India’s 12 Indigenous Foundation Models Drive AI Mission

India’s AI push is moving from rhetoric to rollout, with twelve home‑grown foundation models selected under the IndiaAI Mission and backed by a national compute pool of 38,000 GPUs [3]. The models will be open‑sourced ahead of the India AI Impact Summit in February 2026, aiming to fuel sector‑wide digital transformation.

Key takeaways

• Twelve indigenous foundation models were chosen from over 500 proposals for the IndiaAI Mission [3].
• The models will run on a national compute capacity of 38,000 GPUs, making the infrastructure publicly accessible [3].
• Open‑sourcing the models is intended to promote inclusive innovation and AI sovereignty [3].
• OpenAI CEO Sam Altman now says India should pursue a “full‑stack” AI approach, contrasting his 2023 view that competing on foundation models was “hopeless” [1].
• Power‑intensive AI data centres could strain India’s electricity grid, with Andhra Pradesh’s projected AI load potentially exceeding its total consumption [2].

Building a Home‑grown AI Stack

India’s strategy, outlined by IT minister Ashwini Vaishnaw, emphasizes designing GPU chips, training foundation models, and developing applications on top of them [1]. The twelve selected models span diverse domains, from Avataar.ai’s language‑specific AI avatars for agriculture and healthcare to IIT Bombay’s BharatGen open‑source project [3]. By making these models open‑source, the government hopes to lower entry barriers for startups and academia, fostering a broad ecosystem of AI‑driven services. Altman’s recent endorsement of a full‑stack approach underscores the shifting perception of India’s capability to host large‑scale AI workloads [1].

Infrastructure and Energy Challenges

The ambitious rollout hinges on a massive compute platform—38,000 GPUs—that will be shared across the nation [3]. However, the energy demands of AI data centres raise concerns. In Andhra Pradesh, the target of 6 GW installed capacity for AI data centres by 2030 would represent roughly 77 % of the state’s 2024 electricity consumption, and accounting for cooling overheads could push the requirement to over 9 GW continuous load [2]. Such demand exceeds current generation capacity and would require new thermal, nuclear, or large‑hydro projects, each with multi‑year gestation periods [2]. Similar power‑supply hesitations have emerged in U.S. locales, suggesting that India must address grid reliability and sustainability before the compute capacity can be fully utilized.

Why it matters

The launch of twelve indigenous foundation models signals India’s intent to achieve AI sovereignty, reducing reliance on foreign models and fostering home‑grown innovation. Open‑sourcing these models could accelerate sectoral adoption, from agriculture to public services, while the national GPU pool provides the compute backbone needed for large‑scale training. Yet, the success of this vision depends on resolving infrastructure bottlenecks, particularly electricity supply, to sustain the energy‑intensive AI workloads. As the AI Impact Summit approaches, policymakers and industry players will watch whether India can balance rapid AI expansion with the practical limits of its power grid.