India is advancing its AI ambitions with Avataar.Ai’s indigenous video foundation model, Varya, under the IndiaAI Mission, signaling a push toward sovereign AI capability.

The discussion sits against a backdrop of scrutiny over US access to Anthropic models, highlighting global tensions around AI infrastructure and access.

A core issue is the compute gap: India relies on foreign infrastructure, fueling calls for domestic compute, data, and models to reduce dependence on US and other foreign tech giants.

Training the model involved roughly 100 billion tokens from a curated 1 trillion-token corpus, with pre-training costs around $15,000 and full training costs projected near $100,000, offering efficiency gains over from-scratch approaches.

The project demonstrates resource efficiency by training on a single 8-GPU node, showcasing innovative optimization rather than the usual large-scale GPU farms.

Taxation on technology equipment is cited as a factor raising startup costs and dampening productivity.

Sustainability is central to expanding data center capacity, emphasizing power, water use, and the shift toward independent and renewable energy sources.

AI is just the tip of the pyramid; many enabling technologies exist but require time, talent, and investment rather than just big budgets.

A reusable pipeline—training infrastructure, orchestration, data curriculum, India-focused tokenizer, and expert-divergence scheduling—constitutes the core achievement, not the single model.

Equally important is building engineering infrastructure to enable sustainable, repeatable large-model training.

Investors in early discussions include Peak XV Partners, Accel, and family offices linked to large industrial groups, with strategic backing from Aditya Birla Group.

Aditya Birla Group and other backers are weighing in, signaling strategic interest alongside Peak XV and Accel.