Designing a Data Centre Is Far More Complicated Than Building a Large Building: What XtraNet, Tata Projects, Techno Electric & ESDS Have to Get Right

ayushi ntech — Sun, 06 Sep 2026 09:27:20 +0000

A data centre may look like a large industrial building from the outside.
Inside, however, it is a completely different animal.
A conventional commercial building can tolerate a power interruption, a failed air-conditioning unit or a maintenance shutdown.
A data centre cannot.
A few seconds of downtime can disrupt a bank, cloud platform, government service or enterprise application running across thousands of servers.
That makes designing a data centre an exercise in engineering for failure.
The question is not simply:
How do we build a facility that works?
It is:
How do we design a facility that continues working when individual components fail, equipment needs maintenance, demand suddenly increases, or the workload changes completely?
This is the complexity that companies such as Tata Projects, Techno Electric, XtraNet Technologies and ESDS have to manage when developing or designing data-centre infrastructure.

The first problem: How much power will the data centre actually need? Everything starts with power. Suppose a developer wants to build a 50 MW data centre. That doesn't simply mean arranging a 50 MW electricity connection. The designer has to determine: How much power will the IT equipment consume? How much will cooling consume? How much backup capacity is required? How much additional capacity should be built for future expansion? What happens if one power source fails? Can maintenance happen without shutting down the facility? And AI makes this harder. A traditional enterprise rack might consume several kilowatts. An AI rack can require dramatically more. Recent industry discussions cite potential rack densities ranging from roughly 6 kW to 150 kW, depending on workload. Techno Digital's Chennai facility, for example, is designed for power densities of 10 kW to 50 kW per rack and beyond, illustrating how the physical design changes when facilities are built for high-density AI workloads. So the first design question becomes: What kind of computing will this building eventually contain? Because the answer determines almost everything else.
Redundancy: Designing for failure
This is probably the most important concept in data-centre design.
A data-centre engineer assumes that something will fail.
A transformer may fail.
A UPS may fail.
A generator may fail.
A cooling unit may fail.
A network connection may fail.
A maintenance engineer may need to switch off a piece of equipment.
The facility therefore needs redundancy.
You will hear terms such as:
N+1
If five cooling units are required, install six.
One can fail and the system can continue operating.
2N
Instead of one complete system, provide two independent systems.
If one entire system fails, the second can continue carrying the load.
Concurrent maintainability
The facility should allow critical components to be maintained or replaced without shutting down the IT workload.
ESDS, for example, describes its Tier III facilities as having dual utility feeds, N+N UPS architecture and N+1 generator backup.
The complexity is that redundancy costs money.
More transformers.
More UPS systems.
More generators.
More electrical rooms.
More physical space.
More capital expenditure.
The designer therefore has to balance:
Reliability vs cost vs utilisation.
Power and cooling cannot be designed separately
This is where data-centre engineering becomes particularly interesting.
Almost every unit of electricity consumed by a server ultimately becomes heat.
More computing means:
More electricity → More heat → More cooling
So if the customer suddenly wants to deploy thousands of GPUs, the problem isn't just finding additional electricity.
The facility must also be able to remove the additional heat.
Traditional facilities use systems such as:
CRAC/CRAH units
chillers
cooling towers
air distribution
hot/cold aisle containment
AI workloads are increasingly pushing designers towards:
higher-density cooling
rear-door heat exchangers
direct-to-chip liquid cooling
specialised cooling distribution systems
This is one reason AI is changing the architecture of data centres.
A facility designed around a relatively low-density workload cannot necessarily be converted into a high-density AI facility simply by adding more GPUs.
The power architecture, cooling architecture, floor loading, rack configuration and network design may all need to change.
The building itself has to behave differently
Even the physical building is specialised.
The designer has to think about:
floor loading
rack layouts
equipment movement
ceiling heights
cable trays
electrical rooms
generator yards
fuel storage
cooling equipment
fire compartments
security zones
maintenance access
And there is a hidden constraint:
Space must be designed for future equipment.
A data centre may operate for 15–20 years.
The servers occupying the building in year one may be completely different from those installed five years later.
So designers such as Tata Projects and other infrastructure developers are not merely constructing today's building.
They are trying to create a facility that can accommodate tomorrow's technology.
That is why modularity and scalability have become increasingly important.
The network is another system that cannot afford a single point of failure
Imagine the data centre is operating perfectly.
The servers are running.
The power is available.
The cooling is working.
But the fibre connection is cut.
For the customer, the data centre might as well be offline.
So network architecture has to be designed with the same philosophy as power.
Multiple carriers.
Multiple fibre routes.
Redundant switches.
Independent paths.
Low-latency connections.
Inter-data-centre connectivity.
A data centre therefore isn't an isolated building.
It is a node in a much larger digital network.
This is particularly important for companies such as XtraNet Technologies, whose data-centre work extends beyond physical infrastructure into network management, virtualisation, DC/DR, NOC/SOC and managed IT operations.
Disaster recovery changes the design again
A customer doesn't just ask:
“Is your data centre reliable?”
It may ask:
“What happens if the entire data centre becomes unavailable?”
That introduces disaster recovery.
A second data centre may need to be located sufficiently far away to protect against regional events.
The two facilities may then need:
high-speed connectivity
data replication
backup systems
synchronised infrastructure
recovery procedures
regular disaster-recovery testing
This creates another design problem:
The data centre has to be designed not only as a standalone facility, but as part of a larger infrastructure network.
XtraNet, for example, explicitly offers DC/DR planning, implementation, backup, recovery and continuous monitoring as part of its data-centre services.
Fire and physical security become engineering problems
A data centre contains extremely valuable equipment.
But putting conventional fire-suppression systems around sensitive electronics creates its own challenges.
Designers therefore have to consider:
early smoke detection
fire compartmentalisation
pre-action suppression
fire-rated walls and floors
controlled access
biometric authentication
CCTV
security zones
visitor management
The objective is unusual:
Detect a problem early, isolate it quickly and protect the equipment without damaging the systems you are trying to save.
ESDS, for example, describes VESDA aspirating smoke detection, zoned pre-action suppression and multiple layers of physical access control across its facilities.
Designing for maintenance is almost as important as designing for operation
This is an underrated complexity.
Imagine a transformer needs maintenance.
In an ordinary facility:
Switch it off → service it → switch it back on.
In a data centre:
You cannot simply switch it off.
The workload may be running 24×7.
Therefore the facility has to be designed so that engineers can:
isolate equipment
perform maintenance
replace components
test backup systems
conduct inspections
without interrupting the customer's IT environment.
This is one of the reasons data-centre designs contain so much redundancy.
The goal is not merely:
“Nothing should fail.”
The goal is:
“Something can fail, and we can still keep operating.”
Then comes the IT infrastructure
Once the building is ready, somebody still has to design what goes inside it.
This includes:
Racks → Servers → Storage → Networking → Security → Virtualisation → Monitoring
This is where the roles of companies such as XtraNet and ESDS become particularly relevant.
XtraNet describes capabilities spanning server, storage and network management, virtualisation, cloud integration, facility management and 24×7 monitoring.
ESDS, meanwhile, operates Tier III data centres and combines physical infrastructure with colocation, cloud, managed services and AI infrastructure.
So the design doesn't end when the building is commissioned.
The IT environment has to be designed to run on top of the physical infrastructure.
AI is making the design problem harder
This may be the biggest change underway.
A traditional data centre could largely be designed around a relatively predictable distribution of computing loads.
AI introduces:
Higher rack densities
Larger power requirements
More heat
More demanding networking
Different cooling architectures
Rapidly changing hardware
And there is another complication.
Not every workload needs the same infrastructure.
AI training may favour huge, centralised facilities with enormous computing capacity.
AI inference may benefit from being closer to users and sources of data.
This means the future may not be:
One giant data centre
but a combination of:
Hyperscale campuses + regional facilities + edge data centres.
That makes the job of companies such as Techno Electric, XtraNet, Tata Projects and ESDS increasingly complex.
They are not designing a static building.
They are designing an infrastructure platform that has to accommodate changing computing architectures.
So who does what?
The roles of the companies we are discussing become clearer when viewed from the design perspective.
Tata Projects
Primarily relevant to the large-scale project development and construction side — translating the engineering design into a physical facility.
Techno Electric
Interesting because it is moving from power infrastructure into digital infrastructure and data-centre development, including AI-ready facilities. Its Chennai facility demonstrates the move toward higher-density, flexible infrastructure.
XtraNet Technologies
Sits deeper into the design, integration and operational technology layer — from site assessment and architecture through IT infrastructure, DC/DR, monitoring and managed operations.
ESDS
Combines data-centre ownership/operations with cloud, colocation, managed services and AI infrastructure, giving it exposure across both the physical and digital sides of the facility.
And that is why these companies can all participate in the same data-centre boom while doing very different jobs.
The real challenge: Everything is connected
The most important thing to understand about data-centre design is that there is no isolated decision.
Increase the number of GPUs:
→ power demand rises
→ UPS capacity changes
→ generator requirement changes
→ heat generation rises
→ cooling architecture changes
→ floor and rack configuration changes
→ network requirements change
→ electrical rooms may need to expand
→ capital cost rises
→ operating cost changes.
A seemingly simple decision at the rack level can therefore affect the entire facility.
That is what makes data-centre engineering so difficult.
The data centre is becoming an engineered ecosystem
A modern data centre is simultaneously:
A building
A power plant
A cooling system
A telecommunications hub
A security facility
A computing platform
And increasingly, an AI factory
That is why designing one requires far more than construction expertise.
It requires the ability to bring together power, mechanical systems, civil engineering, networking, IT infrastructure, security, cloud architecture and operations into one system where almost nothing can be allowed to fail.
And as AI pushes power densities higher and computing architectures change faster, the biggest challenge may no longer be simply building more data centres.
It will be building data centres that can adapt to the next generation of computing without having to be rebuilt from scratch.

DEV Community: ayushi ntech

Designing a Data Centre Is Far More Complicated Than Building a Large Building: What XtraNet, Tata Projects, Techno Electric & ESDS Have to Get Right