Multicloud interconnect is free in preview, built for distributed AI workloads
In sum – what we know:
- Joint managed interconnect – Microsoft and AWS now offer a provider-managed Azure–AWS link, provisioned via API or IaC and exposed as a single logical resource through a standardized OpenAPI spec.
- Telco disintermediation – The service moves private connectivity, historically carrier-sold MPLS and private-line revenue, under hyperscaler control, though telcos can reposition as integrators and POP hosts.
- Built for AI workloads – Private high-throughput links let teams train models in one cloud while keeping regulated data in another, and free preview pricing seeds adoption among data-heavy pipelines.
Microsoft and AWS have jointly launched a managed multicloud interconnect designed to create private, high-speed connectivity between Azure and AWS. The collaboration merges AWS Interconnect – multicloud and Azure Multicloud Interconnect into what the companies describe as a single logical resource, exposed through a standardized OpenAPI specification. In practice, that means customers can treat a cross-cloud link as just another cloud-native resource rather than a bespoke networking project.
Both companies frame this as removing the complexity that has historically defined multicloud networking. Instead of procuring circuits, configuring routers and BGP sessions, and doing capacity planning across two providers, teams provision an interconnect via API, console, or infrastructure-as-code — the same way they’d spin up a VM or a storage bucket. Microsoft says the Azure–AWS pairing is the combination customers ask about most often.
Ironically, both Microsoft and AWS spent years downplaying the need for deep multicloud integration, but now they’re building a bridge directly between their infrastructures. Whether that’s customer demand finally winning out or competitive pressure in the AI era, the result is the same — cross-cloud networking is now a first-class, jointly engineered product.
How the tech works
Under the hood, the service is built on familiar foundations. Azure Multicloud Interconnect rides on Azure ExpressRoute and AWS Direct Connect, but with a key difference from the traditional model — the physical and logical connectivity is provider-managed rather than stitched together by the customer through colocation cross-connects. On the AWS side, the tech provides Layer 3 private networking between VPCs and other cloud environments, integrating with AWS Transit Gateway and Cloud WAN so routing and segmentation across clouds fit into existing AWS constructs.
The service is currently in public preview across four cross-cloud region pairs — US East, US West, Australia/APAC, and Germany/Europe. And while the companies commit to up to 100Gbps from day one at general availability, current preview options top out at 1Gbps (one Interconnect per customer per supported region during the preview), with formal SLAs and higher-bandwidth tiers still to come.
It’s also worth noting that Azure is a relative latecomer to this party. AWS’ multicloud interconnect service launched in preview in late 2025 as a Google Cloud-only service, hit general availability in April 2026, and only then expanded to Oracle Cloud Infrastructure. The Azure pairing is the third cloud to adopt the spec — but arguably the most significant, given how often the two clouds coexist inside large enterprises.
The impact on telcos
For telecom operators, this announcement deserves close attention, because it shifts a traditional value chain directly into hyperscaler control. Private circuits, MPLS links, and IP/VPN connectivity have long been the telco’s business when enterprises needed to stitch together data centers and clouds. This joint interconnect is, in effect, a cloud-managed virtual private line — with AWS and Azure orchestrating the physical path, routing, redundancy, and encryption that carriers historically provided and billed for.
To be clear, telcos aren’t cut out of the picture entirely. They still own and operate the underlying fiber, metro, and long-haul transport these interconnects ultimately depend on. But the customer relationship and the abstraction layer are moving to the cloud providers, and that’s where the margin lives. When an enterprise can provision a private Azure–AWS link through an API instead of ordering a circuit from a carrier, high-margin private line and managed MPLS revenues start to erode. That’s the disintermediation scenario, and it’s a real one.
That said, there are plausible paths for telcos to reposition rather than retreat. One is to act as strategic integrators — bundling multicloud interconnects with 5G network slices, SD-WAN, and edge computing to offer end-to-end connectivity from device to multiple clouds, something neither AWS nor Microsoft can deliver alone. Another is physical. Telcos and colocation providers can serve as landing zones and POP hosts for these interconnects, offering proximity, last-mile access, and custom SLAs or security controls layered on top of the hyperscaler service.
The standardized APIs are interesting from a telco perspective too. They hint at alignment with telecom network API initiatives, pointing toward a future where an AI workload could request specific connectivity properties across both the cloud layer and the underlay network in a single programmable motion. If that materializes, telcos that expose comparable interfaces for their networks become participants in the stack rather than plumbing beneath it.
And there’s a more mundane, internal opportunity. Telcos are themselves large multicloud customers, and many run OSS/BSS and network analytics workloads split across Azure and AWS. A high-throughput private path between the two clouds could simplify consolidating those systems — the interconnect as a tool for their own operations, not just a threat to their product portfolio.
Distributed AI workloads
The clearest use case for all of this is AI. Distributed AI architectures increasingly span clouds for practical reasons — GPU training capacity might sit in AWS while proprietary data, compliance tooling, or a preferred AI platform lives in Azure. Historically, connecting those environments meant either routing traffic over the public internet or engaging in telco-style procurement for private circuits at every step. A jointly managed interconnect reduces that friction considerably, letting teams treat the cross-cloud link as another programmable resource in their existing IaC workflows.
Security and compliance are a big part of the appeal. Private links let organizations move large language model training data, model checkpoints, and inference traffic between providers without ever touching the public internet — which matters for regulated sectors, and for anyone who wants predictable performance for data-heavy pipelines. It also makes resilience and vendor diversification strategies more attainable. Training in one cloud while keeping regulated data in another stops being a bespoke networking project and becomes an architectural choice.
The pricing is telling, too. During the preview, Azure isn’t charging service fees or Azure-side data egress charges for the interconnect — which is an aggressive incentive to seed adoption among exactly the data-heavy AI workloads this service targets. Production pricing remains unspecified, and that’s the open question. Free previews have a way of looking different at GA, and whether the eventual pricing undercuts traditional telco private line rates will go a long way toward determining how disruptive this actually is.
Table of Contents
