Data Centre Power Infrastructure in the Age of AI: An India Perspective

AI is changing not just how organisations compute, but also what data centres need to deliver. As AI workloads grow, power is becoming as important as compute, connectivity and cooling.

Dense GPU deployments need more power and generate more heat, increasing the need for advanced cooling. In India, the challenge is therefore not just having enough computing capacity, but having the power and cooling infrastructure to support and scale it.

The Shift at a Glance

• More power: AI needs more power in the same space.

• Power + cooling: Higher density means more heat and advanced cooling needs.

• Built to scale: Infrastructure needs enough power and cooling capacity to support growing AI workloads.

Why Power Became the Primary Constraint in AI Infrastructure

Previously, the data centre design was all about computing capabilities, storage and network infrastructure. However, AI is now disrupting this equation. The use of AI accelerates GPU utilisation and requires significantly greater power input compared to classic processors.

As a result, the available power becomes a critical component of AI infrastructure development. A company could have everything it needs, from models and applications to computing capabilities yet lack the ability to develop due to insufficient power infrastructure.

This consideration becomes especially critical for AI data centre power planning. Not only the overall facility capacity should be considered, but the rack and workload one as well.

How AI Workloads Rewrote Power Demand

With an increase in rack power density demand from 10 kW to 150 kW and higher, there is an obvious shift happening at rack level. Enterprise facilities were usually built with far lower rack densities in mind. In AI clusters, many powerful GPUs are housed within a compact area.

That is one of the reasons why rack power density has become so important in designing facilities capable of handling AI workloads. STTGDC Indiahas developed its AI-ready infrastructure with flexible power and cooling that supports racks with densities between 4 kW to150 kW per rack.

Higher density isn't only about supplying more power; it has implications on power distribution, redundancy, floor planning, and thermal management.

Training vs. Inference Power Profiles

The workloads related to AI tasks also differ based on the nature of the workload. While training can result in running the GPUs in clusters consistently in a sustained manner, inference workload can depend on the traffic load.

This becomes important when planning power considerations in an AI data centre. The infrastructure should be designed in such a way that it can accommodate sustained loads, while still having enough capacity for peak loads.

Core Components of AI Data Centre Power Infrastructure

An AI-powered electrical architecture must account for power supply from the utility grid through to the rack level. This will entail considerations on scalability, resiliency, and visibility along the whole power path. The important aspects will involve:

• High-capacity power supplies through utility grid and substations

• Switches and busways that can handle increased load density

• Smart PDUs to monitor and control power at the rack level

• Power continuity through UPSs and backup systems during power disruptions

• Scalable power distributions that will support changing rack configurations

The power systems of STTGDC India have been designed with scalable architecture for supporting dense GPU deployments, without any sacrifice in terms of reliability. Its modular design also facilitates gradual expansion of its IT power capabilities and floor space.

Cooling and Power Are Now One System

More power per rack always means more heat that needs to be dissipated. This makes cooling an integrated part of data centre power infrastructure rather than a separate facility issue.

With increased rack density, traditional air-cooling becomes difficult to implement. Liquid cooling solutions solve this problem through heat removal closer to high-density computing equipment. Solutions like direct-to-chip liquid cooling, rear-door heat exchanger and liquid immersion cooling make it possible to accommodate workloads which put considerably higher thermal load on the facility.

AI-ready facilities of STTGDC India offer support for modern cooling solutions such as In-row Cooling, Rear Door Heat Exchangers, Direct-to-Chip Liquid Cooling and Liquid Immersion Cooling with a rack density capability of up to 120 kW.

The important thing to note is that power and cooling should be planned as one system. Increasing power supply without the ability to dissipate the resulting heat is just transferring the limitation elsewhere.

The Power-Demand Surge

The emergence of AI is happening against the backdrop of growth in India's digital infrastructure ecosystem. Cloud computing, hyperscale needs, enterprise digitisation, and AI computing loads together are raising the demand for secure data centre capacity.

That is why India data centre power demand is becoming a factor that must be considered increasingly. It is not enough just to get more power, but rather to make sure that it will be accessible in the needed location, at the needed scale, and with future-proof capacity.

For the data centres’ operators, it is an essential criterion of site selection and construction. And for enterprises, it becomes an important part of their AI strategy.

Powering AI Sustainably: Renewables, PPAs & Grid Strain

Increased energy demands also pose another question of sustainability. With more energy being demanded due to the use of artificial intelligence in technology, organisations will be forced to consider not only how much energy they consume but also from what source this energy comes.

Availability of grid and grid interconnection data centre needs can influence the schedule of implementation, especially if there is the need for any additional energy or infrastructure. The source of energy can be the renewable energy that can solve the issue of increased consumption of power.

STTGDC India collaborates with renewable energy firms to purchase green energy for their data centres through Power Purchase Agreements, green tariffs and Energy Attribute Certificates. Its sustainability strategy includes green energy purchasing along with energy-efficient programs and smart energy management for AI and high-density racks.

What This Means for Indian Operators & Enterprises

As companies plan their AI projects, power should become part of the infrastructure discussion early on. The relevant questions will include more than just the current demand.

• Power availability: Is there enough capacity available now?

• Scalability: Can power capacity be scaled along with AI loads?

• Rack density: Can the facility house the desired GPU configuration?

• Cooling capacity: Can it handle liquid cooling at higher densities?

• Expansion capacity: Is there enough capacity to add future hardware?

• Efficiency: What is the efficiency with which the facility power capacity is transformed into IT capacity?

• Sustainability: Can enough renewable power be acquired to meet additional environmental goals?

STTGDC India’s AI-ready infrastructure answers these questions with scalable power and cooling distribution, high-density racks and capacity expansion capability. Its data centres have been engineered for a maximum of 150 kW per rack and are geared for future AI and HPC workloads.

The real issue in AI data centres is not how much power the AI installation uses, but whether the facility can deliver it and cool it and scale with the load.

Frequently Asked Questions

How much power does an AI data centre use?

It is impossible to provide an exact number. The energy consumption of an AI data centre depends on the number of GPUs, workloads, utilisation, rack density, cooling, and size of the facilities. Larger AI deployments may need more energy than regular enterprise deployments.

How much power does an AI rack use?

Depending on hardware configuration, different racks may have different energy needs. Some racks may go way beyond the normal density. STTGDC India has AI-capable data centre infrastructure that provides racks up to 150 kW.

Why do AI data centres use more power than traditional ones?

The use of GPUs and accelerated computing for AI applications consumes more energy than CPU-based data centres do. Utilisation of such infrastructure results in increased power consumption.

How do AI workloads affect cooling?

Increase in power density produces more heat. With increasing densities, new cooling techniques such as direct to chip liquid cooling, rear door heat exchangers and immersion cooling become effective at removing the heat generated.

What is a good PUE for an AI data centre?

PUE compares total facility energy to IT energy. Generally, a low PUE suggests high efficiency levels, but it will depend on facility construction, climate, workloads, and the cooling system being used.

Why are grid connections important for AI data centres?

The AI data centres usually need reliable and adequate power connections. Where the grid capacity is limited, additional infrastructure and permits may take time and hence are vital in such cases.

How much will India's data centre power demand grow?

The power needs for India's data centres will increase due to the growth of artificial intelligence, cloud computing and digitisation. The level of increase will depend on a number of factors like adoption, infrastructure development, efficiency improvements and the evolution of computing technologies.