The focus on AI in the past two years has pushed the demand for faster and more powerful processors to process data, analyse and learn from the information as well as generate responses to human queries.
Yet, the data networks that are needed to deliver the greater volumes of data are often secondary to a conversation on AI infrastructure.
Generative AI can be run from servers in the cloud, but the responses it generates has to be distributed quickly to people around the world, says Rodney Kinchington, BT International’s managing director for Asia-Pacific, Japan and Greater China.
And not enough focus has been placed on the high-speed and flexible networks that have to evolve from more rigid legacy infrastructure, he argues.
“AI’s demands on low latency, high availability, and scalability push beyond what legacy networks can deliver,” he tells Techgoondu in this month’s Q&A. “The focus has long been on compute power, but in reality, it’s the network that often becomes the bottleneck.”
NOTE: Responses have been edited for style.
Q: By most accounts, AI seems to run fine on cloud servers operated by hyperscalers. What new demands is AI making of today’s enterprises and telecom networks that make things challenging?
A: You’re right – AI does run on hyperscaler cloud infrastructure, but that’s only part of the story. What we’re seeing now is that AI, especially generative AI, creates massive volumes of distributed data that need to move quickly, securely, and sustainably between clouds, users, and devices across the globe.
This is creating a surge in east-west data movement, from cloud to cloud, edge to cloud, and cloud to user. Traditional telecom networks were never architected for this. They’re often too rigid and too slow to adapt to over the kind of real-time, deterministic routing needed for AI workloads.
AI’s demands on low latency, high availability, and scalability push beyond what legacy networks can deliver. The focus has long been on compute power, but in reality, it’s the network that often becomes the bottleneck.
Q: What are some businesses and telcos doing to address the issues?
A: We’re seeing more businesses and telcos rethink how their networks are set up because traditional infrastructure just isn’t built for the kind of scale, speed, and complexity that AI demands.
A lot of them are moving towards Network-as-a-Service models that are cloud-first and much more flexible. The focus now is making networks easier to scale, faster to adapt, and smarter to manage.
That means including real-time visibility, software-defined routing, and better energy efficiency because AI consumes a huge amount of power too. Across the industry, the direction is clear: networks need to be more agile, secure, and adaptable to fully support the potential of AI.
At the end of the day, it’s really about moving away from static infrastructure to dynamic, programmable networks that are ready to handle whatever comes next, especially as AI continues to evolve.
Q: When we discuss interconnectivity, where may be the missing links, so to speak, to harness AI better?
A: One of the biggest gaps right now is how data moves between different clouds, whether that’s public, private, edge, or Software-as-a-Service (SaaS).
AI workloads are often spread across all these environments, but the routes that data takes between them aren’t always clear or efficient. That can cause latency issues, performance dips, and unexpected costs, especially around egress fees.
That’s where smarter networking comes in, because what’s really needed is deterministic, programmable networking that allows enterprises to define exactly how, where data travels, and making sure it stays within the right jurisdictions for compliance.
The ability to manage that routing intelligently is becoming just as important as the compute itself. So when we talk about interconnectivity these days, we are talking about the connections that are reliable, secure, and smart enough to keep up with the demands of AI.
Q: How does Asia-Pacific’s unique circumstances, like regulatory regimes, come into the picture when it comes to ensuring good connectivity?
A: Asia-Pacific is a highly diverse region, not just culturally and economically, but also in terms of digital infrastructure and regulatory frameworks.
Connectivity levels also vary considerably, with some markets operating advanced networks while others continue to scale and modernise. At the same time, governments are introducing or tightening regulations around data sovereignty and cross-border data flows.
For enterprises working across different markets, this landscape introduces a layer of complexity as they need to ensure that data moves in ways that comply with local regulations. This makes control over data routing increasingly important, particularly for sectors like finance and government, where regulatory compliance is non-negotiable.
As regulatory expectations evolve alongside rapid tech adoption, especially in areas like AI and cloud, having a network that can adapt and provide clear visibility into how data moves becomes a real advantage in the Asia-Pacific region.
