By Sean Rinas, Head of Network Operations
AI infrastructure conversations tend to focus on compute – GPUs, cooling, high-density power delivery design. But the network is just as critical. Every model trained, every dataset moved, every inference run depends on a network that performs under pressure.
It’s no longer enough for that network to be fast. It needs to be intelligent – able to adapt, optimise and stay one step ahead of workload demand. Issues like lag, failure and overengineering aren’t rare exceptions anymore. They’re everyday risks, and they’re expensive when left unchecked.
That’s why we’re building intelligence into the network itself – not to replace human expertise, but to take the pressure off the teams who run it. Everything just runs better when the infrastructure can spot trouble early, tune itself in real time while optimising routes and improving behind the scenes with little or no human intervention.
Most networks aren’t wired that way. They’re reactive by nature, built to absorb faults rather than prevent them. And while monitoring tools have come a long way, they still miss the early warning signs – the ones hiding in patterns too complex or too subtle to easily catch manually.
This is where AI starts to shift things. By analysing how signals interact – across power, cooling, load, location and system behaviour – we can start predicting how the network will respond, long before anything actually fails.
You can’t scale AI without interconnection. But managing that interconnection proactively at speed – and without complexity – is what comes next. Here’s what that looks like inside NEXTDC’s AXON virtual interconnection platform right now.
We’ve built a live digital twin of AXON – a software-based replica that mirrors real-time performance. That means we can test how the network behaves under different conditions without touching production. It boosts the operational certainty of the platform. Whether it’s load spikes, link failures or routing changes, we can simulate all of it before making a move.
This work started in partnership with La Trobe University, through a research collaboration focused on practical ways AI can improve network operations. What began as a concept has evolved into a working tool we now use to model risk, tune capacity and improve planning.
It helps us scale more precisely and catch issues early, before they affect performance. It also gives us a safer way to trial network changes before they go live, reducing downtime and giving more consistent performance, even during upgrades.
For customers, that means operational certainty – fewer surprises and a more reliable experience overall.
We’re also using graph databases to spot patterns most monitoring tools would miss. Instead of looking at data points in isolation, this approach maps how all the moving parts are connected – for example, config changes, power fluctuations, cooling shifts, traffic patterns – and flags issues before they stack up.
In the past, these signals might have gone unnoticed or been treated as isolated incidents. However, when analysed together, they can reveal developing issues that would otherwise surface too late. We’re already seeing how this improves uptime and reduces troubleshooting time, giving teams space to focus on high-value work instead of chasing anomalies.
The long game is improved customer experience through automation, and this development gives our teams a head start. We can catch problems early and resolve them faster, helping us stay ahead of the kind of noise that can drag down performance. We want to be able to proactively reach out to customers, to say: “hey, you might have a problem”, well before they come to us saying that “we have a problem”.
It’s a cleaner way to manage AI complexity – less time spent chasing anomalies, more time focused on the parts of the network that actually move the needle.
AI infrastructure shouldn’t introduce more complexity than it solves. By building intelligence into the network layer, we’re cutting down on noise, reducing operational drag and giving teams more room to move.
For NEXTDC, it’s about providing the kind of operational certainty foundations our customers are asking for – one that provides stability without overhead, control without friction. And a network that keeps getting better – without constant babysitting.
If you want to see how your network measures up, grab your copy of our Interconnection Excellence Checklist. Or reach out to our team if you’d rather talk it through. We’re always up for a straight conversation about what’s working, what’s not and what comes next.