Time really does fly when you’re reshaping the backbone of global connectivity. Looking back at the last decade of network transformation, we’ve moved from the “elephant in the room”, the sometimes-misunderstood shift of NFV, to a world where disaggregated, cloud-native software is the baseline for scalability and resilience.

But as I read the recent Project Sylva whitepaper on the AI ecosystem, it’s clear we are entering a new phase of “structural reinvention.” We are no longer just “virtualizing” functions. We are re-architecting the infrastructure to be AI-native.

I’ve said before that the “AI marketing fluff” needs to die down so we can focus on the 3–4 use cases that actually matter for telecom. The Sylva whitepaper hits the nail on the head by focusing on the industrialization of the telco cloud for AI workloads.

Here are my three big takeaways from the Sylva perspective, viewed through the lens of what we’ve been building at LF Networking :

• Interoperability is Infrastructure: In the past, we treated interoperability as a feature. Now, it’s the bedrock. Whether it’s CNTi providing best practices for consistent telco cloud-native network functions or Sylva creating a unified cloud stack, the goal is the same: avoiding a fragmented ecosystem that slows down AI deployment.
• From Automation to “Agentic AI”: We’ve started with closed-loop automation in projects like ONAP and Nephio. But the industry is shifting toward Agentic AI – systems that don’t just recommend actions but take them autonomously to maintain the network. This brings us closer to the “Autonomous Networking” vision the Linux Foundation Networking projects have been working toward.
• Security and Sovereignty: As we move AI from experimentation to production, we must address the risks. Our new project, Salus, is designed to provide the guardrails we need, ensuring data privacy and validating AI responses to prevent “hallucinations” in critical network operations.

The next decade will be defined by synergy. Just as we saw Nephio bridge cloud-native intent with telecom requirements, we are now seeing the fusion of CAMARA Project network APIs with the Model Context Protocol (MCP) to create “network-aware” AI.

We are moving from a phase of “wide-eyed observation” to a phase of proof. Open source remains the neutral platform where this innovation happens, ensuring that as the landscape changes, we stay compatible, secure, and—most importantly—driven by community-led standards.

What’s your take? Are we ready for the shift from “AI-supported” to “AI-native” infrastructure, or are there still too many “elephants” in the room?