AI is everywhere, we are told; it will soak up our jobs and leave us idle - perhaps as Bertrand Russell wished for all humanity in In Praise of Idleness. More than 30% of urban Indians have at least tried generative AI, yet India is still to come up with a truly novel AI model. Between them lie our progress and our unfinished business. No sector yet has a credible adoption rate - only official counts of deployment (not use or outcomes) and vendor self-reports. Every sector shows AI adoption, but mostly automation labelled as AI. That is a foundation, not a finish line.
In education, 7,90,520 CBSE students took AI in Classes IX-X in 2025-26, flat on the year, and about 50,361 in XI-XII. India ranks first on Coursera for course sign-ups, though its 89th-of-109 proficiency rank is mostly a GDP-per-capita proxy. IndiaAI fellowships selected 159 undergraduates against a target of 8,000, and AI B.Tech seats are not yet counted separately. In electronics and communications, DoT’s ASTR disconnected 820,000 SIMs by August 2025 and 880,000 by July 2026, a plateau. Operators flagged 24.4 billion spam calls and messages in April-June 2026, though complaints still rose about 40%. Two chip units are in production, both assembly-and-test; there is no wafer fab yet. AI is 3-4% of IT-services revenue, and TCS reported 270,000 “AI-proficient” staff even as it cut 23,460 jobs.
In logistics and transport, DigiYatra has logged 10 crore cumulative journeys; about 30% of passengers use it where it is live. Barrier-free tolling covers 6 of 121 plazas, with about 10% payment failures. Six machine-vision inspection units are the railways’ only system officially called AI/ML; Kavach, ULIP, GatiShakti and IRCTC’s bot-blocking are valuable, but not AI. No logistics adoption survey with a disclosed sample exists yet. Encouragingly, logistics cost is about 8% of GDP, not 13-14%.
{{/usCountry}}In logistics and transport, DigiYatra has logged 10 crore cumulative journeys; about 30% of passengers use it where it is live. Barrier-free tolling covers 6 of 121 plazas, with about 10% payment failures. Six machine-vision inspection units are the railways’ only system officially called AI/ML; Kavach, ULIP, GatiShakti and IRCTC’s bot-blocking are valuable, but not AI. No logistics adoption survey with a disclosed sample exists yet. Encouragingly, logistics cost is about 8% of GDP, not 13-14%.
{{/usCountry}}In public services, Aadhaar face authentication--the only AI system at national scale--more than doubled to 28.29 crore transactions in November 2025, from 12.04 crore a year earlier, though its failure rate is unpublished. Across 62 ministries, 762 AI use cases are identified, none yet counted as in production. SUPACE is still experimental five years on, and the CAG has yet to audit an AI system.
Less happily, fraudulent schemes have exploded, many if not all using easily available AI tools like claude.ai and ChatGPT; the streaming shows about such fraud, in which resourceful elites use clever, poor semi-rural wo/men, are rooted in reality--a sad reflection of our socio-economic and educational make-up. India has adopted AI really fast in cyber-crime, while its positive uses lag behind most of the world. That ingenuity deserves a better target.
Undoubtedly, India, especially its industries, will increasingly need AI in the near future. Sarvam AI has garnered much attention in the last two years with relatively small models accessible to those of us who use AI in our daily work; before it, a different firm laid claim to being India’s first multilingual AI model. Questions have arisen: over a mismatch between cost and performance; over reports that one of these models is really a fine-tuned version of pre-existing European models rather than built from scratch, as its team claimed; and over usage claims of up to a billion people, which some reports suggest are inflated.
All this points to good news: We do have the expertise to understand and rebuild, through fine-tuning and retraining, models made in the West or China, such as Qwen, DeepSeek and Kimi. Each has its edge: DeepSeek, radical training-cost efficiency; Qwen, excellence in two dozen Indian languages and OCR based on them; Kimi, autonomous, agentic task execution.
What Indian scholars and practitioners can offer over and above what already exists is not yet clear. Rebuilding is a start, making is the goal.
We are not out of ideas; the difference lies in our approach. We have tried only top-down: committees and sub-committees decide, and youngsters carry out instructions. The most successful models were made the opposite way. Youngsters trying to solve problems led all the western and Chinese models mentioned here.
The West, and now increasingly China, let young imaginations run wild: Young scientists, techies and, indeed, nerds are spotted early, nurtured and honed--human sides included--into well-balanced wo/men who lead technological revolutions, with genuine resources for their startups, a transparent administrative structure and continued government assistance. An IIT-JEE or NEET examination does not do this for us.
Can we do this in India? Absolutely, if we take risks through our young wo/men and transfer the weight of responsibility onto them. They have done us proud, haven’t they?
We already have schemes to hand-hold startups; let us simplify their compliance burden, within the legal framework. Let us look for multi-disciplinary wo/men instead of choosing by subject, as even institutions of inter/national repute still do. Let us nurture young talent as the West and, indeed, China do; there is no harm learning from those consistently more successful than us.
India has shown it can adopt AI. Its young can make it lead.
(The views expressed are personal)
This article is authored by Siddharth Savyasachi Malu, director, Sourak Nath, graduate trainee (engineer) and Deb Narayan Barik, data scientist, AI Centre of Excellence, Shiv Nadar University, Delhi-NCR.