AIは雲ではなく、電力と水で動く｜インドのデータセンター覇権と住民反発｜AI Runs on Power and Water: India’s Data-Centre Ambitions and Local Resistance
What Reuters reported Amazon, Google and Microsoft are committing tens of billions of dollars to AI infrastructure in India. Amazon plans to invest \$21 billion by 2
アーカイブ移行日：2026年11月11日。原記事の初出日は未確認。 / Archived 11 Nov 2026; original publication date unconfirmed.
What Reuters reported
Amazon, Google and Microsoft are committing tens of billions of dollars to AI infrastructure in India. Amazon plans to invest \$21 billion by 2030, Microsoft has announced a four-year \$20.5 billion programme largely focused on AI data centres, and Google faces opposition to a \$15 billion project in Andhra Pradesh.
Near Mumbai, more than 1,000 residents say an Amazon facility was advanced without meaningful consultation. Their concerns include tree removal, constant industrial noise, pressure on electricity and water, and possible damage to residential property values. Amazon says the facility will use water-free cooling, dedicated power separate from household supply, and noise-mitigation measures.
The conflict is therefore not “India versus AI.” It is a dispute over location, infrastructure capacity, transparency and the distribution of costs and benefits.
India’s strategic choice
India cannot yet match the leading frontier models of the United States and China. Its alternative strategy is to become indispensable to the physical operation of the AI economy: hosting data, supplying compute capacity, expanding transmission networks and attracting hyperscale investment.
Official figures show how quickly that layer is expanding. India’s data-centre capacity rose from roughly 375 MW in 2020 to about 1,500 MW in 2025. The government estimates electricity demand from AI and other large data centres could reach 13.56 GW by 2031–32. It has also onboarded about 38,231 GPUs through 14 approved providers under its subsidised compute programme.
This is industrial policy, energy policy and urban policy disguised as a technology story.
Why the resistance is economically rational
A data centre concentrates enormous capital in a relatively small physical area, but its local benefits are not always equally visible. Construction creates activity, yet long-term employment can be limited relative to the electricity, land and cooling resources consumed. Residents therefore ask a reasonable question: if the strategic and financial benefits are national or global, why should one neighbourhood absorb the noise, environmental risk and loss of amenity?
India has only about 4% of global freshwater resources while supporting roughly 18% of the world’s population. Reuters cited estimates that 60%–80% of Indian data centres could face high water stress this decade, while annual sector water use may more than double from approximately 150 billion litres in 2025 by 2030.
Technology can reduce the burden, but it cannot eliminate politics. Water-free cooling, reclaimed water, direct-to-chip liquid cooling and immersion cooling can improve efficiency. Dedicated substations and power procurement can reduce competition with households. None of these measures automatically solves poor siting, insufficient consultation or distrust.
Why this matters to factory automation
An AI data centre is a continuous-process industrial plant wearing a digital label. It requires:
- power distribution, switchgear, transformers and protection systems;
- cooling plants, pumps, valves, heat exchangers and water treatment;
- sensors for temperature, humidity, vibration, leakage and air quality;
- building and energy management systems;
- backup generation, batteries and microgrid controls;
- predictive maintenance, fire detection and access control;
- auditable data on resource use and environmental performance.
The competitive advantage for an FA supplier is therefore not merely “automation.” It is the ability to integrate electrical, thermal and operational systems, demonstrate energy and water savings, maintain uptime, and produce evidence that can be shown to regulators and local communities.
The investable bottleneck
The AI capex boom creates demand for GPUs, but India’s bottlenecks may reward less glamorous layers: grid equipment, power electronics, cooling, pumps, industrial controls, monitoring software, storage and engineering services.
However, not every supplier will capture attractive returns. Rapid capacity expansion can squeeze engineering resources, invite low-margin bidding and expose contractors to delayed permits or redesigns. Investors should distinguish between companies with proprietary equipment and service revenue, and contractors that merely absorb execution risk.
Questions for an Indian FA professional
- Is the larger constraint currently grid connection, cooling, water availability, land approval or skilled commissioning labour?
- Who specifies the control architecture—the hyperscaler, EPC contractor, local utility or equipment vendor?
- Are customers paying a premium for lower water usage and higher energy efficiency, or treating them as compliance costs?
- How often are projects redesigned after community or environmental objections?
- Which components still depend heavily on imports?
- Is after-sales service more valuable than the initial hardware order?
- Are data-centre standards spilling over into factories, warehouses and industrial parks?
- Which local firms have genuine engineering capability rather than only low-cost installation capacity?
Bull, Base and Bear cases
Bull: Data-centre investment accelerates while grid, cooling and automation spending rises even faster. Efficient infrastructure suppliers gain pricing power and recurring service revenue.
Base: Projects proceed, but with relocation, additional mitigation and slower approvals. Capex remains strong, while returns vary sharply by supplier quality and execution.
Bear: Power shortages, water conflict and political opposition delay projects, while hyperscalers demand lower prices. Local contractors carry the cost and equipment orders shift outward.
Verdict
India’s AI opportunity is real, but it will be won below the software layer. The decisive technologies may be the ones that keep power stable, heat controlled, water use defensible and neighbours willing to tolerate the facility.
