Cheap intelligence, expensive institutions: What the data really says about India

Artificial intelligence has become astonishingly cheap. India’s institutions, however, still move at the same old speed. That mismatch now sits at the heart of India’s AI story. As the year draws to a close, it feels like the right moment to pause — not to speculate about where artificial intelligence might go next, but to assess where it has already taken us.

Over the past year, much of my writing has focused on AI’s impact on jobs, energy, governance, education, and even on how it subtly changes what we count as knowledge. These were not neutral updates. They were opinions, based on data and offered openly.

Yet as the months passed, one question returned more persistently than any other. If artificial intelligence is developing so rapidly — and becoming so cheap — why does real economic and institutional change in India still feel so slow? To answer that, we have to step away from slogans and scrutinise what the data actually says.

Earlier this year, Stanford University released its AI Index Report 2025, among the most comprehensive global assessments of artificial intelligence to date. It tracks costs, adoption, productivity, energy use, investment, and policy trends across countries. Buried beneath its charts and rankings is a simple but unsettling conclusion: intelligence is no longer scarce.

Between 2022 and 2024, the cost of running advanced AI systems collapsed. Using language models with capabilities comparable to GPT-3.5 became more than 280 times cheaper in just two years. Depending on the task, inference costs fell anywhere between nine-fold and nine-hundred-fold annually. What once required elite research labs and large budgets is now available at near-trivial marginal cost. For a country like India, this should have been a turning point. We never suffered from a shortage of intelligence. It produces engineers at scale, exports software globally, and has built some of the world’s largest digital public platforms. Historically, India’s

comparative advantage has not been inventing every technology first, but absorbing and deploying technologies rapidly once they become usable. Cheap intelligence should have aligned perfectly with that strength.Yet the same data tells a more sobering story. Globally, nearly four out of five organisations reported using AI in 2024, and adoption of generative AI more than doubled in a single year. Despite this, the economic impact remains modest. Most firms reported cost savings of less than 10 per cent and revenue gains below 5 per cent.

Productivity has improved, but slowly and unevenly. In small economies, such gains barely register. In India, they should have been transformative. That they have not been telling us something crucial. The primary constraint is no longer technological capability. It is institutional absorption.

Across India —within Government departments, municipal administrations, public service systems, and even large private firms — AI has largely been layered on top of existing processes rather than allowed to alter them. Dashboards multiply, but decisions still crawl. Predictive tools exist, but final authority remains manual. Automation is introduced, then quietly neutralised through additional approvals, reviews, and signatures. Intelligence is present. The trust is not. This is not because Indian institutions are irrational or resistant to technology.  It is because they have been created by history. Decades of scarcity, legal exposure, and uneven accountability have taught organisations a powerful lesson: avoid blame. A wrong decision is visible and punishable.

A delayed decision rarely is. Over time, systems evolve to protect themselves rather than optimise outcomes. When intelligence was expensive and uncertain, such caution made sense.

When intelligence is cheap and abundant, it becomes an invisible economic drag. The AI Index highlights another reality India has only begun to confront: artificial intelligence is no longer just software. It is infrastructure. The computing power required to train leading AI models now doubles roughly every five months.

Dataset sizes double every eight months. Even as hardware becomes cheaper and more energy-efficient, total energy consumption continues to rise because scale grows faster than efficiency improves. Globally, major technology companies are restructuring long-term energy strategies around AI workloads, including renewed investments in nuclear power to ensure stable supply.

This matters deeply for India. A country cannot scale intelligence faster than it can power it. Yet much of India’s AI discourse remains oddly detached from electricity reliability, grid resilience, cooling capacity, land availability, and long-term energy planning. We talk endlessly about skills and models, but far less about the physical systems — power, water, heat, and infrastructure — that make intelligence usable at scale. Without these foundations, AI remains a pilot project rather than a productivity engine. Investment patterns reinforce this diagnosis. In 2024, private AI investment in the United States exceeded $109 billion. China attracted roughly $9.3 billion. India trails far behind both. This gap is not explained by GDP alone or by venture capital depth.  It reflects something more fragile and harder to measure: institutional confidence.

Confidence that policies will remain stable. Confidence that infrastructure will scale. Confidence that long-term, compute-heavy investments will not be undermined by regulatory uncertainty or logistical bottlenecks.

Young Indians experience this contradiction intuitively. Many already use AI daily — to study, write, code, design, and think. For them, intelligence feels abundant. What feels scarce are institutions willing to recognise new skills, reward initiative, and redesign workflows around what technology now makes possible. When intelligence becomes cheap but opportunity remains constrained, optimism curdles into frustration. Seen through this lens, the data does not point to an AI failure. It points to an institutional one. We do not lack intelligence — human or artificial. It lacks readiness. Artificial intelligence will continue to improve.

Models will become cheaper, smaller, and more capable. The global advantage is already shifting away from those who merely invent intelligence toward those who deploy it effectively. Access is no longer the bottleneck. Absorption is.

As we enter a new year, its central AI challenge is not ambition or talent. It is whether institutions — Governments, firms, universities, and public systems — are willing to change how they decide, how they trust, and how they tolerate risk, so that cheap intelligence can translate into real productivity.

History does not reward potential. It rewards preparedness. In an age where intelligence is abundant, hesitation is no longer neutral for India. It is a choice — and an expensive one.

Author is a theoretical physicist at the University of North Carolina at Chapel Hill, US, and the author of the forthcoming book Last Equation Before Silence; views are personal