Analysis
Microsoft and AWS each launched managed interconnect services this week -- Azure Multicloud Interconnect, currently in preview, and AWS Interconnect for multicloud -- providing private high-speed links between the two clouds, The Register reported. The services deliver up to 100 Gbps from day one, four-nines availability, and MACsec encryption, and are built on open API specifications AWS developed jointly with Google Cloud.
Before this, connecting the two required coordinating physical cross-connects, routing, provisioning, monitoring and lifecycle management by hand, a process that could run weeks or months. Narayan Annamalai, who heads product for Azure Networking Services, and Robert Kennedy, AWS VP of network services, both framed the work as removing that operational tax. AWS Interconnect now covers Azure, Google Cloud and Oracle Cloud Infrastructure. Pulse has previously covered AWS across its cloud and chip infrastructure moves.
The regulatory backdrop is the reason this is interesting rather than routine. AWS told UK competition regulators that multicloud posed "no significant technical barriers." Microsoft told them customer interest in multicloud was minimal outside "fairly specialized situations." Both statements were made while the practical experience of connecting the clouds took months. Shipping the fix is an implicit correction of the record, arriving after the CMA's cloud market investigation had already forced the industry to cut egress fees.
“Moving terabytes across the public internet between them is slow and expensive; a private 100 Gbps path with predictable latency changes the architecture.”
The AI workload pattern is what actually forced it. Enterprises are increasingly training or fine-tuning where the accelerators are available and serving where their data and applications already live, which are frequently different clouds. Moving terabytes across the public internet between them is slow and expensive; a private 100 Gbps path with predictable latency changes the architecture. Anthropic sells through Bedrock and Microsoft Foundry; OpenAI's models reach enterprises through Azure. Customers running both are common now in a way they were not in 2023.
There is also a competitive read for startups. A whole generation of networking companies -- Aviatrix, Alkira, Megaport and the cross-connect layer of Equinix and Digital Realty -- built businesses on the difficulty this announcement removes. Megaport in particular sells network-as-a-service specifically because stitching clouds together by hand was painful. When the hyperscalers ship a first-party managed service with four-nines availability, the independent layer either moves up the stack into policy, observability and cost governance, or it gets commoditized. Aviatrix has been making that move toward multicloud networking policy for two years, which now looks prescient rather than defensive.
What is still missing from the announcements is pricing. Neither company disclosed the fee structure, and egress economics are where multicloud has always died. A 100 Gbps link that is technically excellent and priced per gigabyte transferred can still make cross-cloud architectures uneconomic. Until the rate cards are public, this is capability, not portability.
