Archive · 28 Aug 2026
AI Data Centres Face Rising Power and Cooling Demands
AI data centres are very large buildings filled with powerful computers.
These computers use much more electricity than ordinary computers and also produce a great deal of heat.
Some large AI centres may need as much electricity as a small power plant.
India is planning several such facilities, including projects in Andhra Pradesh, Gujarat and Hyderabad.
The Hyderabad project is expected to use 9,000 advanced NVIDIA processors.
The article says operators should study future electricity needs, backup power and cooling before building.
Power could come from a mixture of the grid, gas, nuclear power, solar and wind.
Cryostats are machines that keep certain sensors and specialised computers extremely cold.
The article says this cooling can reduce noise and help superconducting or quantum systems work accurately.
Andhra Pradesh reportedly approved AMAI Factory Pvt. Ltd.’s ₹31,387-crore AI data centre in Amanam village, Bhimunipatnam, with a planned 512-megawatt green facility.
Reliance Industries is building a 168-megawatt AI-enabled data centre in Jamnagar, Gujarat, which Meta Platforms has agreed to lease.
AM Intelligence ordered 9,000 NVIDIA Vera Rubin GPUs for a 30-megawatt Hyderabad facility, with delivery scheduled for the first quarter of 2027.
Hyperscale AI centres can require 100 to 500 megawatts continuously, while next-generation superclusters may need 500 megawatts to more than one gigawatt.
The article urges feasibility studies covering generation capacity, power sources and cooling, while presenting cryostats as a possible technology for ultra-cold computing and sensing applications.
- Who
- The Andhra Pradesh government, AMAI Factory Pvt. Ltd., Reliance Industries, Meta Platforms and AM Intelligence are identified in connection with the projects; the article is written by a former member of the Central Board of Indirect Taxes and Customs and former Director General of Revenue Intelligence.
- What
- India is developing large AI data centres and evaluating their electricity, generation and cooling requirements, including a possible role for cryostats in specialised systems.
- When
- The article cites AI infrastructure investments during 2024 and 2025, projects NVIDIA Rubin GPU delivery in the first quarter of 2027, and an International Energy Agency projection for 2026.
- Where
- The projects mentioned are in Amanam village in Bhimunipatnam, Andhra Pradesh; Jamnagar, Gujarat; and Hyderabad.
- Why
- AI processors and cooling systems require large, stable supplies of electricity, while ultra-cold systems may be needed for superconducting, quantum, sensor, space and defence applications.
- Andhra Pradesh project
- A proposed 512-megawatt green AI data centre in Amanam village, Bhimunipatnam, with an investment reportedly exceeding ₹31,387 crore.
- Jamnagar project
- Reliance Industries is building a 168-megawatt AI-enabled data centre that Meta Platforms has agreed to lease.
- Hyderabad deployment
- AM Intelligence ordered 9,000 NVIDIA Vera Rubin GPUs for a 30-megawatt first phase.
- GPU delivery
- The NVIDIA Rubin systems are slated for delivery in the first quarter of 2027 and are projected to provide about 450 exaflops of low-precision performance.
- Typical AI power demand
- Hyperscale AI training centres may require 100 to 500 megawatts continuously; next-generation superclusters may require 500 megawatts to more than one gigawatt.
- Rack consumption
- The article compares traditional racks at about 7–15 kilowatts with modern AI racks requiring roughly 40 kilowatts to more than 100 kilowatts.
- Proposed power mix
- The article suggests considering combinations of grid power, LNG, pipeline gas, nuclear power, solar and wind, depending on availability.