Dimitra Simeonidou: Embedding AI into telecom networks

• Professor Dimitra Simeonidou of Bristol University on how connectivity will be delivered, managed and monetised in future telecom networks 
• The age of “end-to-end AI” is dawning
• AI to be embedded in the deep fabric of the new network and will no longer be just an app running on top of it
• Disruption and major changes to traditional network architecture necessary for the potential of embedded AI to be realised

Professor Dimitra Simeonidou of the University of Bristol has been working in the telecom sector for more than 30 years. Her research focuses on the areas of high-performance networks, AI-native networks, programmable networks, wireless-optical convergence and smart city infrastructures.

She is also the director of the UK university’s Smart Internet Lab, one of the most renowned research laboratories in Europe. Under its aegis, she is also the leader of JOINER (Joint Open Infrastructure for Network Research), a major UK national-scale initiative that links universities and the telecom industry via a collaborative experimental platform that connects multiple research labs and testbed innovation in 6G technology and future networks: Simeonidou discussed JOINER development previously with TelecomTV in this video interview conducted at FYUZ25.

In an exclusive interview, Professor Simeonidou told TelecomTV, “We believe that with open platforms and democratised experimentation we can drive and pilot innovation at a network scale and by doing so we can help the industry, regulators and policy makers to be forward-looking. We can culture innovation in ways that service providers’ infrastructure cannot… we look very much like a service provider network, but it’s an open network where we can look at the effects of AI, security, architecture, and so on.”

Importantly, and impressively, Simeonidou has also co-founded two spin-out companies, the most recent being venture-capital funded Zeetta Networks, which was spun out of the University of Bristol and which develops software-defined networking (SDN) platforms for enterprise networks. She has proven, in advance of many other academics, that she can talk the talk and walk the walk of spin-out commercialisation as the UK’s big university research sector begins to ride the growing wave of entrepreneurial activity within British academia to commoditise their research by transforming high-tech innovations from their R&D labs into operations that can be quickly spun out as market-ready companies with real scalability and high-growth potential.

TelecomTV first met the Professor back in March 2018 when the University of Bristol’s Smart Internet Lab, in collaboration with BT and Nokia, showcased a live outdoor massive MIMO (multiple-input multiple-output) technology demonstration during the ‘Layered Realities’ public 5G event in and around Bristol’s Millennium Square. The trial used a 128-antenna array to deliver simultaneous, spatial multiplexing of live video streams to multiple users over a 3.5 GHz carrier frequency. 

The alfresco event took place during what was, for England, an exceptionally snowy and Arctic-cold period that the daily press called “The Beast from the East”. It disrupted travel and communication across much of the country, but the day-long Bristol demonstration maintained the age-old British tradition of “carrying on regardless”. The demonstration certainly tested the MIMO technology under less-than-optimal, frigid, conditions but nevertheless proved to be a great success. 

Eight years later, in the intense heat of an exceptionally rare tropical British summer, Simeonidou is strongly focused on AI and advanced connectivity solutions and is running the biggest network project ever funded in the UK. It began in late 2025 and will run until the end of 2027. It brings together all the major communications networking equipment vendors, including the likes of Ericsson, Nokia and Samsung, as well as a large number of universities. 

New AI management and AI orchestration layers will be required

We are now witnessing the dawning of the age of 'end-to-end AI” where the technology will be present in many parts of the network including the radio access network (RAN), packet and transport domains. AI will also be important in software data-driven platforms. Simeonidou and her team are examining network architecture and how it will have to change as AI becomes embedded in the network fabric rather than being an application running on top of it. 

She says that from now on, AI must be considered as part of the network and nested both in network design and in the delivery of the network itself. The professor adds that Bristol University’s own research, taken in combination with data and other inputs provided by the likes of Nokia and Samsung, strongly supports the pressing need for major changes to be made to network architecture at the systems level to optimise particular apps at particular times. 

Another important finding of the project is that a new AI management layer must be instigated – and that will require a new layer for AI orchestration. 

Today, most AI systems are structured in a four-layer stack comprising infrastructure (silicon hardware/cloud), foundation models, orchestration and tooling, and applications and interfaces. However, in the near future, tools that enable global observability and monitoring of networks will be strategically very important in ensuring optimisation and to guarantee quality of service in mobile comms, cloud services and global internet connectivity. 

Thus, Simeonidou states, “We have to make [the] hardware fabric programmable, optimise the hardware side… on the other hand, there is also a need to rethink the software layers and to manage AI in ways similar to the managing of network resources. This is very important. Where AI for networks is concerned, the industry is now aligning with this view, both in terms of observability and actual management of AI. This is in line with ‘native AI’ in terms of network infrastructure.”

She adds: “There is already a lot of compute embedded in the network infrastructure and [as a result] there is also an opportunity for [telecom] service providers. They have an incentive because they have real estate [assets] where more compute can be installed.” 

Furthermore, there is an emerging trend to deploy CPUs (central processing units) and DPUs (data processing units) at the network edge to enable AI processing to take place ‘locally’ as much as possible and to enable compute resources to be aggregated and allocated in an agile way as needed. 

In regard to the ongoing debate about “native AI”, Simeonidou referenced a highly successful demonstration of how to use networks to enable third-party apps. In the demo, held at MWC26 in Barcelona, UK national operator BT Group collaborated with Nvidia and a coalition of global telecom operators and vendors, including Deutsche Telekom, Ericsson, Nokia, SK Telecom, SoftBank, T-Mobile US and Cisco. They utilised 10 distributed nodes of the University of Bristol’s JOINER network to show the need for AI to be embedded in the network. 

The collaboration focused on transforming ‘traditional’ telecom infrastructure into distributed AI compute grids using GPU-accelerated AI-RAN (AI radio access network) architecture. 

What made the demo unique was that rather than issuing yet another re-tread of the nebulous “AI vision” that was routinely trotted out at MWC in previous years, it delivered real field trial results, along with commercial product launches and open-source toolkits. Indeed, the demo participants committed to building next-generation AI-native 6G and AI-RAN infrastructure on open AI-native platforms designed to support the 6G networks of the future and physical AI deployments. 

The intent is now explicit, the necessary changes in telecom networks infrastructure is beginning to reshape and restructure how connectivity will be delivered, managed and monetised. 

In the upcoming second part of TelecomTV’s exclusive interview, Simeonidou covers the most pressing problems network architects currently face in deploying AI-native networks. 

– Martyn Warwick, Editor in Chief, TelecomTV