There are very few industries today that are not affected by
Artificial intelligence (AI)The world of networking is one that is constantly being touched. It is hard to imagine that any network, no matter how large – be it a home router or an office local network – can’t be accessed.
AI can “just” improve the quality of life.
Take Mark Dusener’s words, Swisscom’s chief technology officer, about the partnership between his company and
Cisco Outshift will be deployed
His organisation is implementing agentic AI (more on this later). “The goal of moving into an agentic AI environment, operating networks, and connectivity, is to reduce the impact of service change, reduce the risk of downtime, and costs, and therefore, level up our customer experience.” It seems simple, right? As we all know, nothing is easy in this world, and AI cannot be “just” turned on to achieve such gains. The benefits of AI for networking cannot be fully realised without considering AI networking.
Start with Nvidia.
Any investigation of the technology would be logical.
AI and networking or, indeed, AI in general should be the first step.
Nvidiais a company which has played an important role in developing the AI technology ecosystem and will continue to do so.
In 2024, Nvidia CEO Jensen Huang spoke at a conference about the importance of AI in business.
GenAI (generative AI) has arrived and enterprises must embrace “the most consequential technology of all time”. He told his audience that the greatest fundamental computing platform change in 60 years is happening, from general-purpose computing towards accelerated computing.
We’re sitting on a huge mountain of data. We’re all doing it. We’ve been collecting data in our businesses for years. We have been collecting data for a long time, but until now we didn’t have the ability to refine it, discover insights, and codify them automatically into our natural experience as a company, our digital intelligence. Every company will be an intelligence producer. Domain-specific intelligence is the foundation of every company. “For the first time, this intelligence can be digitised and turned into our AI. The corporate AI,” said he.
AI is a cycle that lasts forever. We want to turn our corporate intelligence (or business intelligence) into digital intelligence. Once we have done that, we will connect our data to our AI flywheel in order to collect more data, gain more insight, and create better intelligence. This allows us to offer better services, or to be more productive. We can run faster, be efficient, and do things on a larger scale.
This exactly what Swisscom aims to achieve. Swisscom is the largest telecoms company in Switzerland with over six million mobile subscribers and 10,000 mobile antennas that need to be managed efficiently. When network engineers make changes to infrastructure, they are faced with a common problem: how to update systems serving millions of customers without interrupting service.
The partnership with Outshift was the solution to “redefine customer experiences” by using AI agents. This is using
Outshift’s Internet of Agents ( ) will deliver meaningful results to the telco while also meeting customers’ needs through AI innovation.
Hon Kit Lam, Tata Communications’ VP of hybrid services, believes that AI has a wider impact on networks. AI is rewriting networking rules, turning static infrastructures into self-adapting, ever-smarter ecosystems. AI automates complex processes, improves real-time monitoring, and detects faults pre-emptively before they reach our customers.
These technologies are helping to shift from reactive firefighting operations to proactive, self healing operations, especially as isolating the root causes and launching correctional measures is done without any human intervention,” he notes. This technology dynamically divides the network to contain and control threats and allocate resources efficiently. It’s a combination intelligence and agility which not only improves operational efficiencies but also elevates customer experience, making network truly future-ready.
Stephen Dodge technology director at IT service provider
Bistech believes that the software-defined wide-area network (SDWAN) is a good example of how AI can improve networks. “Cloud-managed architecture is a key advantage of SD WAN over traditional solutions such as MPLS. This foundational design allows leading vendors to easily integrate AI, such as generative AI, into SD-WAN offerings,” he suggests.
As AI evolves, “we anticipate that it will enhance SD-WAN by simplifying configurations and improving performance by dynamically adjusting to user behaviour and unusual patterns of traffic, as well as bolstering observability and safety.” Generative AI creates smarter, more efficient, and agile networks, reducing manual intervention and simplifying the management. AI can help small and medium businesses gain access to enterprise technology, allowing them to focus on their growth while eliminating the costs and infrastructure issues that arise with managing complex IT infrastructures.
Engineer networks for AI
In a broader context, Swisscom and Outshift also showed that to make AI work effectively, something new is required: an infrastructure which allows businesses to communicate and work together in a secure manner. Here’s where AI and networking play a role. David Hughes, HPE Aruba Networking’s chief product officer, stated that there are pressing issues regarding the use of AI within enterprise networks. This is especially true when it comes to leveraging the benefits GenAI can provide. Hughes argues that there are subtle, but fundamental differences between “AI for networking” (also known as “networking for AI”) and “AI for network”.
From an engineering and data-science perspective, “AI for Networking” is where we invest our time. It’s about [questioning] using AI technology to make IT admins super-admins, so they can handle their escalating work load independent of GenAI. GenAI is a burden on top of all the other things, like escalating cyber threat and privacy concerns. He noted that the business was constantly asking IT to do things and deploy new apps, but the number of employees remained the same.
We are beginning to see and expect more AI computing at the edge, to eliminate the distance from the prompt to the process.
Bastien aerni, GTT.
“Networking for AI” is first and foremost about building the kind of switching infrastructure needed to interconnect GPU clusters [graphics processing unit] . Then, a little beyond that, think about the impact of collecting data on a network. And the changes to the way people want to build their network.”
There is impact. Many firms that are currently researching AI in their business find themselves wondering how to manage mass adoption of AI with regards to networking and data flow, such as what kind of bandwidth and capability is required to facilitate AI generated output, such as text images and video content.
This, says Bastien Aerni, vice-president of strategy and technology adoption at global networking and security-as-a-service firm GTT, is causing companies to rethink the speed and scale of their networking needs.
To achieve the return on their investment in AI initiatives, companies must be able secure and process large volumes of data quickly. Their network architecture needs to be configured to support such a workload. He says that a platform embedded into a Tier 1 [internet protocol] network backbone will ensure low latency, high capacity and direct internet access worldwide.
We are beginning to see and expect more AI computing at the edge, to eliminate the distance between prompt and process. Software-defined wide area networks [SD-WAN] built into the right platform can be used to route AI data traffic efficiently, reducing latency and security risks and providing more control over data.
Managing Network Overload
By the end of 2023 BT revealed its networks were under enormous strain due to the simultaneous online broadcast of 6 Premier League football games and downloads of popular video games. The update ofCall of Duty Modern Warfare (was particularly cited. AI is expected to add to the headache.
Speaking to Mobile World Congress 2025 attendees, BT Business Chief Technology Officer (CTO) Colin Bannon stated that a robust, reliable network is a prerequisite for AI to function. It also requires effort to remain relevant and meet the ongoing challenges faced by BT’s customers, mainly multinationals, governments, and international businesses. In a world in which “slow is new down”, network performance is critical to support an AI-enabled future.
Bannon said that Global Fabric, BT’s network-as – a-service, was built before AI “blew up” and that BT had been thinking about how to deal a hyper-distributed workload on a network, and be able make it fully programmable.
In describing the challenges and how the new networking will solve them, Bannon said: “[AI] only makes distributed and complex workflows more large, which makes it even more important for a fabric type network.” You need a network which can [handle data] burst and is programmable. It should also be able to [control] bandwidth as needed. This level of programmability [is something businesses] has never been seen before. I would argue that the networking is the computer and that the network is necessary for AI to function.”
This would result in constructing enterprise networks which can cope with the massive pressure placed on utilisation by AI, especially when it comes to what is required for training models. Bannon said that there are three main network challenges to deal with AI. These are the training requirements, the inference requirements, and general requirements.
According to him, AI workloads are dynamic, so networks must be scalable, agile, and equipped with visibility tools for real-time monitoring and issue detection. AI training requires the movement of large datasets over the network. This calls for high-bandwidth networks.
In addition, he described “elephant flows” of data – i.e. continuous transmission over a period of time and training that takes place over several days. He warned that network inconsistencies can affect the accuracy of AI models and their training time, and that tail latencies could have a significant impact on job completion times. This means that congestion management is required to detect congestion and redistribute traffic on the network.
AI training models can cause network problems. And now the conversation is turning from the use of generic large language models ( seePreparing networks for Industry 5.0 box ) to application/industry-dedicated small language models.
He says, “[That is] literally deployment on the edge of the networks to avoid flooding the network. Also, privacy concerns are addressed, as well as sustainability concerns about some of these large language models that are very specific when creating domain context.”
Examples of this can be used for video analytics, media analysis, and capturing conversations in real-time, but not by deploying them out to flood the networks. The flip side was that there was a lot of power in these central hyperscale models and capabilities. You also need to know more [about] about what is the right network background and what’s a good balance of your network infrastructure. If you want to stream real-time media from a [sports stadium] while editing all the content on-site or remotely, you’ll need a different network backbone. “This looks like an interesting area of technology relevant for the supporter experience at a stadium. It’s dampening and sound targeting. Then we return to the AI story’s edge. This is exciting for us. “That is the frontier.”
But when we return from the frontier of tech to the everyday operations of businesses, even though the IT and communications community is confident it can handle any technological issues that may arise regarding AI and network, the businesses themselves may be less sure.
Roadblocks for AI PlansResearch by managed network as a service provider Expereo published in April 2025 revealed a number major roadblocks for AI plans in UK businesses. There are many roadblocks, including unreasonable demands from employees as well as poor infrastructure.
One of the key findings from Expereo’sEnterprise Horizons 2025was that many UK technology leaders felt that the expectations of their organisation about what AI can achieve are growing faster than the ability to meet those expectations. While 47% of UK organizations noted that their infrastructure/network was not ready to support any new technology initiatives such as AI in general, another 49% reported that the performance of their network was preventing or restricting their ability to support large-scale data and AI projects.
Expereo CEO Ben Elms, assessing the key trends revealed by the study, says that as global companies embrace AI to transform the employee and customer experience will be crucial to ensure that AI delivers long-term benefits, and not just a quick fix.
While the potential of AI is enormous, its successful integration demands careful planning. He says that technology leaders must recognize the need for robust networking and connectivity infrastructure in order to support AI at a large scale, as well as ensuring consistent performance on these networks.
Elms summarizes the current state of the industry by stating that strategic investments in IT infrastructure and technology are needed to meet both present and future demands. Swisscom’s goal is to reduce the impact of changes in service, reduce costs and downtime, and improve customer service. This is reflected in Dusener’s statement. Stephen Dodge, a Bistech expert, warns that while the industry is rushing to push “AI-ready” technologies, most networks aren’t yet ready. He believes that many are still struggling with the demands of cloud computing, as-a service models, and distributed workforces, and AI could push these networks beyond breaking point.
Before investing in AI, businesses should ensure their networks are capable of handling the workload. This means eliminating performance bottlenecks, scaling up to meet increasing demands, and simplifying the management. As AI advances cyber threats, a secure, strong network becomes the first line of defence. “Speed, security and simplification are core networking requirements for the AI era. Organisations who invest in modernising networks now will be best placed to unlock AI’s potential in the future.”
Simply turning on any AI system, and believing that there is a solution “out there”, won’t work. Your network may very well tell you the opposite.
