Closing the autonomous network storage gap: A live discussion

"It's always going to perform as well as the slowest part of the link."

Telcos have spent years optimizing compute and networking for cloud-native architectures. But the data and storage infrastructure underneath those systems were traditionally built for archival and regulatory compliance versus the real-time demands of autonomous operations and AI workloads.

On this week’s episode of Zero-Touch Live, Rakuten Symphony CMO Geoff Hollingworth caught up with Patrick Lopez, CEO of {Core Analysis} and a telecom infrastructure specialist focused on the data and storage architectures operators need to support autonomous operations and AI at scale.

📺 Watch the replay below.

When the pipeline is the bottleneck

Traditional network storage was designed as a waterfall with data extracted from each network function, cleaned, moved to a data lake, then a warehouse and made available for analytics. That architecture was built to answer questions about last week whereas autonomous networks must answer questions about right now.

Geoff and Patrick explored the architectural rethink that real-time autonomy and distributed AI inference require, focusing on key takeaways that included:

• Pipeline latency is an autonomy blocker. Waterfall architectures that move data through multiple stages before making it available to decision systems are structurally incompatible with autonomous network operations. Closing the gap means blurring the line between memory and storage, and building an end-to-end data infrastructure that ingests from all network functions in real time.
• AI workloads impose memory demands that telecom infrastructure was never designed to meet. Models with billions of parameters require holding vastly more data in active memory than traditional network functions. The assumption that existing storage capacity is sufficient breaks quickly once AI optimization use cases move from concept to deployment.
• Distributed inference means distributed data infrastructure. As operators look to reduce latency below what centralized cloud can offer, inference will be pushed toward central offices and SMOs. That requires distributing not just compute but the storage and data management systems that feed it, and those systems do not yet exist in most operator networks.
• Greenfield AI environments are being built inside brownfield networks. The AI inflection point is giving operators a rare opportunity to rearchitect without a full forklift. The vendors and supply chain assumptions that governed previous generations may not carry over, and operators are still in the learning phase of figuring out what does.
• Open, disaggregated architectures are the practical answer to supply chain fragility. With silicon supply concentrated in a small number of vendors across geopolitically exposed geographies, the most realistic path to resilience is architecture that allows component swapping without systemic disruption, not rebuilding the supply chain itself.

The technology exists. What has been missing is the architectural intent to treat data infrastructure with the same rigor applied to compute and networking.

👉 Watch the full interview for a deeper look at what end-to-end data infrastructure actually needs to look like as operators move toward autonomous operations and AI-native networks.

‍