Login
Register
Intent-driven architectures and implementations underpin Jio's ambition to enable large-scale AI infrastructure and services, as Dr. Sudhir Mittal, Executive Vice President & Chief Architect, Jio Platforms, explains in an interview with TM Forum's Inform.
Dawn Bushaus
Jio collaborates to enable AI at scale
In the first quarter of this year Jio’s subscriber base grew by just over 6 million to reach 488.2 million. Average revenue per user also rose, despite recent consumer price hikes, to reach ₹206.2 (US$2.42). Now, the company has its eye on the next stage of growth.
Reliance Industries Limited reported during a presentation of the group’s 2024/25 annual results that 5G networks carry 45% of Jio’s total wireless traffic. With per capita consumer data consumption significantly higher on 5G networks, Jio expects this percentage to grow and provide a base for monetization.
A key focus for the company is to enable large-scale AI infrastructure and services that will add an intelligence layer to all Jio services.
Dr. Sudhir Mittal, Executive Vice President & Chief Architect, Jio Platforms, plays an important role in ensuring the company can introduce, scale and monetize services efficiently. In an interview with Inform, he explained how the company is collaborating with TM Forum members to develop architectures that have AI and data at their center.
To illustrate just how much data Jio is already handling Mittal cites the example of India playing in the final of the ICC cricket championship: “On that single day, 500 petabytes of data was consumed in the mobile application.”
“Data is turning out to be a new oil,” he says. “[There is] a lot of knowledge, which can be utilized by various technical teams, business teams, marketing teams.” But its use must be architected securely and efficiently.
“The modernization of our overall architecture is happening perpetually – every day we are improving on our architectural constructs,” says Mittal. “What we have brought to the core of our strategy is the AI implementation of componentization and intent-based management.”
Mittal describes AI as being “like an untamed animal. Everybody is really going this way and that way. There are a lot of hallucinated outputs,” he explains. “Even with structured data in the back, you ask the same question twice, thrice, N number of times, and you get different answers. So, you need to standardize the framework and define the guardrails and how different things need to be done.”
AI implementation is a major focus across TM Forum’s three missions (Composable IT & Ecosystems, Data & AI, and Autonomous Networks) and in the Innovation Hub, a physical lab in Mumbai, India, which is hosted by Jio. Going forward, much of the collaborative work happening in the Hub will center on making the Open Digital Architecture (ODA) AI-native, which includes developing an intent-based architecture and framework to enable AI models to communicate as CSPs move toward autonomous networks and adopt agentic AI, according to Mittal.
“We have brought AI implementation to the core of [TM Forum’s] strategy,” he says. “So, what we are looking for are major thought processes around componentization and intent-based implementation. These are the core of the strategy now… AI and data are at the center of the other two themes, and everything is getting integrated through the intent-driven architectures and implementations.”
Some of this collaborative work focuses on bringing AI into the ODA Canvas. For example, TM Forum’s AI Virtual Assistant (AIVA) is planned to become an ODA Component in the Canvas. “In addition, every other Component is also supposed to become AI-native,” Mittal says.
“So, a lot of futuristic architecture is being planned, which gels with what leading operators are planning in their architectures,” he adds, pointing to the participation of operators including Deutsche Telecom, Orange, Telenor and Vodafone in the Innovation Hub.
Specifically, Mittal sees a need for business architecture integration to allow different large language models (LLMs) to interact. “There are some technologies we are working on for implementing AI-to-AI kind of interactions,” he says, adding that it’s important to continue work on TM Forum’s intent API (TMF921).
In addition, Mittal believes that framework standardization efforts need to extend to agentic AI, and he says that TM Forum is in a position to lead this work.
“One of the big problems with large organizations is how do they adopt GenAI? Because the moment you are on any commercial model and as you proliferate across the organization for different use cases, the bill – the cost – goes really out of the roof,” Mittal explains. “Now, the thought process is that we should have very, very specialized agents that can work on smaller LLM models, which are more open source, and train them with the organization’s data and behavior.”
With training and retraining, these agents could become quite powerful so that operators wouldn’t need a huge number of graphics processing units. “We need to come up with that kind of agentic framework in which it can scale and still remain cost effective, as well as have a lower carbon imprint,” says Mittal.
