Airtel runs a small AI model on engineers' phones, cutting Rs 30-45 crore of cloud cost

In an August 2026 ETTelecom report on the Q1 FY27 earnings call, Airtel India's MD and CEO Shashwat Sharma said nearly 30,000 field engineers were using an in-house small language model on their endpoint devices, which he described as regular handsets. He said Airtel uses patented technology that allows inference on the device rather than the cloud, for real-time image processing to standardise workmanship.

Sharma said this was leading to a step change in the quality of installs and fault repair, and in adherence to safety measures on the ground. Executive vice chairman Gopal Vittal said Airtel was spending Rs 30-45 crore on these workloads in the cloud, and that running the model on the device had taken that cost to zero.

Vittal called it a very big breakthrough and said it could be extended to stores. Sharma said the effort is driven by a homegrown agentic platform focused on cost-efficient automation and guardrails for privacy and sovereignty.

Image checks run for every install and repair, so cloud inference costs scale with usage.

A small model can fit on an ordinary handset.

On-device inference avoids both the cloud bill and a network round trip.

Patented endpoint inference keeps data on the device.

If a task runs on every worker's phone, push inference to the phone instead of paying for it per call in the cloud.

The cost figures and the quality gains are management statements on an earnings call, reported by ETTelecom (self-reported). The article gives no independent measurement; Vittal said it could extend to Airtel's stores.