Terrill Dicki
Feb 18, 2026 01:10
India partners with NVIDIA to build sovereign AI infrastructure with 20,000+ Blackwell Ultra GPUs, targeting $27.7B market by 2032 under IndiaAI Mission.
India just made its biggest bet yet on AI sovereignty. At the AI Impact Summit in New Delhi, the country unveiled partnerships with NVIDIA to deploy over 20,000 Blackwell Ultra GPUs across multiple data centers—the hardware backbone for what officials are calling the IndiaAI Mission.
The $1 billion government initiative, approved in March 2024, aims to transform India from an AI consumer into a producer. With domestic AI market projections ranging from $27.7 billion to $131 billion by 2032 depending on the estimate, the stakes couldn’t be higher.
The Hardware Play
Three cloud providers are leading the infrastructure buildout. Yotta is constructing what it calls Shakti Cloud, powered by those 20,000-plus Blackwell Ultra GPUs across facilities in Navi Mumbai and Greater Noida. E2E Networks is deploying NVIDIA HGX B200 systems at L&T’s Vyoma Data Center in Chennai.
Perhaps more significant for long-term strategy: Netweb Technologies is manufacturing NVIDIA GB200 NVL4 platforms domestically under the “Make in India” program. Each system packs four Blackwell GPUs and two Grace CPUs—serious horsepower for model training and inference, built on Indian soil.
Why Sovereign AI Matters Here
India recognizes 22 official languages. Its census has recorded over 1,500 more. Building AI that actually serves 1.4 billion people means training models on local data, in local languages, on local infrastructure.
The model development already underway is substantial. BharatGen, a government-backed initiative, has built a 17-billion-parameter mixture-of-experts model from scratch using NVIDIA’s NeMo framework. Sarvam.ai is open-sourcing its Sarvam-3 series trained across 22 Indic languages with models ranging from 3 billion to 100 billion parameters.
Gnani.ai claims a 15x reduction in inference costs after fine-tuning NVIDIA’s speech models for Indic languages—enabling the company to handle over 10 million calls daily for telecom and banking clients.
Production Deployments Already Live
This isn’t vaporware. CoRover.ai has deployed multilingual speech AI for Indian Railways, supporting 10,000 concurrent users and processing 5,000 daily ticket bookings. The National Payments Corporation of India is testing FiMi, a financial model built on Nemotron, to power multilingual customer service across the banking system.
Tech Mahindra is targeting education—building an 8-billion-parameter model to translate classroom materials into Hindi, Maithili, Dogri, and other regional languages.
The Funding Pipeline
NVIDIA is partnering with Peak XV, Elevation Capital, Nexus Venture Partners, and Accel India to fund AI startups building for both domestic and international markets. Over 4,000 Indian AI startups are already enrolled in NVIDIA’s Inception program.
The Anusandhan National Research Foundation will receive complimentary access to NVIDIA AI Enterprise software and technical mentorship, with bootcamps and hackathons planned to develop talent.
India led AI adoption across Asia Pacific in 2024, with 93% of students and 83% of employees actively using generative AI according to Deloitte. The infrastructure announced this week suggests the country intends to move from adoption to production. Whether that transition succeeds will depend largely on whether these GPU clusters can actually train competitive models—something the next 12 to 18 months should reveal.
Image source: Shutterstock
