Evaluate India's IndiaAI Mission as a response to growing dependency on foreign foundational AI models.

Foundational models are the base layer on which all AI applications sit. Indian startups increasingly build on low-cost Chinese open-weight models, access to which analysts expect to be graduated or restricted by around late 2028 [5]. The IndiaAI Mission (₹10,371.92 crore, seven pillars) [1] is India's structural hedge — directionally correct, but not yet a substitute.

Where the Mission answers the dependency

• Compute floor: common compute capacity has crossed 34,000 GPUs, offered to startups and researchers at subsidised rates, removing the single largest entry barrier [2].
• Sovereign models: support for indigenous large and small language models tailored to Indian languages and local use-cases, shifting firms from pure wrappers to home-built bases [3].
• Full-stack design: datasets platform, FutureSkills, startup financing and Safe & Trusted AI address talent and governance, not compute alone [1].
• Strategic optionality: a domestic stack preserves fallback capacity against unilateral withdrawal of foreign weights [5].

Where it falls short

• Sub-frontier by design: the model portfolio targets linguistic coverage, not frontier parity [3] — it does not replace what foreign frontier models currently supply [5].
• Outlay mismatch: ~₹10,300 crore spread across seven pillars over five years [1] is small against single frontier training runs abroad.
• Dependency substituted, not removed: subsidised compute is largely imported silicon hosted domestically [2], trading a model dependency for a hardware one.
• Thin value capture: open weights yield inference savings, not pre-training capability; fine-tuning a foreign base is limited sovereignty.
• Evaluation gap: no published red-teaming or model-assurance standard for foreign open-weight models, leaving departments with blanket rather than graded decisions.

On balance, the Mission is a credible foundation but an incomplete shield: it builds an ecosystem where a capability floor is needed. The Economic Survey's steer — decentralised, application-driven systems over capital-intensive frontier models, to avoid fragile dependencies [4] — is the right calibration. Pairing it with mandated model portability, efficiency-focused research and domestic archiving of checkpoints would convert a cost hedge into genuine technological self-reliance.