The most interesting networking launch this week did not ship a box
It shipped an abstraction — and abstractions are where blast radius hides.
September 2, 2026: Equinix announced Fabric One, a managed any-to-any connectivity service across enterprise, cloud and AI environments, with AWS and Google Cloud as lead integration partners.
WHAT ACTUALLY CHANGED
The model flips. You no longer describe the path — you describe the outcome. State what needs to reach what, and under which requirements, and the service determines and delivers the connectivity.
It runs on the open-source OpenAPI 3.0 Interconnect specification that AWS and Google Cloud collaborated on, which is the genuinely notable part: two competing hyperscalers agreeing on an interoperability contract rather than each defending a proprietary on-ramp.
Requests can arrive through a portal, APIs, automation workflows, agent-based requests — or a natural-language prompt.
Underneath, the platform manages routing, cloud connectivity, encryption, resiliency and failover on your behalf.
WHY NETWORK ENGINEERS SHOULD READ THIS CAREFULLY
Distributed AI broke the old interconnect model. Training in one place, inference in another, data somewhere else, all across multiple clouds — the cross-connect-and-ticket workflow simply does not keep pace.
So intent-based interconnect is the right direction. But be precise about what it changes for you.
It does not remove the network. It removes your visibility into it. When routing, failover and encryption are decided by a control plane you do not operate, "any-to-any" becomes an architectural claim you cannot verify — and a flat reachability model you did not explicitly design.
Your value shifts too. Less time on CLI, more on defining segmentation intent, validating that the delivered path matches the requested one, and knowing which failure domains your circuits actually share.
Automation does not eliminate network engineering. It raises the cost of not understanding your own topology.
WHAT TO DO NOW
WRITE SEGMENTATION INTO THE INTENT — any-to-any is a capability, not a policy, so declare what must never reach what before you declare what should.
DEMAND PATH TRANSPARENCY — ask any managed interconnect vendor which underlying carriers and facilities a service uses, and get it in the contract.
MONITOR FROM OUTSIDE THE ABSTRACTION — independent active path testing is your only proof that the delivered connectivity matches the requested SLA.
MODEL THE CONTROL-PLANE OUTAGE — if the orchestration layer fails, know exactly what stays up and what you can still change by hand.
Declarative networking is coming to interconnect. Declare carefully.
Would you hand routing, failover and encryption to a provider's control plane — and how would you verify it?