per-GB charge on the AWS side.
  • Free tier: one local 500 Mbps interconnect per Region per partner, about 160 TB a month at full use.

Now the reality check. The tier is decided by the farthest place your Direct Connect gateway can reach. The same 10 Gbps link is $12.33/h at Tier 1 and $51.78/h at Tier 4, if your Cloud WAN has Core Network Edges in both N. Virginia and Singapore (AWS pricing). The other cloud bills its own side separately, and Azure's paid rates are still unconfirmed. Azure's own free tier is 90 days only, without SLA. IP ranges must not overlap across the clouds. And the UK is not available yet: no AWS–Azure UK region at GA, and AWS–OCI is US-only, with a UK rollout pending demand (CMA update).

So, should you use it?

Good fit Not so good fit
Steady cloud-to-cloud flows (replication, pipelines, split-stack apps) in a supported region pair Regions not paired yet, like the UK for AWS–Azure
SaaS/ISV vendors who need private reach into customer clouds Your own global backbone or NaaS fabric already reaches every cloud
Post-M&A estates that need a quick private bridge Overlapping CIDRs that would need a NAT redesign
FinOps teams replacing per-GB egress with a fixed line Global Cloud WAN designs that land in Tier 4–5
Small or test links under 500 Mbps in one Region Spiky, low-volume traffic where VPN is still cheaper

AWS Interconnect – multicloud does not make multicloud easy. It makes the network the easiest part of it. For architects the decision is now about region pairs, IP plans and tier maths, not colo contracts. For FinOps, cloud-to-cloud traffic becomes a capacity line instead of a per-GB guess. Customer demand or regulatory pressure? I think it matters less now than whether the coverage keeps growing.

If you already tried it, or you are still stitching clouds together the old way, please leave a comment. I am curious how your bill looks before and after.