Mobile operators are moving quickly to explore how to add native on-network AI capabilities to voice calls, attracted by the ability to provide features such as in-call agentic assistance, live translation, fraud detection and enhanced audio quality.
Sharath Keshava Narayana, CEO and Co-Founder, Sanas.
Sharath Keshava Narayana, CEO of five year old voice AI start-up Sanas, said he has spoken to over 25 MNOs over the past three months, and found that operator excitement is “very real” around the opportunity.
Sanas has today announced a partnership with Mavenir to offer native voice AI services to telecoms operators. Mavenir has integrated Sanas’s voice and language models as part of its Voice AI software offer, offering AI voice capabilities natively within its on-prem cloud-native offer.
Sanas formed in 2021 and raised a Series B round of $65 million in 2025, and already has exposure with over 200 enterprises in the enterprise and call centre space. Such speed of growth is not often associated with telco-land, fairly or unfairly. But Narayana said that he has been “pleasantly surprised” by the reaction within telco-land to the voice AI opportunity, with pilots slated at over half of those 25+ operators.
“Thanks to Mavenir we are engaging directly with the C-suite. We’re getting approvals to start a pilot very quickly and for some of the largest operators, the pace at which they’ve moved has honestly surprised me. I think I think more than half of them have either given us a go-ahead to start a pilot, or have already started a pilot. And that, for me, for everything I’ve heard about telco in the last five years – that you’ll be stuck in five-year sales cycle – in 90 days getting something going has been a pleasant surprise.”
A question of scale
Pardeep Kohli, CEO, Mavenir.
Mavenir’s CEO Pardeep Kohli points out that what is required in the telco space, versus enterprise voice AI, is scale.
“AI has been in the enterprise space and also in the contact centre, that has been going on for a little while, but nobody has been able to take it and scale it to millions of users, real-time, in live conversations. We’ve seen T-Mobile announce its live translation service, but there’s a lot of other aspects you can add. So this is where the opportunity is, and we’re definitely seeing other operators also wanting to do it.”
For Kohli, the Sanas-Mavenir partnership is the answer to that scale challenge.
“We already have a lot of integration points in the network, for 911, legal intercept, billing and charging. Voice AI becomes like a bump in the wire. You need trained models, but nothing changes in all those other things around it. Working in contact centres means Sanas already has experience of dealing with many different accents and different data environments, so they already have the trained models which can easily be taken from that environment and put into the telecom environment. They have been doing it for you know 400-500 people in a contact centre. We’re going to take that model and work with them to scale it to millions of users. But as far as training a model and using the data sets, those already exist.”
For Narayana, the issue was not so much scaling up AI parameters as understanding the telco environment itself.
“For the last three months we’ve been working in lockstep with Mavenir just to be telco ready. For us, as long as the model can work on one GPU, it’s the same model that works on 100 GPUs. But I think it was more about all the interconnecting issues, just to understand how data packets come and go, understanding redundancy and concurrency. That’s what both teams have worked together for the last 90 days, and today we feel ready that we are on the Mavenir platform. Today, we can switch on a telco operator at scale in an instant.”
A sovereign model
When T-Mobile launched its Live Translation service earlier this year, it described the service as using language learning models embedded into its core to translate voice into one of 80 different languages anywhere in the world.
T-Mo’s German majority owner Deutsche Telekom also launched an agentic voice service this year, allowing users to say “Hey Magenta” to open up a series of services, including live translation. While Deutsche Telekom said it had built its service using Eleven Labs and Radisys.
While T-Mobile’s integration was on-net, Narayana says that the Mavenir-Sanas integration has an advantage over deployments such as DT’s with Eleven Labs as it allows telcos to deploy on-premises in a sovereign cloud environment.
“I think the biggest advantage that Sanas and Mavenir will bring to the table is that every operator today wants a sovereign model. They want the models to be run within their infrastructure.
“Today, their choice is you can go to OpenAI, you can go to Eleven Labs, but then you’ll have to go to their cloud. I think for an operator, if you tell them, ‘I’ll give you the speech model, but you have to send all the data to my cloud,’ it defeats the whole sovereignty principle. With Mavenir already being the core network provider for a lot of these large telco operators, us now providing the speech models that can be housed within the Mavenir core and IMS provides them the on-premise infrastructure that they never had access to.”
While T-Mobile has not released details of its technology partners, it seems highly likely that Mavenir, as its voice core and vIMS provider, is involved in the integration. If true, that should give it good experience to replicate T-Mobile’s headline grabbing launch with other customers.
It’s not just language – it’s the end-to-end summarisation to AI calling, enabled as an AI assistant, to accent, to all the security services. So we can be a very strategic value add to any operator today.
Market opportunity and challenges
So is telco likely to become the next big platform for voice AI? Sanas certainly sees the opportunity. The company acquired Tomato.AI earlier in 2026, and made Tomato’s Ofer Ronen, himself ex-Google, head of its telecom efforts.
Narayana states that speech has become mainstream in the last 12 months -“voice is the new keyboard” is his soundbite – and this is something that telcos can now tap into.
“I think telco as an industry has the largest opportunity because most of the telco operators – where voice and speech natively sits – missed out on the speech wave in the last 12 months. But they do have all the users, they do all the speech traffic.”
Well, one might rather obviously point out that telcos do not, in fact, do all the speech traffic. Yet despite the rise and rise of apps and platforms such as WhatsApp and Teams, for Mavenir’s Kohli, telco voice still has primacy over OTT apps in business and in legally mandated areas such as 911 calls, and other situations where users might not have a direct link over a company WhatsApp voice channel, of example.
For Narayana, Voice AI can give telcos “an answer” to WhatsApp and other OTT providers. And he throws in some surprising stats to back up his point.
“One of the telco operators told me that a year ago the percentage of voice sessions over their network was 3% OTT, and today it’s 17% . And very soon there’ll be no difference between the telco who charges 30-40 bucks a month and somebody else who charges eight bucks a month. The only way for the telco to differentiate is to add speech services on top of what it provides to maintain value, and that’s the opportunity where we can go in and provide a telco operator an entire end-to-end speech stack. It’s not just language – it’s the end-to-end summarisation to AI calling, enabled as an AI assistant, to accent, to all the security services. So we can be a very strategic value add to any operator today.”
Expanding on these points, this well-read LinkedIn post from Sanas’ Ronen also lays out clearly where he sees the Voice AI opportunity for telcos.
Analyst Dean Bubley of Disruptive Analysis, visible in the comments on that page, told TMN that there is certainly a play for telcos in the space.
Reacting to the Sanas-Mavenir announcement he said, “It is really good to see the telecoms industry pay attention not just to the roles of AI in voice communications, but also the quality and control of acoustics for better speech. Better clarity – and intelligent adaptation of accents and language – can add significant value to voice services.”
Bubley said that adding AI into the call and signalling paths can change the central, unchanged, calling paradigm into ‘voice’ services that add value to their customers.
“Telcos have never done ‘voice’. They have just done phone calls, a specific service which has evolved little over the last 140 years since Alexander Graham Bell. But adding AI into both the call path and the signalling helps bring telephony up-to-date, and enables operators to offer much greater utility than simple A-calls-B, by adding security, context, translation, transcription and more. I see AI voice as a central plank in operators’ emerging AI applications efforts and revenue opportunities”
However, Bubley also cautions that “there are still many outstanding questions around consent, liability and compliance with rapidly-changing AI content and data rules”.
Mavenir’s Kohli said that while achieving explicit consent from users by playing an announcement or similar does “mess up” the user experience, it will be a price users are willing to pay in order to access anti-fraud services. And he points out that DT is looking at making consent part of its T&Cs on certain packages, which would limit the in-call consent process at least for on-net calls where both parties are DT customers.
Ecosystem evolving
Whatever the opportunities and challenges, it’s certainly an area attracting a great deal of interest and investment, and Mavenir and Sanas are far from alone in targeting the opportunity. German containerised IMS company ng-Voice has partnered with Radisys, while Tallence describes its Thor Voice AI platform as an intelligent service layer for telcos to control and deploy Voice AI models. Meanwhile the big NEPs won’t give up the ground easily. Ericsson Ventures has invested in companies Hiya and Cartesia, and says it is integrating AI capabilities into IMS voice calling – listing the familiar use cases of spam and fraud detection, quality improvement, and live translation.
Comments
0