India's Government Bets on Homegrown AI Foundations
In a move that signals a fundamental shift in India's technology policy ambitions, the Ministry of Electronics and Information Technology (MeitY) has selected four Indian AI startups to receive comprehensive support under the â¹10,000 crore IndiaAI Mission â including access to compute infrastructure at a heavily subsidised rate of just â¹67 per hour, alongside a 100% compute subsidy for foundational AI model development. The chosen companies â Sarvam AI, SoketAI, Gan AI, and Gnani AI â represent India's first coordinated attempt to build sovereign large language models that reflect the country's linguistic diversity, cultural context, and specific governance requirements.
The announcement arrives at a critical juncture. India's AI market is projected to become a $126 billion opportunity by 2030, with a potential GDP impact of $1.7 trillion by 2035. Yet until now, India's AI ecosystem has been characterised primarily by talent export â Indian engineers building AI products for American companies â and AI consumption, with Indian enterprises deploying models built in Silicon Valley. The IndiaAI Mission represents a decisive government commitment to change that equation.
Why Foundational Models Matter for India
The distinction between AI applications and foundational models is crucial for understanding why this programme matters. Applications â chatbots, content generators, code assistants â can be built on top of existing models like GPT-4o or Claude. But foundational models are the underlying language systems that determine what an AI actually understands, what biases it carries, what languages it speaks fluently, and whose cultural context it reflects.
India is home to 22 officially recognised languages, hundreds of dialects, and a cultural and religious diversity that defies simple generalisation. Existing Western foundational models, trained predominantly on English-language internet data, perform poorly on many Indian languages â particularly those written in non-Latin scripts like Tamil, Telugu, Kannada, Bangla, and Odia. A foundational AI model built with India's linguistic reality in mind would unlock AI applications for hundreds of millions of Indians who are currently underserved by existing technology.
The Four Selected Startups
Sarvam AI, arguably the most prominent of the four, has already raised $53 million in Series A funding from marquee investors including Lightspeed Venture Partners, Peak XV Partners (formerly Sequoia India), and Khosla Ventures. Founded by IIT graduates with deep AI research backgrounds, Sarvam has focused on building multilingual AI for Indian languages, with existing models covering several major Indian languages with state-of-the-art performance. The government partnership provides compute resources that would otherwise require hundreds of crore rupees in investment.
SoketAI focuses on enterprise AI infrastructure, building the tooling and deployment systems that allow Indian enterprises to run AI workloads efficiently on domestic compute. Gan AI specialises in synthetic media and generative AI, with applications in advertising, entertainment, and digital content creation. Gnani AI has built a strong position in voice AI for Indian languages, with enterprise customers in financial services, retail, and telecommunications using its conversational AI platform.
The Economics of Subsidised Compute
The â¹67 per hour GPU access rate deserves special attention. Commercial GPU compute on major cloud platforms â AWS, Google Cloud, Azure â typically costs between â¹400 and â¹1,500 per hour for the high-performance A100 and H100 chips required for large model training. At â¹67 per hour, the government is providing compute at roughly 5â15% of market cost, a subsidy that can translate into tens or hundreds of crore rupees in cost savings for a startup training a foundational model.
This matters because compute cost is the primary barrier to foundational model development. Training a competitive large language model requires thousands of GPU-hours â expenses that have historically confined serious model development to well-capitalised American companies and Chinese state enterprises. By providing subsidised access through the IndiaAI Mission's compute infrastructure (procured through partnerships with domestic and international GPU providers), the government is leveling a playing field that capital markets alone could not balance.
The IndiaAI Innovation Challenge and Broader Ecosystem
Alongside the foundational model programme, the IndiaAI Mission has launched the IndiaAI Innovation Challenge 2026, an open competition inviting startups across India to develop AI applications addressing specific national priorities â healthcare diagnostics in regional languages, agricultural advisory systems for smallholder farmers, legal aid tools for underserved communities, and financial inclusion applications for the unbanked. The challenge carries prize pools and follow-on funding for winning applications, creating a pipeline of AI-driven public goods built on the foundational models that Sarvam, SoketAI, Gan AI, and Gnani AI are developing.
The broader IndiaAI ecosystem is also booming without government intervention. India crossed 4,500 AI startups in 2026, with the sector being having raised approximately $1.48 billion in Q1 2026. Neysa, a GPU cloud infrastructure provider, raised a landmark $1.2 billion Series B â the largest AI startup round in Indian history â reflecting deep investor conviction in India's AI infrastructure buildout.
Geopolitical Dimensions of AI Sovereignty
The IndiaAI Mission also carries geopolitical significance. As the US and China compete for global AI leadership, India is positioning itself as a third pole â a democracy with a massive talent base, enormous domestic market, and the ambition to build AI infrastructure that serves Indian needs. For Indian policymakers, AI sovereignty is increasingly framed as a dimension of national security and cultural preservation, not just economic competitiveness. The mission has officially begun.
