Goh Boon Huat of Singtel Singapore explains to Contributing Editor Annie Turner how identifying emerging tech trends, anticipating changing business needs and rigorous cost discipline allows the operator to stay ahead of the curve

Singtel Group plans to become the world’s most intelligent connectivity company. Goh Boon Huat is contributing to that goal in Singapore’s domestic market as VP, IP & Data Network Engineering at Singtel Singapore. Goh has been in his current role since July 2021, responsible for the island’s IP transport, broadband and data networking infrastructure, reporting to the operator’s CTO.

“Infrastructure is the core foundation for many things. Network planning has always been long term for us,” Goh says. The number one driving factor is Singtel’s emerging business needs. We also look at developing technology to see “what is possible next,” he says. “Operationally, we assess what we have to change to run a tighter, more effective operation.”

World-first commercial 5G SA

About 10 years ago Singtel began planning the consolidation of its almost 20 exchanges and network infrastructure across Singapore, which is about half the size of Greater London. The plan was to consolidate the transport infrastructure into four sites, across the north, south, east and west of the island. 5G and the network services were to run on top of this infrastructure with its superior redundancy and capabilities.

“5G requirements are more stringent than just bandwidth – you need 10Gbps capacity to support each base station site, The clocking requirement is extremely precise, down to sub micro-seconds. If you don’t plan for it, you can’t support the services,” he states. “Five years ago, we were the first in the world to launch a country-wide commercial 5G SA [standalone] network. That’s an example of our forward planning.”

Consolidating IP transport infrastructure

Next Singtel turned its attention to the IP transport network which comprises numerous platforms and diverse technologies. Singtel recognised that two major trends – cloud-based, software-defined networks and segment routing over IPv6 (SRv6) – were driving tech convergence. Goh notes that consolidation was essential “to prepare for what was to come.”
In parallel, much of the last two to three years have involved extensive preparation for AI. “We need to think about the network for AI as well as AI for the network,” he says. “AI will drive huge demand for capacity and it will be a game-changer for operational efficiency.”

To meet that future need for increased capacity, Singapore’s government offers an incentive for operators to make 10Gbps available to every home. Singtel already held a majority market share and, having completed the necessary infrastructure upgrades in what Goh describes as “record time”, now has a fully 10Gbps-ready network. Although people don’t yet need that capacity, he predicts, “It will become the norm. We are already testing 25 Gbps in the lab and will also be looking into 50 Gbps.”

Being ahead of the demand curve is part of the Singtel’s strategy; another is having controlled, predictable costs. When planning the network consolidation, Singapore Singtel presented all its requirements to vendors as a five-year partnership. “For 10Gbps I wanted the cost to be little more than for 2Gbps, for example,” he explains.

Dealing with data

Goh continues, “We are also exploring AI [for the network] – you need the data and telemetry from the network to implement AIOps. The network telemetry data, the network configuration information, have to be made available to the AI layer before AIOps can happen. We need to go deep into the network to extract the data to create AI use cases that make the life of engineers easier.”
This is not easily achieved. He says, “I think a lot of AI starts at the fringe with chatbots because getting into the core is deeply technical. You want deep AI that can answer questions about your network, to use AI to make critical decisions. It took us a long time to streamline the data so that we could develop these AI use cases.”
Singtel Singapore recently completed three or four critical AI pilots. One was for fault management. The aim was to ask AI if the network is degrading and if so, to provide a recommendation. AI is beginning to demonstrate it can ‘converse’ in this way, Goh says. He adds, “I can ask the AI tool to give me a network health-check report and more”.
However, he warns that applying AI in the core network remains “hugely complex and a lot of time is required to test the use cases and prepare the data,” which adds difficulty although “the pilots…show it can work in specific network modules.”

A thousand flowers bloom

The approach to developing use cases is, explains Goh, is to “let a thousand flowers bloom with a top-down framework. Then depending on whichever [use case we choose], to create its own bottom-up business case. This is a very stringent approach. It has to boil down to tangible outcomes for the business. If we cannot be sure [about the value], we will stop the pilot.”

Of course, different stakeholders sometimes have different priorities. A vendor might recommend a solution but the engineer say it does not address an operational pain point. This requires a multi-party investigation into the impact of the fault highlighted by the engineer, the potential factors involved in fixing it and whether that work can be scaled up. Goh states, “An AI project is only good if it can perform a useful job.”

This emphasis on outcomes and the number of stakeholders involved has made the overarching “process of network consolidation extremely challenging,” Goh states. “We consolidated the exchanges and network but if a business stream is not convinced, or we cannot articulate value, then it’s a problem. This is not about technology. It’s about finding the best way to run the network considering the multiple stakeholders’ interests and their multiple constraints.”

Policy and frameworks

In parallel, Singtel is constructing and setting policy around the use of AI data. “We have established an architectural framework,” he states. “This too is both a top-down and bottom-up approach. We are guided by a blueprint with actual use cases. We also need our AI architect to look at what model to choose, what MCP [Model Context Protocol] gateway we have to construct, how we do the layering to make sure that once the service layer and resource layers are completed, we can plug back into the overall orchestrator.”

One example is broadband transport. He says, “We can see the fault and recommend a solution, but this is just one section of the whole network. To be complete, we need to extend the view to the core network’s back-end systems and all the way to the submarine cable system. We do the access parts first to make sure it works and progressively extend it to the rest of the network.”

This process includes making data available to agents in call centres for customers’ calls. “We have to stitch that part in as well to both our ‘western stack’ and ‘eastern stack,” he notes. “The architecture must be thought through in advance so that all the pieces can come together.”

Next steps

Looking forward, Goh says, “Superior uptime of the network is critical. We’re challenging ourselves with a framework we call 1, 5, 60. This represents the goal of knowing there’s a problem within one minute, understanding its nature within five minutes and fixing it within 60.”

Sixty minutes to remedy may be unrealistic currently but making it achievable is part of operator’s next phase of resiliency planning. As Goh observes, the expectation is that the network never goes down because network outages cause such serious disruption.

The key to achieving 1, 5, 60 is greater automation, including that enabled by AI. For instance, the team makes thousands of network changes annually – typically carried out between 1am and 4am – but the volume is growing and so is the pressure to complete them faster.

“We have no choice but to automate – that’s why we came up with a plan to get to Level 3 or even 4,” explains Goh. “We are cautious regarding Level 4 because it is fully autonomous – but that’s the objective. It’s not 100% AI, but AI will go a long way to helping us reduce complexity if it’s done right.”

He concludes, “Today I need to troubleshoot the network to see if there will be any configuration problems. Tomorrow I can ask my AI agent to scout the network to find any weakness that needs patching before it fails. If that can be done…we should never reach the point of an incident.”

Hear from Singtel execs, including Tay Yeow Lian, Chief Technology Officer, Singtel Singapore, at FutureNet Asia, 22/23 September 2026 – Register here.