India's AI Infrastructure Race Is Now Real — Not Just Policy
For years, India's AI ambitions existed primarily in policy documents and budget presentations. That has changed decisively in 2026. The IndiaAI Mission — the Union government's Rs 10,371.92 crore ($1.25 billion) initiative to build sovereign AI infrastructure — has now deployed more than 38,000 GPUs across providers including Yotta Data Services and NxtGen Cloud Technologies. Compute is available to Indian startups at Rs 115-150 per GPU-hour, roughly 42% below market rates.
The government's target is 1 lakh (100,000) GPUs by end of 2026 — an ambitious buildout that, if achieved, would put India among the top five countries globally by sovereign AI compute capacity. The mission also shortlisted 12 teams to build indigenous AI models, signalling that the government wants India not just to use AI, but to build it from the ground up.
Sarvam AI: India's First Sovereign LLM Goes Live
The most significant milestone in India's sovereign AI journey came on February 18, 2026, when Sarvam AI — the Bengaluru-based startup selected as the IndiaAI Mission's flagship LLM partner — launched two open-source models: Sarvam-30B and Sarvam-105B. The 105B model uses a mixture-of-experts (MoE) architecture with approximately 9 billion parameters activated per token and a 1,28,000-token context window, competitive with comparable-tier international models.
Sarvam AI, which had reached a $1.5 billion valuation by early 2026, also launched a startup programme offering AI credits and developer tools to Indian startups building on its models — creating a flywheel of cheap compute, Indian-language-optimised foundation models, and a growing developer ecosystem.
The Data Sovereignty Argument
Why does India need its own AI infrastructure when AWS, Azure, and Google Cloud all have India data centre regions? For government ministries, defence contractors, and regulated financial institutions, routing data through US-headquartered cloud providers creates legal complexity under India's data protection framework and SEBI/RBI regulations. A sovereign GPU cluster operated by Indian companies under Indian law eliminates that complexity entirely. The IndiaAI Mission's compute is specifically positioned for government agencies to train and deploy AI systems on sensitive datasets — population health data, agricultural productivity records, tax administration systems — that cannot be sent to foreign cloud providers.
NVIDIA's Role in India's AI Mission
India's GPU buildout has NVIDIA's fingerprints all over it. The company has supplied over 20,000 NVIDIA Blackwell Ultra GPUs for Yotta's "Shakti Cloud" sovereign AI platform. Larsen and Toubro (L&T), one of India's largest conglomerates, is building NVIDIA AI factory infrastructure in Chennai (30 megawatts) and Mumbai (40 megawatts). NVIDIA CEO Jensen Huang has personally visited India multiple times, recognising it as one of the fastest-growing AI infrastructure markets globally.
The Budget Reality: Only 40% Utilisation in FY26
Despite the momentum, a critical data point deserves attention: the IndiaAI Mission utilised only INR 800 crore of its INR 2,000 crore FY26 allocation — just 40% of the earmarked budget. For FY27, Finance Minister Nirmala Sitharaman has allocated INR 1,000 crore. The underspend reflects real challenges: procurement delays, vendor onboarding complexity, and difficulty finding enough qualified Indian AI researchers and startups to absorb the compute capacity.
What Indian Startups Should Know
If you are building an AI startup in India, the IndiaAI Mission's compute access programme is worth applying for. GPU access at Rs 115-150 per hour represents a genuine cost advantage for training custom models — and Sarvam's open-source models provide a strong starting point for applications requiring Hindi, Tamil, Telugu, Bengali, or other Indian language capabilities. For the first time, Indian startups can build and train AI models at competitive costs on Indian infrastructure with Indian foundation models. That is a meaningful shift for India's technology sovereignty, and its full implications will play out over the next five to ten years.
