India Weighs Mandating Domestic AI Models in Defence and Aerospace

India's Union Government is likely to direct critical-sector departments at central and state levels to stop using frontier AI models built by non-Indian vendors and to deploy indigenous systems instead, Business Standard reported, citing senior government officials. Defence and aerospace sit at the centre of the plan. The departments are also being asked to review how domestically developed AI systems can be deployed in these sectors, according to the report.

The move targets sovereign control over AI (the ability of a state to run its own models rather than depend on foreign providers) in areas where data sensitivity and national security are paramount. If formalised, the policy would redirect procurement away from US and Chinese frontier model providers toward a small pool of Indian startups still shipping their first releases.

How capable are the domestic models?

One official told Business Standard: "The idea is to have enough technological capability to use these models in crucial sectors. We are in constant talks with all the companies (who won bids to develop indigenous AI models) to see if we can expand the scope of the models they have built so far."

The same official said some domestic models have reached up to 80 per cent of the capacity of leading frontier AI models. That figure frames the government's core bet: near-frontier performance is sufficient for critical-sector workloads if the model runs on Indian-controlled infrastructure.

The government has approved the development of 20 AI models under the IndiaAI Mission. The portfolio spans general-purpose large language models (LLMs, AI systems trained on broad text corpora) and sector-specific small language models (SLMs, compact models tuned for narrow domains). Twelve startups selected under the mission are building them.

Released models so far include:

• Several models from Sarvam, one of the mission's highest-profile winners
• Multiple releases from BharatGen
• Varya, a video-generation model
• Soket's EKA-1 foundation model
• ZenteiQ's Dhir 2B and Dhir 5B

Companies including Gnani, Gan AI, Fractal, Tech Mahindra and Shodh AI plan to release their LLMs and SLMs over the next 12 to 15 months, the official told the publication. The government is in talks with the selected companies to expand the capabilities and scope of the models already delivered.

How much money has actually moved?

The IndiaAI Mission carries a five-year outlay of Rs 10,371.92 crore. Actual disbursement, disclosed by the Ministry of Electronics and Information Technology (MeitY) in a Rajya Sabha reply, remains a fraction of that figure.

As of April 2026, the numbers stand as follows:

• Rs 21.79 crore released in 2024-25, against revised estimates of Rs 173 crore
• Rs 379.15 crore released in 2025-26, against revised estimates of Rs 800 crore
• No funds released for 2026-27 against a budget estimate of Rs 1,000 crore

Combined, roughly Rs 400 crore has moved against a Rs 10,371.92 crore commitment. That gap between approval and disbursement will shape how quickly the 12 vendors can close the remaining capability distance to frontier models.

What else is in motion under the mission?

Two related developments contextualise the critical-sector push. MeitY has invited AI compute providers to supply 10,000 GPUs under the IndiaAI Mission, addressing the hardware constraint that limits domestic model training. Separately, the government could take a minority stake in Sarvam through IndiaAI Mission support, aligning vendor incentives with state interests.

For CIOs in Indian defence, aerospace and adjacent critical infrastructure, the signal is direct. Procurement roadmaps that assume access to foreign frontier models may need revision. Vendor conversations should now account for a domestic stack — Sarvam, BharatGen, Soket, ZenteiQ and their mission peers — whose delivery timelines stretch 12 to 15 months out.

The capability question remains open. An 80 per cent threshold, reported by an official rather than benchmarked independently, leaves a performance margin that critical-sector deployments may not tolerate. The government's stated answer is scope expansion: continuous negotiation with the 12 selected companies to stretch what they have built.

No formal directive has been published. The policy exists, for now, as reported intent from senior officials. Departments reviewing indigenous deployment options are acting in advance of any mandate, which suggests the direction of travel is set even if the instrument is not.