The grid, the chip and the future of AI

Artificial intelligence is becoming less a contest of algorithms than of electricity, chips and State capacity. The race for artificial intelligence is often portrayed as a contest over code. Increasingly, it is a contest over power supply, semiconductor access and institutional competence.

The material weight of “weightless” intelligence

From Arizona to Shenzhen to Navi Mumbai, data centres now anchor the AI economy. Some large facilities consume as much electricity as a town of 100,000 people. Training advanced models requires tens of thousands of specialised chips operating

continuously. What is marketed as a cloud rests on land, cooling systems and high-voltage cables. Competitive advantage in AI will depend not only on designing better models, but on building and sustaining the infrastructure that runs them.

India’s ambitions are expansive. The Government has committed funds under the IndiaAI Mission to expand domestic computing capacity, with tens of thousands of

high-end graphics-processing units deployed and more planned. Subsidised access is meant to allow local firms to train and adapt models without relying entirely on foreign cloud providers.

The grid behind the cloud

Compute capacity is only as reliable as the grid beneath it. AI workloads require uninterrupted electricity and predictable pricing. India’s power system has improved markedly over the past decade, yet transmission bottlenecks, distribution losses and uneven State finances remain constraints.

Expanding renewable generation adds further complexity. Data centres also require substantial water for cooling, raising questions about siting and environmental trade-offs. The economics are exacting. Building a large data centre demands billions in capital expenditure and long planning horizons. Securing advanced chips involves navigating supply chains increasingly shaped by export controls and geopolitical rivalry. If domestic power is unreliable or costly, firms will default to foreign infrastructure. Technology policy therefore intersects with energy reform, industrial policy and fiscal management. Public subsidies may accelerate early capacity building. They also create opportunity costs. Funds directed toward compute clusters compete with spending on universities, rural connectivity and semiconductor fabrication. The sequencing of investment will matter as much as its scale.

Adoption, not applause

Infrastructure is necessary but insufficient. The economic impact of generalpurpose technologies has historically depended on diffusion. Electricity raised productivity when factories reorganised around it. The internet transformed commerce when firms rebuilt supply chains and distribution channels. AI is unlikely to be different.India’s digital public infrastructure provides a platform for diffusion. Aadhaar and the Unified Payments Interface operate at national scale, generating vast volumes of transaction data and lowering the cost of digital service delivery. These systems could support AI applications in credit scoring, fraud detection and logistics optimisation.

Yet adoption remains uneven. Many small and medium-sized enterprises lack the capital, data quality or managerial capacity to integrate AI tools into daily operations. In the public sector, deployment requires procurement reform, interoperable data standards and retraining of officials. Without organisational change, sophisticated models will not translate into measurable productivity gains. The constraint is not merely technological. It is institutional.

The human bottleneck

Human capital presents a parallel challenge. India produces large numbers of engineers and has a growing pool of AI specialists. Hiring in AI-related roles is increasing, and training initiatives are broadening exposure to the field. Frontier research, however, remains concentrated in a few global hubs, most notably in the United States and increasingly in China. America hosts a large share of leading AI laboratories and benefits from deep capital markets and dominant cloud providers. China combines State direction with manufacturing scale and a formidable hardware ecosystem. India’s comparative advantage lies in scale, digital infrastructure and a youthful workforce. Whether that proves sufficient will depend on the depth of research funding, the quality of universities and the ability to retain high-end talent.

The risk is bifurcation. A thin layer of advanced capability may coexist with limited diffusion into the broader economy. Bridging that gap requires not only elite research institutes but also managerial expertise and sector-specific knowledge across manufacturing, agriculture and services.

Capacity over ambition

As AI shifts from laboratory breakthrough to industrial system, administrative competence becomes as important as technical ingenuity. Expanding power capacity, attracting semiconductor investment, strengthening research universities and enabling enterprise adoption require coordination across ministries and between public and private actors. Regulatory predictability will shape capital flows as much as headline funding commitments.

The contest over algorithms will continue. Model performance will improve incrementally. But durable advantage will accrue to those able to align energy infrastructure, chip supply and institutional reform. For India, as for others, the central question is not whether it can produce capable models, but whether it can build and operate the system that sustains them.

Clever code travels only as far as the grid allows Artificial intelligence may be written in software, but it is forged in the fires of heavy industry, sustained by a middle class of applied talent, and protected by institutional willpower. The race for the smartest algorithm is over. The race for systemic capacity has just begun. And in this new industrial revolution, code is no longer the destination — it is merely the first step.

Writer is a General Secretary, PanIIT Alumni India; views are personal