ChatGPT's Real Water Use, India's AI Boom & What Hyperscalers Can't Do | Ft. Sify CEO

2026-06-20 · Source: AIM Network · Field: Technology & Digital — Cloud Computing & IT Infrastructure, Artificial Intelligence & Machine Learning · Depth: Intermediate, long

Summary

Sify CEO Sharad Agarwal discussed India's data center landscape, clarifying that ChatGPT queries do not consume vast water, as Indian facilities largely employ closed-loop chiller systems, requiring only 3,000-10,000 liters per megawatt initially and minimal annual top-ups. He attributed Mumbai's data center saturation to historical stickiness and hyperscaler demands for geographically dispersed sites. Agarwal noted India's AI infrastructure investment currently prioritizes inferencing. Sify's competitive edge is its capability to build data centers from land acquisition to production in under 24 months. He stated Tier 3 redundancy (N+1 or N+2) is generally adequate, with only 2-3% demand for Tier 4, which is often wasteful. Sify is expanding into Tier 2 cities like Patna and Lucknow, driven by hyperscalers, GPU neo clouds, enterprises, and edge computing, aiming to diversify revenue beyond its current 70-80% reliance on five hyperscalers.

Key takeaway

For data center operators or AI/ML directors planning infrastructure in India, prioritize closed-loop cooling systems to ensure water efficiency, aligning with local environmental realities and reducing operational costs. You should also design for Tier 3 redundancy (N+1 or N+2) as it meets most demands while minimizing resource waste. Consider expanding into Tier 2 cities, as these locations offer strategic advantages for serving diverse demands from hyperscalers, GPU neo clouds, and enterprises, leveraging local expertise for faster deployment.

Key insights

Data center design in India prioritizes water efficiency and optimized redundancy while addressing local market demands.

Principles

• Closed-loop cooling significantly reduces data center water consumption.
• Tier 3 redundancy (N+1/N+2) is typically sufficient and more sustainable than Tier 4.
• Local market knowledge accelerates data center deployment.

Method

Data center cooling can use evaporative (3-5M gal/MW/yr) or closed-loop systems (3k-10k L/MW one-time, 100-1k L/annum/MW leakage). Gray water can supplement evaporative cooling.

In practice

• Adopt closed-loop water chiller systems for new data center builds.
• Design for Tier 3 redundancy unless regulatory demands mandate Tier 4.
• Explore Tier 2 cities for data center expansion to meet diverse demands.

Topics

• Data Center Infrastructure
• Water Conservation
• AI Inference
• Tier 3 Redundancy
• India Data Centers
• Hyperscale Cloud

Best for: CTO, VP of Engineering/Data, Investor, Director of AI/ML, MLOps Engineer, Consultant

Related on AIssential

• Bridge Data Centres: Sustainability at the Hyperscale Level — Hyperscale data centers are prioritizing sustainable, vertically integrated, and rapidly deployable infrastructure for AI workloads.
• India’s AI Infrastructure Bet Faces Its First Real Test — Reliable compute infrastructure, not talent, is India's primary AI scaling challenge.
• The great American data center divide — Rural communities are increasingly resisting data center expansion due to concerns over water, energy, and environmental impact.
• The Silicon Sanctuary: Why the World is Relocating its Digital Brain to India — India is becoming a global digital infrastructure hub by integrating chip manufacturing, green energy, and data residency.
• Startup’s nuclear-inspired cooling system could make data centers more sustainable — Adapting nuclear reactor heat transfer principles can significantly improve data center cooling efficiency and sustainability.
• HSCALE: Fulfilling the Power Requirements of AI Clusters — HSCALE's Milan campuses offer flexible, high-density AI infrastructure with committed power and renewable energy, addressing critical operational challenges.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AIM Network.