Artificial intelligence is often imagined as something that lives in the cloud. In reality, it lives on the ground, drawing vast amounts of electricity, occupying land, and reshaping industrial geography. As India embraces AI, the challenge is no longer just technological ambition, but how the country fits into the physical and economic systems that make AI possible.
Much of the public debate in India has focused on semiconductors, specifically chips, supply-chain vulnerabilities, and subsidies for chip manufacturing. These efforts are important. But they represent only one part of a much larger system. At scale, AI is an industrial ecosystem operating across five interlinked layers: energy, capital, infrastructure, and geopolitics as much as by algorithms. Energy determines affordability; chips determine who can build systems; infrastructure enables scale; models concentrate control over intelligence; and applications capture economic value. Each layer has distinct economic and strategic implications, and no country dominates them all. India’s AI trajectory reflects activity across these layers, marked not by dramatic lag or leapfrogging, but by deliberate choices shaped by scale, affordability, and sovereignty.
For AI-intensive workloads, energy and infrastructure are inseparable. Power requirements rise steeply with scale. Small enterprise data centres consume 1-5 megawatts (MW); large cloud facilities operate at 10-30 MW; hyperscale centres draw 50-100 MW; and frontier AI training clusters increasingly require 100-500 MW or more of continuous power. With one megawatt capable of supplying electricity to roughly 1,000 Indian households, a single AI campus can draw power equivalent to that of an entire district. Electricity is also the dominant operating cost, accounting for 40-60 per cent of expenditure in large data centres. As a result, AI infrastructure gravitates towards locations where power is cheap, reliable, and contractually secured over decades. This is not an IT optimisation problem, but an energy economics problem.
Strategic infrastructure
Governments worldwide now treat AI data centres as strategic infrastructure. In the US, states offer tax incentives and discounted power; China has created state-backed “AI power zones”; and in the Middle East, campuses are co-located with gas and solar facilities to secure long-term low-cost energy. India’s position reflects both ambition and constraint. With 1.5-2 GW of data-centre capacity projected to reach 10-14 GW by the mid-2030s, growth is underway, but uneven 24×7 industrial power has led to clustering in select States. Nuclear energy has therefore re-entered policy discussions to ensure firm power for AI infrastructure.
Chips are the engine and the geopolitically sensitive layer. The US dominates chip design and software ecosystems; Taiwan leads advanced fabrication; South Korea controls memory; and Europe supplies critical manufacturing equipment.
India is a global hub for semiconductor design talent, yet it has almost no presence in advanced manufacturing for AI accelerators. This is not a policy oversight. Leading-edge fabrication plants cost $15-25 billion each, require continuous reinvestment, and take years to stabilise, risks that only a handful of economies can absorb.
India’s strategy prioritises foundational capabilities: mature-node fabrication, assembly and testing, advanced packaging, and reliability engineering, often through partnerships. Though unlikely to yield frontier AI chips soon, these efforts reduce vulnerability and strengthen depth in a strategic supply chain.
The model layer, foundation models, large language models, and multimodal systems are where AI intelligence is created and where concentration is sharpest. Training frontier models demand vast computing power, significant capital, and tolerance for repeated failure, limiting ownership to a small number of organisations, largely in the US, with China developing a parallel ecosystem.
India does not host or control frontier-scale models, but Indian researchers are deeply embedded in global development, contributing to architecture design, optimisation, safety, and deployment. The constraint is not talent, but access to concentrated compute and capital. Domestic efforts are focused on adaptation, Indic-language models, domain-specific systems, and fine-tuning global models for local data and regulatory contexts.
AI applications with factories, banks, hospitals, and governments do not require ownership of chips or frontier models; instead, it rewards domain expertise, integration capability, and scale. In India, AI is used to predict defects, optimise yields, manage fragmented supply chains, detect fraud at a population scale, assist diagnostics, and augment digital public infrastructure. However, reliance on external platforms raises concerns about long-term costs and autonomy. .AI is not a single race but a layered industrial system. India’s transition across these layers is shaped by energy economics, capital intensity, global integration, and domestic scale.
Aparna is Director and Co-Founder, and Gaurav is Director, Centre for Innovation and Trade Economy
Published on May 26, 2026