The Intelligence Utility: India’s ₹65-per-hour Gambit

Infrastructure as Sovereignty

India is pivoting from digital regulation to aggressive infrastructure provisioning. By late August 2026, the IndiaAI Mission has expanded its national compute network to 45,000 empanelled GPUs. The centerpiece is the National Informatics Centre (NIC) facility at Shastri Park, Delhi—the government’s first dedicated AI Data Centre, delivering 1.1 Exaflops (EFLOPS) of localized power.

Subsidizing the Stack

The strategy is "The UPI of Compute": turning elite processing power into a public utility. At a subsidized rate of ₹65 per hour, the government is undercutting commercial cloud incumbents to force a democratization of Large Multimodal Model (LMM) development. This compute layer is fed by "AIKosha," a national repository now housing over 15,000 sector-specific datasets, including the high-value Indic language corpora essential for localized intelligence.

The Shastri Park Node

Shastri Park is not a standard web-hosting facility. It is a purpose-built training and inference hub. By anchoring "Sovereign AI" in physical, state-controlled nodes, India ensures that sensitive data in healthcare and governance remains within the legal and geographic jurisdiction of the state, shielding it from extraterritorial leverage.

BharatLens Deduction: The Compute-Commons Paradox

The paradox of India’s Sovereign AI is the tension between centralization and innovation. While AIKosha and Shastri Park solve the "entry barrier" problem, they create a "governance bottleneck." By aggregating 15,000 national datasets into a single state-managed silo, the mission risks trade-offs between speed and privacy. To avoid the Compute-Commons Paradox, the legal frameworks for AIKosha must be as modular as the GPUs themselves; otherwise, the sovereignty meant to protect Indian startups may inadvertently trap them in a centralized dependency that mirrors the Big Tech cartels it seeks to replace.