The ₹10,372 crore AI gamble: Why software gloss cannot mask hardware vacuity

By Sanjay Pandey

The IndiaAI Mission, with its multi-year outlay of ₹10,372 crore (approximately $1.2 billion), represents a serious commitment to building sovereign artificial intelligence capabilities. As deliberations at the India AI Impact Summit in New Delhi this week highlight compute expansion, green infrastructure, and semiconductor synergies, the narrative is one of acceleration. Government-backed programs have contracted and allocated compute capacity equivalent to more than 38,000 GPUs through public-private partnerships, with an additional 20,000 targeted under a proposed “AI Mission 2.0.”

This is progress. But progress is not sovereignty.

Having overseen security systems serving more than 120 million citizens, I see a strategic risk: an emerging sovereign trap where we master software layers while remaining structurally dependent on external hardware foundations.

The Black Box Problem and the “Autonomous Defamer”
A recent reported incident involving the Python plotting library Matplotlib illustrates a growing oversight challenge. Developer Scott Shambaugh alleged that an AI agent published a critical blog-style post targeting him after he rejected its code contribution. While the degree of autonomy and possible human involvement are debatable, the episode demonstrates how AI systems operating with minimal supervision can generate harmful defamatory content at scale.

We are moving beyond conventional misinformation toward scenarios where semi-autonomous AI agents can produce persistent, logic-structured narratives that blur accountability. When such systems operate through opaque model weights and foreign-controlled infrastructure, retrospective audits provide limited comfort. The governance question is no longer whether content can be generated — but who retains command authority when it is.

In early February 2026, French prosecutors conducted a raid on the Paris offices of X as part of an ongoing criminal investigation into alleged unlawful data practices and harmful AI-generated content, including deepfake-related concerns. This action does not represent a final legal judgment, but it signals a clear shift: states are increasingly intervening where algorithmic systems intersect with public harm.
India must anticipate similar tensions.

The European Lesson: Software Sovereignty Meets the Hardware Wall
Europe’s experience is instructive. Mistral AI emerged as a flagship European alternative in the large-language-model race. Yet its models run largely on hardware supplied by firms such as NVIDIA.

Advanced AI chips like the NVIDIA H100 and NVIDIA H200 have become geopolitical chokepoints amid US–China export controls. Access can be restricted, throttled, or repriced. Hardware, not software, is the leverage layer.

Europe did not fail — but it learned that model-layer independence without compute-layer autonomy remains partial sovereignty.

India’s roadmap currently emphasizes “compute as a service,” which is pragmatic for rapid deployment. Yet if this compute is predominantly foreign-designed and externally controlled, our sovereign AI architecture risks resembling tenancy rather than ownership.

Semiconductor Ambitions and Strategic Alignment
India’s semiconductor push under the broader ₹76,000 crore incentive framework is a structural correction long overdue. Budget 2026–27 announcements include approximately ₹1,000 crore in additional allocations toward ecosystem components such as equipment, materials, design IP, and supply-chain strengthening.

This momentum is encouraging.

However, AI-specific compute demands tighter alignment between the IndiaAI Mission and the India Semiconductor Mission. Rather than leap immediately to bleeding-edge general-purpose GPUs — a multi-billion-dollar, multi-year endeavour — India could prioritize trusted domestic AI accelerators and specialized compute architectures aligned with national workloads.

The goal is not symbolic chip-making. It is strategic compute resilience.

The Thirsty Machine: AI and the Water-Security Nexus

AI’s physical footprint extends beyond silicon.

Depending on cooling architecture and climate conditions, large hyper scale data centers can consume from hundreds of thousands to several million gallons of water per day, particularly when evaporative cooling systems are used. In water-stressed geographies, this introduces a strategic trade-off between digital infrastructure and community resource security.

Globally, resistance to large-scale data center expansion is rising where groundwater stress and ecological impacts are perceived to outweigh local benefits. For India — already confronting regional water scarcity — scaling AI infrastructure without a hardware-water policy would replicate avoidable externalities.

Sovereignty cannot come at the cost of sustainability.

Three Pillars for a More Resilient AI Mission
1. Close the Infrastructure Gap: Rapid scaling through contracted compute is sensible. But long-term sovereignty requires indigenous, trusted hardware stacks — whether through domestic accelerators, joint ventures, or strategic fabrication partnerships.

2. Build Active Command Layers: Supervisory filters and command architectures must sit between users and high-capability models, particularly when foreign libraries and pre trained systems are integrated. Governance must be embedded in runtime systems, not limited to post-facto audits.

3. Institutionalize Sustainability: Standards Mandate water-efficient cooling systems — such as dry or closed-loop technologies — and prioritize non-potable sources for new data centers. AI policy should integrate environmental safeguards as first principles, not afterthoughts.

Own the Bed, Not Just the Sheet
India’s AI trajectory is promising subsidized compute access, indigenous model experimentation, and semiconductor ecosystem milestones with initial commercial outputs expected soon. But sovereignty requires ownership of the foundational layers that power innovation.

Data sovereignty protects information. Model sovereignty protects cognition. Silicon sovereignty protects continuity.

A nation that builds intelligent software atop imported hardware while straining local resources risks strategic contradiction. If India’s AI mission is to embody “People, Planet, Progress,” its hardware foundations must match its software ambition.

Otherwise, we risk polishing the sheet while renting the bed.

— Sanjay Pandey is a Cyber-Policy & Digital Jurisprudence Expert with a distinguished career at the intersection of technology, law, and national security. An alumnus of IIT Kanpur, Harvard University (MPA), and an LLB graduate, he brings deep expertise in cyber governance, digital regulation, and institutional resilience. He has served as Former Director General of Police, Maharashtra, and Former Commissioner of Police, Mumbai, leading one of India’s most complex security ecosystems. His work spans national security strategy, cybercrime response, digital evidence frameworks, and strengthening institutional capability in an era of rapid technological disruption.