Every day, millions of people ask artificial intelligence (AI) to execute a wide range of personal and professional tasks. From writing emails, generating images, summarising documents, planning holidays, to solving coding problems and content creation – AI is doing everything. Consequently, AI-systems are swiftly gaining popularity globally. OpenAI's ChatGPT crossed 900 million weekly active users by February 2026 (adoption growth being fastest in Africa and Asia), Google's Gemini reported over 750 million monthly active users by February 2026 and China’s DeepSeek (a later entrant) also attracted tens of millions of users within months of its launch.

This is relevant since nearly 82% of global population above 10 years of age owns a mobile phone and the gap between mobile phone access and internet connectivity is narrowing, despite significant disparities between high- and low-income countries. Each new smartphone user, internet subscriber, and cloud-connected device, becomes a potential user of AI services. In fact, a joint report by Flipkart and Counterpoint Research found that 89% of consumers considered AI-enabled features to be important while making smartphone purchase decisions.

As AI gets increasingly embedded into everyday life, it is worth pausing and asking a simple question: what does an AI prompt "actually" cost – not in terms of money paid by the user, but in terms of the required infrastructure, electricity, and resources?

Every AI interaction depends on an enormous physical system

AI, a digital technology, often triggers a sense of detachment from the physical world because it appears to exist in the "cloud". There are no visible machines or factory floors, and no typically visible supply chains behind its instantaneous responses. Further, user interactions with AI systems, even if short, make the process feel highly efficient and almost costless. In reality, though, every AI interaction depends on an enormous physical system. Behind every prompt are data centres, servers, fibre-optic cables, cooling systems, storage facilities, power stations, and global supply chains that stretch across multiple continents. The scale of AI adoption helps explain why this matters.

The rapid expansion of digital technologies (including AI) has increased demand for critical minerals such as lithium, silicon, cobalt, and rare earth elements. Global lithium demand increased by almost 30% in 2024, reflecting its growing importance in electric vehicles (EVs), energy storage systems, and digital technologies (International Energy Agency, IEA, 2025). Silicon, the key element for modern semiconductor manufacturing, is also used in advanced chips that train and operate AI models. Meanwhile, cobalt and several rare earth elements are also crucial for batteries, electronic systems, and other technologies that support the broader digital infrastructure on which AI relies. In other words, an AI prompt represents the final step of a much longer supply chain that begins with physical extraction and processing of resources globally.

At the centre of the physical AI infrastructure are data centres. Often described as "the cloud", data centres are, in reality, large facilities filled with servers, graphics processing units (GPUs), networking equipment, storage devices, cooling systems, and backup power infrastructure. They store, process, and transmit the information that supports modern digital economy. Whether someone is streaming a film, storing photographs online, accessing a banking application, or interacting with an AI assistant, the request is ultimately handled by equipment housed within these facilities.

AI adds a new dimension to this infrastructure by increasingly asking them to perform large-scale, simultaneous, and computationally-intensive tasks. With AI adoption, demand for energy-intensive accelerated servers, specialised computing hardware, data-centre capacity, and electricity will increase. While every expanding industry competes for scarce resources, the speed and scale of AI adoption may place new demands on electricity systems, infrastructure planning, and resource allocation. This necessitates prior planning.

Data centres and energy

Data centres account for roughly 1.5% of global electricity consumption, growing at 12% per annum (IEA, 2025). A typical large data centre may consume 1.2-2.4 million kWh of electricity daily, larger hyperscale facilities consume more. To put this into perspective, an average urban Indian household consumes roughly 3-6 kWh of electricity per day. A single large data centre can, therefore, consume electricity equivalent to 200,000-800,000 urban Indian households, daily. Thus, AI infrastructure is becoming an important component of the overall electricity demand. With present AI infrastructure spending increasingly outpacing earnings, electricity consumption from data centres – growing 17% in 2025 – is projected to double by 2030 and that from AI-focused facilities – growing faster (over 17%) – is likely to triple (IEA, 2026). Then a critical question pertains whether power systems can expand quickly while maintaining affordability, reliability, and sustainability.

Some estimates suggest that there approximately 11,800 data centres were operating worldwide by June 2025, with the highest concentration in countries like the US, Germany, the UK, China, and France (Lee and West 2025) (Figure 2). However, many emerging economies are increasingly competing to attract data-centre investments as part of broader digital economy strategies. India is no exception (Figure 3). To integrate within the global AI economy, it is simultaneously expanding digital infrastructure, attracting investment, and striving to balance technological ambitions with broader development goals. Mumbai is presently India's largest data-centre hub, accounting for over a quarter of India's live data-centre capacity (Government of India, 2025). Other major clusters include Chennai, Bengaluru, Hyderabad, Delhi-NCR, Pune, and Kolkata. Both, at the national and sub-national levels, measures have been introduced to attract investment in the AI infrastructure.

Maharashtra's 2023 data-centre policy included 100% electricity-duty exemptions, stamp-duty waivers, and regulatory relaxations. Similarly, Karnataka's Data Centre Policy (2022-2027) provides (10%) land subsidies and 100% stamp-duty exemptions (Government of India, 2025). Some other states – Telangana, Uttar Pradesh, Tamil Nadu, Gujarat, Odisha, and West Bengal – with a data-centre policy, also include commitments to support renewable energy integration and advanced cooling infrastructure. At the national level, the Union Budget 2026-27 proposed a tax holiday till 2047 for foreign firms providing cloud-based services through Indian data-centre infrastructure.

These incentives reflect a broader policy trade-off. While aimed at strengthening India’s AI competitiveness, they may also reduce the extent to which firms bear the full costs of substantial resource consumption. Governments view AI infrastructure as a source of investment, innovation, and economic growth, and are understandably reluctant to be left behind. Simultaneously, companies operating at the forefront of the AI industry are generating extraordinary revenues. OpenAI, for example, reportedly crossed billions of dollars in annualised revenue within a remarkably short period. This raises questions about cost-benefit distribution, when supporting this resource-intensive industry. Additionally, externalities such as environmental impacts (higher water consumption and carbon emissions), costs of expanding power infrastructure, and pressure on energy systems may not be entirely borne by private investors. Therefore, the policy challenge includes balancing strategic benefits of AI infrastructure with the efficient use of scarce resources, while accounting for broader economic and environmental costs.

In this context, every new data centre and increase in AI adoption raise a broader question: can the energy needs of an increasingly AI-driven economy be met in an environmentally sustainable and reliable manner? India's recent heatwave offers a glimpse of this challenge. In May 2026, peak electricity demand reached a record 270.8 GW, fuelled by demand for cooling, met only partly (34%) by renewable energy (Goyal 2026). Episodes of power outages owing to additional pressure on power grids are not uncommon during Indian summers. Importantly, fossil fuels remain the backbone of electricity generation in India, supplying more than 70% of power.

This makes the expansion of AI infrastructure a critical energy security question in addition to a climate one, primarily because pressures on electricity systems can have uneven consequences across users. These emerging electricity-system pressures are situated within a broader energy security context for India, as recent Middle-East tensions have once again highlighted the vulnerability of global energy markets to geopolitical shocks. Despite diversification, crude oil imports (particularly from the Middle East) play an integral role in supporting the Indian economy. More broadly, much of Asia remains dependent on external energy supplies – 60% comes from Middle East – making it particularly sensitive to disruptions in global energy flows and fuel markets.

The future of AI is not just about innovation

As AI becomes more deeply integrated into everyday life, the electricity needed to support data centres, cloud infrastructure, and advanced computing systems will continue to grow. The question is no longer only whether countries can build better, more innovative AI models. It is also whether the underlying energy and infrastructure systems can expand fast enough – and remain stable enough – to support them, while managing the associated economic and environmental costs. Discussions about AI have often focused on what algorithms can do, privacy, and digital dependence. Although important, these concerns sometimes overshadow another question: what does it physically take to power AI? Every AI prompt depends on electricity, infrastructure, and supply chains that extend way beyond the screen in front of us. The future of AI, therefore, is not just about innovation. It is also a question of energy security, infrastructure, and whether systems powering the digital economy can expand sustainably, while keeping pace with the technologies they support.

Note:

1. Rapidly growing AI revenues do not necessarily imply commensurate returns on investment. Current growth of AI revenues lags behind the scale of AI investment. See, The Economist (2026).
2. A survey by the Stagwell National Research Group in the United States revealed concerns among nearly one-third smartphone users about AI systems making decisions on their behalf without explicit consent.

Neha Arya holds a Ph.D. from the Department of Humanities & Social Sciences of the Indian Institute of Technology (Delhi). Her research interests centre around the impact of technology on the labour market.

This article was originally published on Ideas For India.

The Wire is now on WhatsApp. Follow our channel for sharp analysis and opinions on the latest developments.