Mavenir and Sanas Embed Native Voice AI Directly into Telecom Core Networks

Telecom software provider Mavenir and speech intelligence startup Sanas have launched an integrated Voice AI platform designed to operate natively within mobile network operators' core infrastructure. Rather than routing conversational audio to external cloud providers or handling processing on user devices, Mavenir has embedded Sanas's real-time voice and language models directly into its cloud-native telecommunications stack. The deployment architecture targets in-call capabilities including live speech translation, agentic assistance, automated fraud detection, and speech accent and clarity adaptation across live carrier streams. The initiative addresses a critical architectural bottleneck in modern conversational AI: latency and compute placement. Standard voice AI pipelines typically capture audio at the client application, stream it over public networks to third-party APIs for transcription and inference, and return synthesized audio back downstream. This architecture introduces variable network jitter and multi-hundred-millisecond latencies that disrupt natural conversational cadence. By embedding AI models natively within carrier core networks, voice workloads can process in-stream audio with minimal latency, operating concurrently on high-throughput telecommunications traffic. This shift reflects a broader convergence between cloud-native telecom infrastructure and real-time generative AI. As conversational AI transitions from turn-based text interactions to real-time, multimodal voice agents, enterprise demand has shifted toward infrastructure capable of supporting sub-second, bidirectional speech processing. While major public clouds have expanded their managed voice and agent runtimes, telecommunications providers retain control over raw voice traffic and edge data paths. Integrating speech models into carrier-grade network functions allows operators to reclaim high-margin value-added services rather than serving as passive data conduits. For DevOps, cloud, and voice engineers, network-embedded voice AI introduces both architectural opportunities and operational trade-offs. On the operational side, deploying language and voice models inside on-premises and private cloud-native carrier cores demands rigorous model optimization, low-latency containerized inference runtimes, and specialized hardware acceleration at scale. Teams developing voice agents should monitor carrier API integration points, as native in-call intelligence could streamline contact center routing, simplify compliance by keeping audio streams within regulated carrier boundaries, and reduce reliance on client-heavy voice SDKs.