GLOSSARY: India’s AI Mafia

India's AI story has moved beyond policy drafts and grand announcements.‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌

India's AI Mafia

THE BELAMY

Weekly Newsletter of AIM

Monday, Feb 23, 2026 | By Mohit Pandey

Now, subscribe to our Digital & Print Editions >

India's AI story has moved beyond policy drafts and grand announcements. It is now measured in models trained, systems deployed, and infrastructure owned. Self-reliance is now being tested in code repositories, GPU clusters, hospitals, police stations, factories, farms and call centres.

What follows is not a leaderboard, but a map of the stack. From foundation models to quantum hardware, from public health systems to industrial simulation, these 15 companies show how wide the sovereignty debate has become.

BharatGen & Sarvam sit at the top of this stack.

Led by CEO Rishi Bal and founded by Ganesh Ramakrishnan, BharatGen is an IIT Bombay-anchored consortium that secured the largest GPU allocation under the IndiaAI Mission.

Its flagship release, Param2 17B, is a 17-billion-parameter multilingual model built from scratch using an MoE architecture and trained extensively on Indian datasets. It supports 22 Indian languages and spans reasoning, mathematics, code, speech, text and vision.

Over 60 engineers across nine academic institutions are working towards a larger target: a trillion-parameter model—a philosophy that Ramakrishnan calls "Virasat bhi aur vikas bhi".

Sarvam AI offers a contrasting lesson.

Founded by Vivek Raghavan and Pratyush Kumar, Sarvam faced criticism for its earlier 24B model but has since recalibrated. At the India AI Impact Summit, it launched Sarvam 30B and Sarvam 105B, signalling a renewed push at the large-model layer while deepening its Indic focus.

It also launched Indus, a chatbot positioned to compete with ChatGPT, which has seen strong domestic reception.

Bulbul V3, its text-to-speech model, topped a blind study with over 20,000 votes across 11 languages. Sarvam Vision reported 84.3% accuracy on messy documents across 22 Indic scripts.

Besides, Sarvam has signed an MoU with the Odisha government for a sovereign AI park, partnered with Tamil Nadu, and signed an LoI with Maharashtra to build sovereign AI infrastructure across the state to create 20 MW of AI-optimised compute capacity.

Then there's Gnani.ai. Led by Ganesh Gopalan, it launched a 5-billion-parameter voice-to-voice model, Inya VoiceOS, along with Inya.ai, a no-code agentic platform. Enterprises report 4-6% higher collection rates, nearly 2x loan disbursals and 80% faster fraud detection using voice biometrics.

This followed the launch of its Vachana STT and TTS models at the Summit.

Others are pursuing sharper specialisation.

Genloop, led by CEO Ayush Gupta, is building three 2-billion-parameter models under the IndiaAI Mission: Yukti for multilingual reasoning, Varta for edge deployment in healthcare and agriculture, and Kavach as a guardrail model to secure AI inputs and outputs for Indians.

Genloop argues that Western public models are expensive and linguistically misaligned, and claims its enterprise analytics platform is being piloted by firms including Lenskart and Emeritus.

Meanwhile, Soket AI Labs, led by CEO Abhishek Upperwal, is developing Project EKA—a 120-billion-parameter open source model trained on India-centric datasets. Its roadmap scales from 1B to 120B, with air-gapped and on-device deployments planned for sensitive sectors. The company speaks openly about national security risks of relying on foreign models in defence contexts. Its COOM framework promises transparent updates and energy-efficient training.

ZenteiQ moves the conversation to factories and labs.

Founded by Prof Sashikumaar Ganesan, it builds physics-aware AI for engineering and industrial intelligence—from thermal optimisation and battery modelling, to semiconductor design and simulation acceleration. The argument is that sovereignty is not only about chatbots, but about control over AI embedded in manufacturing and energy infrastructure.

AIM Network >>

In this episode of The Power Exchange, we explore the intricate world of healthcare innovation through a special conversation with Rohit Agarwal, SVP and Head of Transformation, Innovation, and Enablement at Optum Global. Rohit redefines innovation not as a sudden "shiny object" but as a persistent journey of step-by-step changes aimed at making the health system work better for everyone.

Sovereignty for Bharat

Shodh AI pushes into science. CEO Arastu Sharma is building Project Skanda with NVIDIA, a mesoscale foundation model for energy storage trained on over 10 million simulations and experimental data. Skanda models the internal structure of battery cells.

It operates an AI-driven wet chemistry lab where systems decide, run and refine experiments in a closed loop. Partners include DRDO, DST, Indian Railways and global cloud providers. The vision is autonomous laboratories as standard practice in Indian materials discovery.

Wadhwani AI, led by Shekar Sivasubramanian, embeds AI inside public systems. Its tuberculosis screening tools span 17 states and 103 districts, with drug-resistant diagnostics in 97 laboratories. Its reading fluency platform has completed over 15 million assessments, benefiting 7.9 million children.

Kisan e-Mitra has resolved over 5 million farmer grievances. In total, it says its systems have reached 190 million individuals.

Sovereignty here means population-scale delivery.

CoRover, led by Ankush Sabharwal, builds a multilingual agentic platform powered by BharatGPT models ranging from 0.5–3 billion parameters across 12 languages. It claims deployments with IRCTC, SEBI and LIC, serving over 50,000 enterprises and one billion users.

At the summit, it announced the BharatGPT DeskAI Appliance for offline, on-premise deployment, arguing that sovereignty requires control across models, datasets and application layers.

Developers, Alert! AMD AI Engage is Here >>

Join AMD AI Engage, an AMD-led initiative for developers and AI enthusiasts. Strengthen your AI skills, participate in challenges, and unlock opportunities for advanced certification experiences with AMD and industry experts. Register now.

Full-Stack AI for India

These companies show that India's AI self-reliance has evolved into a layered buildout.

RevRag AI embeds AI agents directly into BFSI apps. CEO Ashutosh Singh says it works with PhonePe, Mahindra Finance, Motilal Oswal and others, reporting 10-15% higher conversion rates and up to 70% lower support costs.

Vahan.ai, led by CEO Madhav Krishna, applies conversational AI to labour markets. It has placed over 10 lakh people into jobs, including more than four lakh in the past year, using a voice-first recruiter for gig and blue-collar workers.

Staqu Technologies, founded by Atul Rai with Anurag Saini and Pankaj Kumar Sharma, built JARVIS, an AI video analytics platform used by over 11 state police forces. Officers can query CCTV footage using text prompts. Deployments run on customer-owned infrastructure in government data centres.

NeuroDx is building MANAS-1, a 400-million-parameter brain foundation model trained on Indian clinical EEG data. Founded by Siddharth Panwar, Dr Puneet Agarwal and Kailash Sati, it reports over 95% accuracy in detecting biomarkers linked to epilepsy and mental health disorders, with a 20-billion-parameter model for deeper signal analysis underway.

Neysa, founded by Sharad Sanghi and Anindya Das, is constructing the infrastructure layer, with $1.2 billion in commitments to deploy more than 20,000 GPUs in India. Through Velocis Cloud, it offers end-to-end AI workflows from notebooks to inference endpoints. The pitch is performance certainty and data assurance within domestic borders.

Finally, QpiAI, led by Nagendra Nagaraja, represents the AI–quantum convergence. It launched QpiAI Indus, India's first full-stack 25-qubit superconducting quantum computer. It is developing a 64-qubit processor called Kaveri and training platforms for quantum education. Sovereignty here extends beyond AI into the next computational frontier.

Together, these 15 startups are powering the next phase of India's AI mission. The harder question now is not whether India can build, but whether these systems will scale, interoperate and endure. Sovereignty is not declared. It is sustained.

Simulated Reality >>

In this Simulated Reality episode, we sit down with C5i (Course5 Intelligence) to talk about turning AI pilots from a cool lab experiment into a serious profit driver. We break down their patent-pending AI Impact Model and examine what it takes for global enterprises to move from proofs-of-concept to billion‑dollar outcomes.