Chief Technology and Information Office Mark Düsener says telcos must think and operate differently to secure a role in the AI value chain.
How Swisscom keeps pace with AI innovation
One of the big challenges for telcos is how to adapt their organizations to keep pace with AI advances and prepare for the “unthinkable” – that is, AI impacts that are not even imagined yet, according to Mark Düsener, Chief Technology and Information Officer at Swisscom. Speaking during FutureNet World, he said the speed of innovation today requires operators to think differently.
The company is already preparing the ground for thinking and operating differently. It has been on a simplification drive to reduce complexity and enhance resilience since Düsener joined in 2020. The telco is on track to deliver significant cost savings in 2025, as part of a cost transformation program that includes deploying AI and automation across IT and network operations. When it comes to services, it has launched Swiss AI Platform, offering AI infrastructure and GPU-as-a-service to enterprises, leveraging NVIDIA’s SuperPOD system in Swisscom data centers. The operator is also a key player in sovereign AI initiatives in Switzerland and Italy, where its subsidiary Fastweb+Vodafone operates.
TM Forum Inform met with Düsener and asked him about Swisscom’s approach to AI, cloud-native transformation, and the role of telcos in the AI value chain.
Q: You talked about being prepared for an unpredictable future given the pace of AI and GenAI advances. How do you manage in such a fast-changing environment and know that you’re heading in the right direction?
A: There are two aspects to it. First, don’t rush. It is an easy reaction is to say we need to be faster, but then you just run without a direction. You need to have the right foundations in place for AI, and that includes the data, data models, and the ability to automate. We talk about agents taking over the operations, but without automation, they can only tell and they cannot act. That is a journey we have been on for years now. The other aspect is that if you think you’re already bold, be bolder. If you set yourself a target, you should double or triple it. Then you understand that this is not a linear, continuous improvement process, it’s a systemic change.
Q: Are there examples of Swisscom being bold?
A: By the end of next year, we will have reduced 70% of our networks and network platforms through our simplification programme. We had 15 IP networks and now we’re down to one. We will have phased out 2G, 3G, and MPLS. Everything is SD-WAN. We have radically simplified.
Our strategy is to reduce operational efforts by 80%. But whether it is 80%, or 70% or 90%, the point is that it is not an improvement of 10% or 20%. It is not linear thinking, it’s disruptive. To achieve that, we need to think differently in a systematic way. The number itself is less important, the consequence of the number is what matters.
Q: What impact will Agentic AI have on networks?
A: I look at Agentic AI in two ways and we’re exploring both with Outshift by Cisco. One is using Agentic AI to manage our networks. That is easier to understand because you substitute a role with an agent. The other angle is exploring what networking will look like with Agentic AI. If you have agents talking to each other, what will be the communication pattern? Do we need to adjust our networks to optimize? A single request might end up with 100 agents talking, so will we have a 100-times more traffic? Most likely not. But can we differentiate and optimize for certain latency requirements? We just don’t know, but it will have an effect. That’s why we’re working jointly with our vendors and partners to explore these things. If we cannot differentiate the traffic, it will be once again just more traffic without more revenues. We need to be part of the value chain of Agentic AI more than just providing the connectivity.
Q: Do you see the role of telcos being reimagined as a result of AI and automation?
A: At Swisscom, our vision is to be innovators of trust. A fundamental part of our reimagination is the continuation of being one of the most trusted technology providers in everybody's daily life. We are trusted. In a world that is changing faster, trust is a key ingredient to success. So how do we build on that and embrace technologies, but have a trust angle to it? A simple example is having a sovereign GPU-as-a-service offering. We’re in a good position to implement sovereignty because we have done it in the past without naming it or even talking about it. We are the digital backbone of Switzerland and with that comes underlying trust and security.
Q: Swisscom launched a proof of concept two years ago with 5G core network functions running on AWS. How has Swisscom’s cloud native strategy evolved?
A: There were two main reasons for driving that project that I still believe. First, today we are running workloads on telco cloud infrastructure. That is the most inefficient cloud usage because it runs only telco workloads, which have the same usage pattern throughout the day. In the busiest hour, I can have hardware idling of 60%. Using a multi-purpose cloud for telco workloads is more efficient.
The other dimension is that telco workloads and telco cloud are tightly integrated. What you couldn’t solve in software, you could solve in the hardware. If you push them onto a public cloud, the software stack needs to be optimized because the hardware cannot be changed. That helps us to optimize cloud native operations.
The move to cloud native is not about technology, it’s about productivity gains. For the virtualized core that we’re running now, it takes about six weeks to deploy a new version. A cloud native deployment takes about an hour, using TM Forum’s Open Digital Architecture cloud-native components and Open APIs. That is the change, and this is why we’re doing cloud native.
Q: Are there some workloads that will remain on premises?
A: For now, we will not shift everything to the public cloud for different reasons, one of them being compliance. Critical workloads that touch customer data would not be feasible, for example. In future, I can imagine that we’ll have not an either/or, but a hybrid cloud solution.