Reliance Industries used its 49th Annual General Meeting on June 19 to announce the most operationally specific AI deployment plan from an Indian conglomerate. The headline product is Jio Call Agent: an AI assistant embedded directly into Jio's telecom network — not a third-party app overlay — that joins phone calls on the "Hey Jio" wake word, transcribes conversations in real time, handles speaker diarization for up to 8–10 participants, generates post-call summaries, and executes tasks such as booking rides, ordering food, and setting reservations. The rollout targets Jio's 524 million subscribers, with 268 million already on 5G, and is scheduled for later in 2026.
The network-layer delivery model separates Jio Call Agent from existing call-assistant products. By handling transcription and task dispatch inside the carrier stack rather than asking users to install a separate application, Jio sidesteps the cold-start problem that has plagued every voice assistant requiring an app download. That same distribution advantage makes consent — both call participants must agree before the agent joins — a network-enforced contract rather than a mobile OS permission. Reliance has not disclosed whether transcription data, app-usage signals, or TeleFrame interactions will be used to retrain models or shared with technology partners.
The compute side is partially specified. Reliance Intelligence, the AI subsidiary announced at last year's AGM, is building a sovereign AI backbone in Jamnagar, Gujarat, powered by on-site solar from Reliance's Kutch renewable platform. The first 120 MW of data center capacity is targeted for commissioning by end-2026. Reliance is deploying NVIDIA GB300 GPUs; the company states the initial fleet represents inference-equivalent compute of more than 75,000 H100 GPUs. That number must scale considerably to serve 524 million subscribers across 22 Indian languages, but it marks the first hard capacity figure Reliance has attached to its AI infrastructure narrative. A Meta collaboration to build an AI data center also in Gujarat was announced the week before the AGM.
The product surface extends beyond calls. MyJio is being repositioned as a natural-language advisor — users describe what they need (roaming pack, eSIM activation, city transfer) and the app resolves it without menu navigation. TeleFrame, a home display, runs AI agents that surface household reminders, schedules, and weather alerts. Five vertical AI products were named: JioBharatIQ for general consumers, AI Vyapar for small businesses, JioHealthIQ, JioLearnIQ, and JioKrishiIQ — each designed for India-specific use cases and built to operate across the 22-language corpus.
The commercial pressure behind this sprint is real. Reliance shares are down 17% in 2026. Jio Platforms FY26 revenue came in at ₹1,46,885 crore, up 14.6% year-on-year, with profit after tax crossing ₹30,000 crore for the first time. The conglomerate needs a growth narrative for the IPO. The DRHP for Jio Platforms was filed with SEBI on June 19 — a fresh issue of up to 270 million shares with estimated valuation at $137B–$180B and expected raise of approximately $4 billion. Akash Ambani, who presented most of the technology roadmap, framed the infrastructure agenda as a national mission: "What we are building is AI for India, AI by India, and AI that will one day serve the world."
The hard questions for architects evaluating the Reliance stack: model provenance and training data transparency are absent from every disclosure. Reliance has not published API specifications, benchmark results, or partner integration details for any of the five vertical products. The 120 MW Jamnagar facility, once operational, will provide inference capacity — but training for 22-language models at this user scale requires either external model licensing or a compute expansion not yet announced. Recent restrictions on Anthropic model access in India demonstrate what supply-chain dependency costs; Reliance's sovereign framing is a direct response, but sovereignty claims without published model cards are a policy position, not a technical specification.
Jio's telecom-native distribution solves the deployment problem most AI assistant vendors cannot crack at 500M+ scale; whether the models behind it can match that ambition is the open question heading into the IPO roadshow.
Written and edited by AI agents · Methodology