Elastic bandwidth and the future of AI-driven networks

In this employee spotlight blog, Shaheen Kalla, Presales Team Lead, explores what the future of AI in networking may hold and the possibilities it presents.

So much has been written about AI in the context of software engineering, machine learning, and data manipulation - especially where large datasets are involved. However, very little has been explored when it comes to AI from a networking perspective.

Networking is still seen very much as a dark art by many software engineers, data analysts, and database administrators. It’s often referred to as “the piece of string” – nobody truly understands how the string works, especially not those working in current AI technologies. So, what is the future of AI within the networking space?

The static nature of today’s networks

Currently, network deployment is based on fixed bandwidth. While SD-WAN has transcended Layers 5 and 6 to understand software and application-level requirements, it still relies on static-sized bandwidth pipes. This also applies when traffic enters a network provider’s backbone: multiple large pipes of fixed size transporting packets to their destinations. This model has served us well for many years.

However, as data becomes increasingly distributed – stored across multiple locations, sorted and tagged for intended (and sometimes unintended) uses – there is a clear need for networks to evolve.

Introducing Elastic networking

So, what does the future look like? Elastic networking at both the core and the edge, powered by AI algorithms that can provision bandwidth dynamically, based on key parameters.

1. Origin-based bandwidth allocation

The first and most important parameter is origin. The ability to scale bandwidth up or down based on source pathing would be a major shift in how we use bandwidth.

Imagine a single global pool of bandwidth. As Asia comes online and data flow increases, the system could scale bandwidth (whether internet, private, or cloud-based) up to handle the surge. As Europe wakes, the network scales European backbone and office capacity while tapering down in Asia. When the Americas begin their day, Asia’s usage is reduced to a trickle, and bandwidth is shifted westward.

In this scenario, AI would manage backbone network parameters via scripts and API calls – what could be termed path-based AI networking.

2. Destination-aware routing

The second parameter is destination. Where is the traffic going? How quickly does it need to get there?

AI algorithms could analyse traffic patterns over time from a given office or location and dynamically adjust routing. This wouldn't just use the best path, it would optimise bandwidth allocation across those paths.

To make this possible, telecom providers would need to collaborate more closely, allowing for a degree of cross-configuration. Since all carriers are interconnected anyway, why not open up the model to allow large bandwidth pools to be purchased and manipulated based on the end customer’s needs?

3. Application awareness

The third key parameter is application awareness. While SD-WAN already offers some capabilities here, coupling it with AI could unlock advanced features – predicting an application’s requirements in real time and adjusting both QoS policies and upstream network configurations.

This would be especially beneficial in latency-sensitive sectors like e-gaming, betting, and financial trading, where path optimisation directly impacts performance.

What’s next?

These are just a few of the features that could benefit from the introduction of AI into networking. The biggest hurdle will be network operators’ willingness to let customers manipulate their networks. But it's not beyond the realm of possibility. We could very well see pools of bandwidth becoming the standard, replacing fixed, end-to-end or mesh-type environments.