AI access is now a question of national security

The issue

The BRICS New Delhi Declaration of 12 September says that international cooperation to make AI resources more accessible, while keeping them safe and reliable, is essential for all countries, especially those of the Global South. Leaders committed to implement a separate BRICS statement on the global governance of AI.

At home, the government approved the IndiaAI Mission on 7 March 2024 with Rs 10,371.92 crore over five years. As the government told the Lok Sabha on 5 August 2026, 20 Indian foundation model proposals (12 large multimodal and 8 small language models) have been chosen from 506 applications. Fifteen compute providers have been empanelled, 237 projects have been given subsidised computing worth 93.18 lakh GPU hours, and a system of about 1.1 exaflops has been ordered for the NIC data centre in Delhi. Thirteen projects on safe and trusted AI, covering bias, privacy and algorithm audits, have been selected. The same reply says that India's computing still draws on GPUs sourced from abroad. The Cabinet approved Semicon 2.0, with Rs 1,27,500 crore, on 15 July 2026.

Our view

AI now runs through government services, banking, health and defence. A country that depends on one or two foreign suppliers for its chips, its computing and its best models can be cut off by a trade dispute or an export rule it has no say in. We think India should treat AI access the way it treats oil: spread its sources and keep a reserve.

That means three separate efforts. On chips, Semicon 2.0 is a long-term answer, and in the meantime India needs agreements with more than one supplier. On models, India should back its own foundation models and also keep the option of open-weight models, whose published weights can be run on Indian servers without asking anyone's permission. On safety, India needs the capacity to test any model it uses, Indian or foreign, for bias, security flaws and failures in Indian languages, before it goes into public systems.

BRICS can help with shared tools and training for countries with weak cyber defences. It cannot replace India's own capacity, or its ties with partners such as the United States, with whom India joined the Pax Silica coalition on chip supply this year.

The other side

• Open-weight models can be stripped of their safety limits and misused, so wider access adds risk.
• Full self-reliance in chips is unrealistic for any country; the supply chain is global, and money spent on duplication could go elsewhere.
• A BRICS AI agenda includes China, India's main technology rival, and may extend its influence more than India's.

Using it in Mains

This fits GS Paper 3 (science and technology, cyber security) and GS Paper 2 (India and global groupings). For BPSC, use it in the science and technology section.

Q. "Access to artificial intelligence is now a matter of national security." Discuss India's approach to building sovereign AI capacity.

• IndiaAI Mission: models, compute and safe and trusted AI.
• Dependence on imported GPUs and the role of Semicon 2.0.
• Open-weight models as a hedge, with their risks.
• BRICS cooperation and ties with the United States side by side.
• Keywords: technology denial, digital sovereignty, AI safety testing, Global South.

Sources: PIB, Government Expands Sovereign AI Infrastructure, 6 August 2026; PIB, BRICS New Delhi Declaration