Building sovereign AI capability requires both compute and governance. Evaluate India's approach in the light of the IndiaAI initiative.

Sovereign AI means a nation can train, host and regulate AI models on its own terms. The IndiaAI Mission was approved in March 2024 with ₹10,371.92 crore over five years [1]. It pairs a compute pillar with a "Safe and Trusted AI" pillar. The design is sound, but delivery is uneven.

Compute: Strengths

• Beyond target: shared capacity crossed 45,000 GPUs by June 2026, against an initial 10,000 [1][2].
• Democratised access: subsidised GPU hours through 15 empanelled providers; 237 projects supported [3].
• Homegrown models: Indian-language LLMs and IndiaAI's high-performance compute access featured at ESTIC 2026 [4][5].

Compute: Gaps

• Import dependence: GPUs are still globally sourced. The government's own ~1.1 EFLOPS system at NIC is only at purchase-order stage [3].
• Thin private R&D: industry funded just 36.4% of GERD in 2020-21 [6]. The ₹1 lakh crore RDI Fund names AI as a sunrise domain [7] but has yet to prove disbursal.

Governance: Strengths

• India AI Governance Guidelines (Nov 2025): seven sutras, six policy pillars and a phased action plan. The approach is pro-innovation [8].
• Institutions: AI Governance Group, Technology & Policy Expert Committee and the IndiaAI Safety Institute for standards and testing [9].
• Targeted rules against AI-generated deepfakes [10]. Safe-deployment frameworks for agentic AI were on ESTIC 2026's agenda [4].

Governance: Gaps

• Soft law: these are guidelines, not a statute. Enforcement rests with sectoral regulators, so accountability may be fragmented [8][9].
• The Safety Institute's testing and benchmarking role depends on resourcing still being built [9].

Compute (GPUs, Indic models) ──┐
                               ├──► Sovereign AI
Governance (sutras, AISI) ─────┘
Gaps: own chips · private R&D · binding law

Fig: Two pillars, three gaps

Verdict: India's approach is right in direction and quick on access, but sovereignty is only partial. India rents compute it does not make and regulates through guidance rather than law.

IndiaAI has turned scarce compute into shared public infrastructure and set a trust-based rulebook. The next steps are:

• link compute expansion to semiconductor initiatives for domestic chips [3];
• bring in private R&D through the RDI Fund;
• strengthen the Safety Institute's mandate over time.

These steps would make AI self-reliance a real pillar of Atmanirbhar Bharat and Viksit Bharat 2047.