The expanding role of AI in critical network infrastructure

When I talk with CSPs and industry leaders about AI, the conversation often starts with hype. But hype doesn’t transform networks—clear, targeted applications do. In my recent conversation with Contributing Editor Annie Turner at FutureNet World, I shared why we at IBM view AI not as a magic bullet, but as a powerful set of tools that must be applied strategically to drive automation, observability, and ultimately, autonomy.

Computing and networking have consistently oscillated between centralization and decentralization. Public cloud brought us into a centralized era, but I believe that we will see growing decentralization, especially when it comes to AI inference.

Why? Because every minute networks generate massive amounts of telemetry and metadata. This data is the operators’ crown jewel, governed by strict sovereignty rules that make it unsuitable for public foundation models. Running smaller, targeted AI models at the edge lets us keep sensitive data under local control while delivering faster, context-rich decisions.

It’s not just about sovereignty, it’s about scalability. Breaking large problems into smaller, parallelizable components means that operators can process tens of thousands of daily events without overwhelming centralized systems.

Automation works only if we can trust the situational awareness it’s based on. In today’s hybrid, multivendor networks, visibility gaps are common and dangerous. That’s why I stress that observability must be baked into AI architecture from the start. Without it, autonomous systems are stuck. At best, they deliver the same signal-to-noise ratio as pre-AI, and at worst, they either fail or become prohibitively expensive to reengineer.

This undertaking is a balancing act. Move too slowly and you miss innovation windows, move too fast without understanding fit and you end up with costly technical debt. In the case of large language models (LLMs), observability is a gap.  And it is the foundation that will make AI systems stick.

I like to think of AI in terms of the hemispheres of the brain. Large language models are the right brain: brilliant at creativity—developing hypothesis, drafting remediation steps, suggesting operational actions—but poor at handling time. LLMs simply don’t understand network telemetry or time-series data.

That’s where IBM’s left brain comes in: IBM® Granite® Time Series Foundation Models (TSFMs). Purpose-built for network data, they deliver analytical precision and a high signal-to-noise ratio. Pairing TSFMs with LLMs allows us to generate creative, actionable plans rooted in solid analytics. When right brain meets left brain, the result is AI that operators can trust.

AI’s role in telecom extends beyond keeping the lights on. Done right, it becomes a competitive advantage that leads to primarily fewer outages but also shorter outages with faster troubleshooting. This approach is the start, not the end. You can build on observability and autonomy to achieve optimized resource allocation and even more intelligent design with the potential for new services.

And then there’s quantum. It’s not just a buzzword in this context. Quantum promises leaps in processing capability that will supercharge AI’s potential. At IBM, we’re already exploring this potential through tools like Qiskit Code Assistant, which uses AI to accelerate the creation of quantum algorithms. In the long run, AI and quantum will amplify each other in profound ways.

The journey to fully autonomous networks won’t happen overnight. It will be shaped by observability, edge inference, hybrid cloud agility and the combination of analytical and generative AI. CSPs will need to invest in the right tools and architecture that ensure adaptability and sovereignty.

For us at IBM, the trajectory is clear: from LLMs to TSFMs, and ultimately toward quantum-enabled optimization. That’s how we intend to push the boundaries of what’s possible in intelligent networks.

Learn more about network management solutions