Cloud Daily edition
CLOUD MULTICLOUD AND FINOPS
AWS and Azure turn private interconnection into a service, not a project
AWS Multicloud Interconnect now links AWS and Microsoft Azure on demand in public preview. The promise is to cut months of circuit provisioning and coordination, but fixed bandwidth, topology, and dual-cloud operations still come at a price.

AWS and Microsoft have opened a managed path to privately connect their two clouds. Since 31 August, AWS Multicloud Interconnect supports Azure in public preview, and Azure offers its end as Multicloud Interconnect. Provisioning is done from the consoles and tools of both providers, without manually contracting and cabling a series of carriers. Initial availability includes North Virginia, Northern California, Sydney and Frankfurt. It is a concrete advance for architectures that are already distributed, not an automatic invitation to split any application across clouds.
The technical foundation combines the well-known domains of Direct Connect and ExpressRoute with a coordinated experience. Each interconnection creates four independent logical paths over two distinct physical locations and routers. The design aims to withstand the maintenance of one router, two additional outages, and even the loss of a site without disrupting traffic. Between the providers' edges, MACsec encrypts the link, and the offering announces four-nines availability. The relevant detail is not just speed, but eliminating some of the grey work of cross-orders, letters of authorization, VLANs, and testing between carriers.
Hourly billing changes based on the topology's most expensive route
The service also changes the economic unit. AWS bills hourly based on the chosen bandwidth and a geographic tier calculated by the path; on its side, it does not add a per-gigabyte transfer charge. There is a free 500 Mbps interconnection per provider and region. In the published example, 10 Gbps at the local tier 1 costs $12.33 per hour, about $9,000.90 for 730 hours. If a Cloud WAN topology incorporates a distant region and pushes the link to tier 4, the same capacity reaches $51.78 per hour, nearly $37,799.40 per month. Azure charges its portion separately.
This model can make spending more predictable for constant and voluminous flows, but it penalises oversized capacity or capacity reserved for long periods without use. Furthermore, the tier is assigned based on the most expensive route the topology can use, not just the proximity of the physical point. Associating a Direct Connect Gateway with a global network without review can multiply the rate. FinOps must come in before the click: measure throughput, peaks, active hours and authorised routes; compare the fixed cost with current transit; and tag each interconnection with an owner, application and review date.
The link does not resolve identity, DNS or shared logical failure
Operationally, the managed network does not resolve DNS, identity, service discovery, overlapping addressing, or firewall policies. Nor does it make an application portable. A database in Azure and a service in AWS can gain stable latency, but they still share a logical point of failure if they depend on the same directory, secret, or deployment plane. Routing must prevent asymmetric routes and leaks; quotas and MTU must be tested; and observability must unite telemetry from both ends. AWS includes Network Synthetic Monitor to locate packet loss on its side, which is useful but insufficient to explain the entire transaction.
The practical application is particularly suited for mergers, regulatory requirements, ISVs serving customers on different clouds, and AI workloads where the data is on one provider and specialised compute is on another. A sensible pilot connects a single pair of non-production networks, limits BGP prefixes, enables logging, measures latency and loss with and without encryption, forces a path to fail, and calculates the cost per useful terabyte. Afterwards, the procedure for provisioning, changing and decommissioning is validated with the managers of both platforms. Provisioning speed is only an advantage if it does not eliminate change controls.
The preview status requires moderating expectations about coverage, support, and final conditions. It is worth monitoring its arrival at general availability, additional bandwidths and regions, the contractual end-to-end SLA, the exact bill from the Azure side, and how incidents are handled when diagnostics cross providers. The news does not prove that multicloud is simpler; it shows that one of its most thankless layers is beginning to look like a product. The rest—architecture, security, cost, and responsibility—still belongs to the customer.
Tags
- AWS
- Azure
- Multicloud
- FinOps
- Networking
- MACsec
BOLDERROR Daily edition Rubén Campoy