IndiaAI Mission Backs 20 Home-Grown Foundation Models — 12 LLMs and 8 SLMs

India's bid to build a sovereign artificial-intelligence stack has a sharper shape. Union Electronics and IT Minister Ashwini Vaishnaw told Parliament on 22 July 2026 that the government has identified 20 indigenous AI foundation-model proposals for support under the IndiaAI Mission — 12 large language models (LLMs) and eight small language models (SLMs).

The disclosure, made in a written reply, is the clearest official accounting yet of how many home-grown models the mission is now backing, and it signals that India's strategy of funding a portfolio of models rather than betting on a single national champion is beginning to deliver.

From ambition to a portfolio

When the IndiaAI Mission was launched, the headline promise was affordable compute and a push for Indian foundation models trained on Indian languages and data. The 20 selected proposals turn that ambition into a concrete slate. The mix of 12 LLMs and eight SLMs reflects a deliberate hedge: large models for the most demanding reasoning and generation tasks, and smaller, cheaper models that can run efficiently for narrower, high-volume applications such as on-device assistants and vernacular services.

Among the projects under development are Sarvam AI's 30-billion- and 105-billion-parameter models, Gnani.ai's speech-to-speech model, BharatGen's multilingual foundation models and Avataar AI's video-generation model. The spread across text, speech and video underscores that India is not chasing a single modality but seeding capability across the frontier.

Compute is the enabler

Underpinning the model push is the mission's compute programme. According to the minister, the IndiaAI Mission has empanelled 15 compute service providers, approved 237 projects and allocated 9.3 million (93 lakh) GPU hours to developers. Subsidised access to graphics processing units — the scarce, expensive hardware that model training depends on — is the single biggest lever the government holds, and pooling it for Indian startups and researchers is what makes a 20-model portfolio feasible in the first place.

The mission is structured around seven pillars spanning model development, applications, datasets, talent development and affordable compute infrastructure, a design meant to address the full stack rather than any one bottleneck.

Why sovereignty matters here

The case for home-grown models is not merely national pride. Frontier AI increasingly shapes how citizens access services, how businesses automate and how governments deliver welfare. Models built abroad may under-serve Indian languages, misread local context, and leave critical infrastructure dependent on foreign providers and their terms. A domestic portfolio — trained on Indian data, tuned for Indian languages and hosted on Indian compute — is a bid for strategic autonomy in a technology that is becoming as foundational as telecom or power.

There are real questions the model count alone cannot answer: how these models will perform against the best global systems, whether the compute allocated is enough to train them to the frontier, and how many will find durable commercial or public-service use. Funding 20 proposals is a portfolio bet, and portfolio bets are expected to produce a range of outcomes.

What to watch next

The near-term signals worth tracking are the release and benchmarking of the flagship models — Sarvam's larger models chief among them — the uptake of the SLMs in real deployments, and whether the 9.3 million GPU-hour pool expands as the mission's compute capacity grows. For now, the 22 July update establishes the scale of India's ambition in hard numbers: 20 models, two size classes, one national push.

Sources

• The Tribune / ANI — Govt identifies 20 indigenous AI foundation model proposals under IndiaAI Mission: Parliament (22 July 2026): https://www.tribuneindia.com/news/business/govt-identifies-20-indigenous-ai-foundation-model-proposals-under-indiaai-mission-parliament/
• Business Standard — Govt to support 20 indigenous sovereign AI models under IndiaAI Mission (23 July 2026): https://www.business-standard.com/technology/tech-news/govt-to-support-20-indigenous-sovereign-ai-models-under-indiaai-mission-126072300226_1.html
• Indian Television — IndiaAI Mission backs 20 homegrown AI models, allocates 93 lakh GPU hours: https://indiantelevision.com/regulators/indiaai-mission-backs-20-homegrown-ai-models-allocates-93-lakh-gpu-hours/
• Communications Today — IndiaAI Mission backs 20 indigenous AI model proposals: https://www.communicationstoday.co.in/indiaai-mission-backs-20-indigenous-ai-model-proposals/