Open-weight AI models are reshaping considerations of data sovereignty and vendor dependence for developing economies. Critically evaluate India's institutional response through the IndiaAI Mission.

Open-weight models — whose trained parameters can be downloaded and self-hosted, unlike closed models reachable only through a vendor's API — let developing economies keep sensitive data within national borders and fine-tune privately. India's institutional answer is the IndiaAI Mission (MeitY, March 2024, outlay ₹10,372 crore) [1], a substantial but still maturing response.

Why open weights matter for sovereignty

• Data residency: self-hosting keeps citizen and enterprise data in approved domestic environments, aligning with DPDP-type consent and localisation concerns.
• Vendor lock-in: reliance on a single foreign frontier lab exposes users to its pricing and roadmap; downloadable weights break that dependence.
• Cost: lower per-token costs, though total ownership cost depends on utilisation and in-house skill.

Strengths of India's response

• Compute as a public good: the IndiaAI Compute Portal has onboarded 38,000+ GPUs and 1,050+ TPUs, shared with startups and academia at subsidised rates under ₹100/hour [1].
• Indigenous models: foundation-model support to teams such as Sarvam AI, BharatGen, Gnani and Socket targets Indian languages and local use-cases [2].
• Governance architecture: the India AI Governance Guidelines, built on seven "sutras" including trust, accountability and innovation over restraint, propose an AI Governance Group, a Technology & Policy Expert Committee and an AI Safety Institute [3].
• Open data commons: AIKosh pools thousands of datasets and sectoral models for public reuse [1].

Limitations

• Guidelines are advisory, not statutory — enforcement and liability remain unsettled [3].
• Hardware dependence persists: GPUs are imported, so compute sovereignty is rented, not owned.
• Capacity gap: running open-weight models reliably at scale demands MLOps talent that most public agencies and MSMEs lack.
• Indigenous models are early-stage against frontier capability.

India has correctly reframed AI from a procurement question to a sovereignty question, pairing subsidised compute with a rights-based governance frame. Deepening the effort — statutory backing for the guidelines, convergence with the Semiconductor Mission, and public-sector AI skilling — would convert present infrastructure into durable strategic autonomy.