EnterpriseWeb presents “CPU-first AI for RAN at the Telco Edge”

At a recent IWPC workshop on RAN Innovation, hosted by Rogers in Montreal, EnterpriseWeb presented “CPU-first AI for RAN at the Telco Edge” (webinar replay link).

Summary

• EnterpriseWeb is an ontology-driven platform for intent-based orchestration
• The platform leverages the ontology to intelligently orchestrate AI, so AI can efficiently optimize the network
• EnterpriseWeb provides Context-as-a-Service to enrich and deterministically validate LLM/SLM inferences
• It also provides persisted structured knowledge, objects and memory to coordinate, optimize and control agent behavior
• The platform’s low-token, low-latency, resource and energy efficient AI orchestration represents a software optimization to put the ROI in GenAI
• EnterpriseWeb’s runtime implements software-based concurrency over CPUs, allowing it to close the performance gap with pricey GPUs
• CPU-first AI for RAN at the Telco Edge provides a low TCO solution that supports sustainability, without CapEx intensive rip-and-replace
• CPU-first, not CPU-only – the platform can dynamically schedule workloads between CPU and GPU where use cases justify the spend

Background
Brute force AI is an expensive and questionable endeavor. Organizations seeking to implement real-world operational use cases are running into walls. While GPUs and Foundation Models continue to evolve, token consumption is exploding with increasing use case complexity erasing any gains. “Token Maxxing” is consuming annual AI budgets in months. Telcos require trusted, consistent AI for reliable and available networks, and they care about improved business outcomes and returns on their AI investments. Telco AI challenges are compounded at the edge where resources are inherently constrained.

Prior Work
The IWPC webinar builds on EnterpriseWeb’s 2024 proof-of-concept project with Intel, Dell, Fortinet, Red Hat and Keysight CPU-first Generative AI: Enables AI-powered 5G Network Automation at the Edge, which earned industry recognition from Fierce Telecom and SiliconAngle. The PoC was benchmarked in the Intel labs.

The new demo runs between Snowflake as a central management and control plane and a set of edge nodes. The latest demo extends the previous by incorporating Agentic AI with LangGraph Agents.

Smarter, Faster, Better

Save CapEx: Leverage existing infrastructure
High-performance AI-powered edge network automation on CPUs at a much lower TCO. No need to rip and replace existing infrastructure. No specialized compute, energy or cooling requirements.

Save OpEx: Do more with less
The ontology-driven orchestrator efficiently enriches and deterministically validates LLM/SLM inferences (Low-token, low-latency, resource and energy-efficient) to maximize constrained edge resources.

Optimize Performance: AI for RAN
AI Edge Orchestrator enables continuous assurance and optimization. It dynamically orchestrates AI to reduce its overhead and improve inference quality so AI can be efficiently and effectively leveraged to optimize RAN performance at the Telco Edge.

Use Case
Compose a Secure Edge Gateway, and order and manage service via Agentic AI

Demo prep

• EnterpriseWeb is a native app in Snowflake deployed in Snowpark containers
• EnterpriseWeb is a central knowledge and control plane for edge automation, including:
  • Telecom Ontology and Network Topology

  • Catalog with VCS and Inventory with CMDB
  • RCA and NetOps

• EnterpriseWeb offers a pre-integrated solution leveraging Snowflake’s rich data, analytic and AI services
  • Zero-copy integration with Snowflake AI Data Cloud
    Integration with Snowflake Cortex

  • Integration with Snowflake Machine Learning and Graph Data Science Libraries
  • Integration with Snowflake Semantic Views

Day 0

• Login to EnterpriseWeb and declaratively compose and configure a Secure Edge Gateway from objects in a catalog
  • Linux Foundation Aether SD-RAN
  • Fortinet NGFW
  • Kamailio IMS

Day 1

• Order new Service via Agentic AI, map service to specific edge node and 5G core in inventory
  • On order, bundle service with EnterpriseWeb AI Edge Orchestrator for RCA and NetOps
  • Map EnterpriseWeb AI Edge Orchestrator to an LLM resident at edge node (Mixtral)
• Agent triggers deployment workflow by Enterprise central control plane
  • Remotely deploys the Secure Edge Gateway to the edge node
    • Connects to CaaS (Red Hat OCP AI)
    • Deploys all software packages, including EnterpriseWeb AI Edge Orchestrator
    • Integrates and configures solution elements
    • Connects EnterpriseWeb AI Edge Orchestrator to LLM for NLP and inferences
• See new service in Snowflake AI Data Cloud

Day 2

Note: Demo configured for Human-in-the-Loop to review each step, but can run autonomously

• Simulate traffic (no physical radio in demo environment)
  • Check service health via Agent / NLP
• Simulate Denial of Service attack
  • Agent detects DoS
  • Agent triggers RCA workflow by EnterpriseWeb AI Edge Orchestrator to determine root cause
  • Review determination (what and why), logs, traces
  • Approve determination
• Agent triggers NetOps workflow by EnterpriseWeb AI Edge Orchestrator to generate remediation plan
  • Review remediation plan, logs, traces
  • Approve remediation plan
  • Agent reports updated service health
  • Query Agent performance in Snowflake Semantic Views