Reflections from the TIP FYUZ Birds of a Feather Session

On a crisp Wednesday morning in Dublin, long before the main stages filled and the expo floor came alive, a group of industry practitioners, innovators, and AI enthusiasts gathered for a Birds of a Feather (BoF) session at Telecom Infra Project’s (TIP) Annual FYUZ Conference. Despite the early hour, the room buzzed with shared purpose: how do we remove the barriers that prevent telecom networks from fully embracing AI at scale?

Across the industry, operators, ISVs, OEMs and enterprises face a common set of obstacles:

Data accessibility remains fragmented, inconsistent, and locked behind siloed systems

Network and architecture complexity limits AI’s role to just an augmentation on the network & services and hinders innovation

Scaling AI depends on both technical and organizational shifts, neither of which are trivial.

These shared pain points shaped the charter for the BoF discussion. Through the TIP Data & AI Foundations initiative, members aim to address these challenges head-on, enabling a future where Agentic AI plays a key role in abstracting network and architectural complexities, and federated data and learning frameworks provide an environment to collaborate and innovate together. 

AWS Spotlight: Network Language Models (NLMs) 

One of the morning’s compelling sessions came from Amazon Web Services (AWS), which presented a visionary perspective on Network Language Models (NLMs) , specialized domain models developed to understand, reason, and act on complex network environments. AWS through the Breaking Barriers Hackathon strategically run just before the TIP’s annual FYUZ conference showcased the Agentic AI frameworks and abstractions of the  Open-RAN interfaces for developers, start ups and ISVs to build innovation solutions. This direction aligns closely with TIP’s broader ambition: to build the shared data and AI foundations that enable NLMs and agentic AI to operate securely, efficiently, and at scale across diverse operator environments.

Orange shared their perspective on what Data and AI Foundations mean for the future of telecom, emphasizing the need for federated learning environments and the need for domain models integrated with agentic capabilities. 

AT&T highlighted key security and operational requirements that must be addressed in the move toward AI-native networks, underscoring the critical role of strong governance and zero-trust security architectures. 

JOINER, the national-scale experimentation platform connecting 14 UK and Irish university and industry labs, advocated for an AI assessment and deployability framework, drawing on their experience in validating and testing AI systems in telecom environments. 

Tech Mahindra reinforced that data is the bedrock of AI and future 6G intelligence, stressing that a strong Data Foundation is essential to enabling the industry’s AI-native journey and they would be excited to be an execution partner with industry leaders. 

Telefónica helped to shape the session in particular from an open optical and IP transport networks perspective. 

In closing, AWS emphasized the need for customized Foundation Models and LLMs tailored specifically for networks, noting that today’s models are text-native while network data is inherently data-native, symbolic, and structurally complex. AWS proposed the concept of a Network Intelligence Fabric, built on Network Language Models trained on multi-operator data and integrated with a robust knowledge layer to enable a new generation of intelligent network agents. They also highlighted how AWS cloud services provide the broad set of tools required to develop, train, and operationalize these specialized models at scale.

The BoF session provided directional guidance as TIP formalizes the Charter and workstreams with industry leaders - some listed as follows:

Security must be built in from the start, not added later in the AI lifecycle.Transparency into agent decision-making requires well-defined telemetry and observability insights.Multi-agent architectures vs Centralized with trade-offs between complexity, scalability, and potential performance gains. Operational guidelines for data and AI are essential to ensure security, reliability and at-scale deployment in live networks.Research learnings should be fed back into standards, validation frameworks, and best practices

Ecosystem requirements for building NLMs include shared datasets, common abstractions, interoperability, and collaboration across operators, vendors, and academia.

A Community Moving Forward - Together

What stood out clearly in this BoF session was the spiritof collaboration. Attendees openly shared experiences, obstacles, and aspirations. There was a collective acknowledgment that no single organization can solve these systemic challenges alone. Curated data products, federated machine learning environments and community-driven efforts will be essential. As the session concluded, the group left with renewed energy to continue shaping the AI-native future of telecom,one in which networks are more programmable, adaptive and autonomic.

TIP is encouraging proactive engagement from interested parties building this community. Please get in touch through membership@telecominfraproject.com. 

Join one of our Project Groups to collaborate with Operators, Infrastructure Providers, and Integrators in conceiving new and innovative ways of building and deploying telecom network infrastructure.