- Mavenir developed GenAI Copilot in collaboration with Nvidia and AWS for operating communications AI networks.
- Analysts predict that these systems will start to be deployed in operational networks within the next 12 to 18 months.
- T-Mobile is already using AWS for its initial operations in this space.
Within the next year, generative artificial intelligence (GenAI) will be introduced into operator network management.
To that end, Mavenir, a pioneer in open radio access networks (open RAN), has moved further into the telecommunications industry's world of GenAI and machine learning (ML), teaming up with AI chip giant Nvidia and cloud provider Amazon Web Services (AWS) to develop its Copilot software to manage network operations.
“Operations Copilot is intended to support our engineering and operations teams. [communication service providers] “It analyzes large volumes of disparate data collected from the network, enabling personnel to quickly diagnose faults and provide appropriate fixes when an outage occurs or the network is not performing optimally,” Bejoy Pankajakshan, EVP, chief technology and strategy officer at Mavenir, told Fierce in an email.
“Copilot is part of the NIaaS (Network Intelligence as a Service) framework with various building blocks of AI/ML workflows that can extract value from various sources of network data to identify patterns in historical data, make inferences and predictions, and provide early warnings on potential hardware failures,” he explained, adding that the software can provide insights that help enable preventive maintenance.
Pancajakshan noted that Mavenir hasn't forgotten its open RAN roots with its latest AI software: “The NIaaS framework with Copilot built-in applies AI/ML to data collected across open RAN fronthaul, midhaul and backhaul interfaces, enabling AI/ML insights into network operations and helping operations and engineering teams ensure their networks are functioning at optimal performance,” he said.
Roy Chua, principal analyst at AvidThink, said Tier 1 operators are already running GenAI pilots for network operations. He said such copilots are likely to arrive “within 12 to 18 months for the low-hanging fruit,” with more general AI network operations software likely to arrive “around late 2025 to 2026.”
“There are mobile operators that are using GenAI as part of their network operations,” Chua says. “One public example is T-Mobile, which is an AWS customer for this use case,” he notes.
As for how operators will initially use the GenAI software, the analyst said AI network operations will start simple. “From an operations perspective, it could be field operations (troubleshooting), network operations (troubleshooting, capacity planning, predictive maintenance, optimization),” Chua said. “Field operations could be looking at manuals or providing instant context about what's going wrong in the field.”
“Simply put, these are low-risk activities or functions that have little to no impact on operations,” Leonard Lee, executive analyst at NextCurve, agreed via email. “Anything that is significant needs to be proven within the context of the operator's environment,” he said.
“The main challenges are data quality and integration, which operators I speak to consistently cite as barriers to adoption,” Lee noted.
Mavenir's CTO argued that all CSPs will benefit from the Copilot software, but that “operators who are deploying open RAN will benefit more because of the richer, more detailed data available across a variety of interfaces.” He said a pre-release version of the software is available now, with a commercial release expected to come “soon.”
Analysts disagreed on whether Mavenir's Copilot is tightly tied to Nvidia and AWS. “The company's Operational Copilot framework is supposed to be cloud portable and cloud agnostic, but because it's based on Nvidia AI Enterprise, it appears to be tightly tied to the Nvidia stack,” Lee said.
“Not necessarily,” Chua countered. “The number of GPUs is increasing. [graphical processing unit] Options (considering high demand) would certainly be led by Nvidia for now, but Intel, AMD and others are also building AI accelerators. Similarly, workflow efficiency, quantization, model pruning or model size reduction (specialized models) could make it run just fine on a CPU. [central processing units]. “
