By Mark Rafferty, CEO, FibreconX
I’ve spoken to network engineers who were completely convinced their infrastructure was resilient. Multiple providers. Separate contracts. Redundant links documented in architecture diagrams.
Until it wasn’t.
Both services were running through the same duct under the same street. When a contractor’s excavator hit that duct, both links went down simultaneously. Everything that was supposed to fail over, failed together.
That’s the difference between redundancy and true path diversity. Redundancy protects you if a cable fails. Path diversity protects you if the infrastructure carrying that cable fails. Most networks in Australia have the former. Far fewer have the latter.
Why ‘Separate Providers’ Doesn’t Mean Separate Infrastructure
The problem is structural. Many legacy fibre networks in Australia were built by the same original infrastructure owners, following the same physical routes, through the same shared conduit systems. Over time, those networks were resold, rebranded, and repackaged by different providers, but the physical layer often remained largely unchanged.
So, when enterprises and operators buy connectivity from two different providers to achieve ‘diversity,’ they’re frequently buying two services that share the same ducts for some or all of their routes. The commercial layer is separate. The physical layer is not.
This is a distinction that matters enormously when something goes wrong — and in infrastructure, something always eventually goes wrong.
“Resilience isn’t about how quickly you recover. It’s about whether failure impacts you at all. And that comes down entirely to whether your physical infrastructure is truly diverse.”
The Increasing Cost of Failure
As infrastructure becomes more distributed and workloads become more interdependent, the blast radius of a network outage grows. In a hyperscale data centre environment, a connectivity failure between facilities doesn’t just pause a workload — it can cascade across dependent systems, affect model serving environments, disrupt real-time decision processes, and create customer-facing failures that are difficult to remediate quickly.
For AI workloads specifically, the consequences are compounded. These environments are always-on, data-intensive, and highly sensitive to interruption. They’re often running inference at a scale where even a brief outage creates backlogs that take time to clear long after connectivity is restored.
What True Path Diversity Actually Requires
Genuine network resilience — the kind that actually protects you when something goes wrong — requires physical separation at the infrastructure layer. Not just separate logical paths. Not just separate providers. Physically separate duct infrastructure, following distinct routing paths, with no shared physical components between primary and backup routes.
That’s a high bar. And it’s one that most networks, if you map their physical routes honestly, don’t meet.
A practical audit of your current resilience should ask: Do you know the physical duct routes of your primary and backup services? Have you confirmed with your provider that those routes don’t share any common conduit? If your provider can’t answer those questions with confidence, you’re working from an assumption rather than a guarantee.
How We Engineered Resilience into FibreconX
FibreconX is the first new fibre network built in Australia in over 20 years. Every other carrier is operating infrastructure that predates the data centre era entirely – networks designed for a different world, now being asked to carry workloads they were never conceived to handle. Resilience wasn’t something we could bolt on after the fact, it had to be a founding principle. So, we built our own duct network, independent of legacy shared conduit. We designed distinct routing paths between key data centre locations, and we built with enough capacity that congestion simply isn’t part of the equation.
The combination of physically diverse infrastructure, independent routing, and low current utilisation creates a meaningfully different risk profile than networks relying on shared legacy duct systems, regardless of how many providers or contracts those networks are procured through.
“AI workloads don’t just need low latency and high bandwidth. They need infrastructure that keeps running when something goes wrong, because something will always eventually go wrong.”
The Question to Ask Your Provider
If you’re procuring connectivity for critical workloads, AI inference environments, multi-site training infrastructure, hyperscale interconnect, the right question isn’t ‘do you have a redundant path?’ It’s ‘can you show me the physical routes of both paths, and confirm they share no common duct?’
The answer to that question will tell you more about your actual resilience than any SLA.
→ Ask us about diverse fibre routing across Sydney and Melbourne: fibreconx.com.au/contact