Checkout the whitepaper on this topic based on my work setting up AI DC hub in Gujarat:Edge AI Data Centers Whitepaper
The whitepaper argues that India's AI infrastructure build-out requires a deliberate two-tier architecture: a small number of GW-scale Phase 1 training campuses — already under development by Reliance, Adani, L&T, and the IndiaAI Mission — for training sovereign LLMs like Sarvam and Krutrim; and a much broader network of Phase 2 distributed Edge AI Data Centers (10–50 MW each), co-located with India's abundant but often-curtailed renewable energy resources in Gujarat, Rajasthan, Tamil Nadu, and Andhra Pradesh, networked together via Jio/Airtel fiber into a federated India AI Grid. The core thesis is that inference — which will constitute the vast majority of AI compute demand by volume — is geographically relocatable and should follow the power, not the transmission line, enabling 12–24 month deployments on existing 33/132 kV distribution feeders rather than 4–6 year waits for 400 kV transmission interconnections, while simultaneously absorbing curtailed renewable generation, minimizing grid upgrade requirements, and positioning distributed edge nodes as grid-balancing assets rather than grid burdens.