Rs 10K crore, 38K GPUs, and 12 firms later, India’s AI mission is still looking for users

- The India AI mission picked 12 companies to build indigenous AI products for social welfare - Two years in, most of those projects have been repurposed for banks, telcos, and enterprises - The government has no procurement framework or cross-ministry coordination to buy AI products - The mission is entering a new phase, with more companies aspiring to be part of the next cohort Enter your email address to receive a daily summary of all our stories. Whenever a new sector takes off, India responds with a mission and about Rs 10,000 crore in funding. For AI, that meant picking 12 companies to build indigenous products for government work and public welfare. That was step one. Step two has been much slower. M worked on Project Indus, Tech Mahindra’s open-source large language model (LLM) designed to support Hindi and its dialects. He remembers meeting officials from the Department of Agriculture to explain the model’s farm-related technical capabilities—latency, dialect variations, cost, and accuracy. “But most of the questions that followed were in a very different vein,” he said. “One of the officials asked me who would take over if the model suddenly glitched or stopped answering in the middle of a session.” These weren’t questions he had expected in a tech-explainer session. “I remember going back to our team and saying this wasn’t a model evaluation,” he said. “They were trying to assess whether the AI could fit into their way of functioning. That is a very different exercise.” There was no fixed standard M could try to meet, or KPIs he could fulfil. This wasn’t how government contracts usually worked, he said. It all began in March 2024, when the world was still discovering new things about ChatGPT everyday. India’s handpicked AI companies would help farmers instead of being a random chatbot, the government said. Improve education, assist doctors, even preserve Indic languages. The shiny new tech would be of the government, for the people, by the startups. Two years later, most of the companies have their products ready. But selling them to the same government that had commissioned them is proving to be a challenge. A version of project Indus has now been optimised as IndusQ for Qualcomm’s AI hub. It runs directly on devices using Qualcomm’s chips, like cars, sensors, and cameras, that are all part of its smart city plan based on Tech Mahindra is not alone in its pivot. Credits Rs 10K crore, 38K GPUs, and 12 firms later, India’s AI mission is still looking for users Know someone who'd appreciate this read? Share this story with them.