By Tech Innovators
September 2026For most people, artificial intelligence lives on a screen.
Ask a chatbot a question. Generate an image. Run an AI agent. Train a model.
What remains largely invisible is what happens behind that screen.
Thousands of accelerators operate inside buildings consuming enormous amounts of electricity. Those processors generate heat that must be removed continuously. Power has to reach the facility reliably. Cooling systems have to operate around the clock. Fibre has to connect the machines. Backup systems have to be available when the grid is not.
And as India moves from consuming artificial intelligence to building its own AI infrastructure, this physical layer is becoming impossible to ignore.
India's AI race is increasingly becoming an infrastructure race.
The country has spent the past several years trying to make computing power more accessible. Under the IndiaAI Mission, more than 38,000 GPUs had been onboarded into the national common compute framework by March 2026.
But GPUs are only one part of the equation.
The larger question is:
Where will India get the physical infrastructure required to keep tens of thousands—and eventually potentially hundreds of thousands—of AI accelerators running?
The answer could determine how far India's AI ambitions can actually scale.
From 375 MW to a Multi-Gigawatt Industry
India's data-centre expansion has already been extraordinary.
Government figures show that installed data-centre capacity increased from approximately 375 MW in 2020 to about 1,575 MW by August 2026.
That is more than a fourfold increase in roughly six years.
And the next expansion could be substantially larger.
JLL estimates India's data-centre capacity could rise from approximately 1.6 GW in mid-2026 to 6 GW by 2029.
That would mean adding more than four gigawatts of capacity in only a few years.
The capital required is equally significant.
JLL estimates that India's data-centre expansion could require approximately $110 billion of investment by 2029, spanning data-centre real estate, power infrastructure, cooling technology and IT equipment.
Artificial intelligence is accelerating this transition.
Traditional cloud computing already requires large data centres. AI changes the density of the infrastructure inside them.
A rack filled with conventional servers is one thing.
A rack filled with high-performance AI accelerators is another.
The more computing power concentrated into a small physical area, the greater the challenge of delivering electricity and removing heat.
This changes the economics of the data centre.
The question is no longer simply:
How many servers can we install?
It becomes:
How much computing power can we reliably support per square metre?
India Is Rapidly Building the Compute Layer
India clearly understands that access to computing infrastructure has become strategically important.
The IndiaAI Mission was approved with an outlay of approximately ₹10,372 crore, with compute infrastructure forming one of its central pillars.
By March 2026, more than 38,000 GPUs had been onboarded through the IndiaAI compute framework.
The government has also announced plans to expand that pool further.
This represents an important shift.
For years, one of the biggest barriers facing Indian AI startups and researchers was straightforward: high-performance computing was expensive.
Training and deploying increasingly capable models requires access to hardware that many small companies and academic teams cannot afford to purchase independently.
Shared national compute infrastructure attempts to reduce that barrier.
But democratising compute creates another challenge.
Every additional accelerator eventually becomes an electricity load somewhere.
The AI economy may appear digital, but underneath it sits an increasingly industrial-scale physical system.
A 1-GW AI Campus Changes the Conversation
The scale of projects now being proposed shows where the market is heading.
In September 2026, a Tata Consultancy Services subsidiary and partners announced plans for an AI data-centre campus in Telangana with planned capacity of up to 1 GW, involving investment of as much as ₹700 billion.
One gigawatt is not simply another technology campus.
It is infrastructure at power-system scale.
And this is unlikely to remain an isolated phenomenon.
India is attracting large commitments from hyperscalers and infrastructure providers as companies race to serve cloud computing, AI inference, model training, sovereign computing and enterprise workloads.
The geography of the industry is consequently expanding as well.
Mumbai and Navi Mumbai have historically dominated India's data-centre market, alongside major clusters around Chennai, Hyderabad, Bengaluru and Delhi NCR.
Government data now identifies states including Andhra Pradesh, Madhya Pradesh, Chhattisgarh and West Bengal as emerging investment destinations.
The location decision is becoming strategic.
A future AI data centre cannot simply ask:
Is fibre available?
It must also ask:
Is sufficient electricity available?
Can the grid support high-density demand?
What renewable power can be contracted?
How much water will cooling require?
What is the local climate?
How quickly can transmission infrastructure be expanded?
The data-centre map of India could therefore increasingly become an energy map.
AI Is Turning Electricity Into a Technology Input
Software companies traditionally thought about electricity as an operating expense.
AI infrastructure is changing that relationship.
Electricity is becoming one of the fundamental inputs into computing.
The International Energy Agency estimates global data-centre electricity consumption could rise to roughly 945 TWh by 2030, more than double current levels.
AI is expected to be the most important driver of that increase.
Electricity consumption from accelerated servers—the machines primarily responsible for modern AI workloads—is projected to grow much faster than consumption from conventional servers.
There is also a mismatch in development timelines.
A data centre can potentially be brought online within a few years.
Large-scale electricity infrastructure, transmission systems and generation projects can take considerably longer.
That difference matters.
You can order GPUs.
You can raise capital for a data-centre campus.
You cannot instantly create a resilient power grid around it.
This is why electricity availability is emerging globally as one of the constraints on AI infrastructure development.
India will not be exempt.
But India Is Also Building Power at Extraordinary Speed
It would be wrong, however, to portray India as simply running toward an electricity shortage.
India is simultaneously undergoing one of the world's largest energy expansions.
By the end of August 2026, the country had approximately 304 GW of non-fossil electricity capacity, according to the Ministry of New and Renewable Energy.
Solar alone had reached approximately 168 GW.
India is working toward 500 GW of non-fossil capacity by 2030.
The International Energy Agency expects Indian electricity demand to grow at an average rate of approximately 6.4% annually through 2030.
Around half of that additional electricity demand is forecast to be met by solar PV, with coal supplying roughly another quarter and wind, nuclear, hydro and gas contributing the remainder.
That means India's AI infrastructure expansion is occurring inside a much larger transformation of the country's power system.
Data centres are not the only new consumers.
India is simultaneously electrifying transport, expanding manufacturing, deploying air conditioning to hundreds of millions of consumers and developing new industrial capacity.
AI therefore has to compete for infrastructure inside an economy whose overall electricity demand is already rising rapidly.
Renewable Energy Solves Only Part of the Problem
There is an obvious response:
Power AI with renewable energy.
And renewables will undoubtedly play a major role.
Data-centre operators globally are increasingly signing long-term renewable power agreements, while India's rapidly expanding solar and wind capacity provides an enormous opportunity for cleaner computing.
But a data centre cannot operate only when the sun shines.
AI infrastructure requires reliable electricity every second of the day.
That means renewable generation must be combined with some mixture of storage, transmission, grid power and other firm sources.
This distinction is important.
Having sufficient annual renewable generation is not the same as having sufficient electricity available at a particular data-centre location at every hour of the year.
The AI infrastructure challenge therefore becomes partly a grid flexibility challenge.
The winners may be regions capable of combining abundant renewable energy with storage, strong transmission infrastructure, reliable grids, fibre connectivity and suitable land.
That could reshape where India's next generation of data centres is built.
Then Comes the Heat
Electricity entering a processor eventually becomes heat.
As computing density increases, removing that heat becomes increasingly difficult.
Traditional data centres have relied heavily on air cooling.
High-density AI racks are pushing the industry toward technologies such as direct-to-chip liquid cooling, immersion cooling, adiabatic systems and closed-loop liquid cooling.
The Indian government has already acknowledged this shift.
In August 2026, the Ministry of Electronics and Information Technology noted that the industry was adopting advanced cooling systems as AI and high-performance computing increased infrastructure requirements.
This is not a minor engineering detail.
Cooling determines how densely computing equipment can operate, how much electricity a facility consumes and, depending on the system, how much water it requires.
India's climate makes this especially interesting.
A cooling architecture suitable for a data centre in Scandinavia cannot simply be copied into Hyderabad or Chennai.
Ambient temperature, humidity and water availability change the equation.
AI infrastructure must therefore increasingly be engineered for geography.
The Water Question Cannot Be Ignored
Water is likely to become one of the most sensitive parts of the AI infrastructure conversation.
Certain cooling systems can consume substantial amounts of water, creating potential tension when large data centres are built in regions already facing water stress.
The Indian government has acknowledged the issue.
Environmental appraisal rules can require large building projects to assess freshwater availability, water balance, greywater generation, recycling and reuse.
Meanwhile, the government says data-centre operators are increasingly adopting technologies designed to reduce water consumption.
Closed-loop cooling is particularly important because water can circulate through a system rather than continuously being consumed.
But the broader question remains:
Should every region that can host a data centre host one?
In the AI era, site selection will increasingly need to account for energy and water availability alongside connectivity and economics.
That may force policymakers to think about data centres more like industrial infrastructure than conventional commercial real estate.
Efficiency Could Be India's Hidden Advantage
There is another side to this equation.
The future of AI infrastructure does not have to be solved only by generating more electricity.
It can also be solved by using computing resources more efficiently.
Better chips can deliver more computation per watt.
Smaller specialised models can perform certain tasks without requiring enormous general-purpose models.
Quantisation can reduce model size.
Workload scheduling can move non-urgent computation toward periods when electricity is cheaper or cleaner.
Advanced cooling can reduce overhead.
Software optimisation can improve accelerator utilisation.
Inference can increasingly move to edge devices instead of sending every request to a central cloud.
This matters enormously for India.
The country may not need to reproduce the most power-intensive version of the American AI infrastructure model.
It could instead compete on efficient AI.
For a country serving more than a billion people, lowering the compute required per useful AI task could be just as strategically valuable as adding more GPUs.
The AI Race Is Becoming a Systems-Engineering Race
The first phase of the AI boom created a simple hierarchy.
Who has the best model?
Then came another question.
Who has the most GPUs?
The next question may be harder:
Who can build the complete system around those GPUs?
That system includes:
Semiconductors.
Servers.
Data centres.
Electricity generation.
Transmission.
Energy storage.
Cooling.
Water management.
Fibre networks.
Cybersecurity.
Capital.
Software efficiency.
And the engineers capable of making all of those layers work together.
This is why the AI race is gradually becoming less like a conventional software competition and more like an industrial infrastructure competition.
India Has an Opportunity to Design Differently
India is entering this infrastructure cycle later than the United States and China.
That sounds like a disadvantage.
It can also be an opportunity.
Much of India's future AI capacity has not yet been constructed.
New facilities can therefore be designed around high-density AI workloads rather than retrofitting infrastructure originally designed for conventional cloud computing.
New data-centre clusters can be planned around renewable generation.
Liquid cooling can be incorporated from the beginning.
Water recycling can be designed into facilities rather than added later.
Grid infrastructure can be planned alongside digital infrastructure.
AI workloads can be optimised for energy efficiency.
And state governments competing for data-centre investment can begin evaluating projects not merely by their investment value, but by their long-term demands on electricity and water systems.
India does not simply need more data centres.
It needs better-designed AI infrastructure.
The Real AI Stack Starts Below the GPU
India's AI ambition is often discussed through models, startups, talent and semiconductors.
Those are essential.
But beneath all of them sits another stack:
Land → Energy → Grid → Cooling → Compute → Models → Applications.
If one layer cannot scale, everything above it eventually encounters a limit.
The encouraging part is that India is simultaneously investing in several pieces of this puzzle: renewable generation is expanding rapidly, national AI compute capacity is growing, private investment in data centres is accelerating and new cooling technologies are entering the market.
The difficult part is coordination.
Building a powerful AI economy will require technology policy and energy policy to increasingly become part of the same conversation.
Because the defining constraint on India's AI ambitions may not ultimately be whether the country can acquire enough GPUs.
It may be whether India can build the physical systems capable of keeping those GPUs running—reliably, affordably and sustainably.
The cloud, after all, has always lived somewhere.
In the AI era, what sits underneath it matters more than ever.
Key Numbers
38,000+ — GPUs onboarded under the IndiaAI common compute framework by March 2026.
~1.58 GW — India's installed data-centre capacity as of August 2026.
6 GW — JLL's projection for Indian data-centre capacity by 2029.
~$110 billion — estimated investment associated with India's data-centre expansion through 2029.
304 GW — India's total installed non-fossil electricity capacity as of August 31, 2026.
168 GW — installed solar capacity as of August 31, 2026.
6.4% — IEA forecast average annual growth in India's electricity demand through 2030.
Editorial Sources
Government of India — Ministry of Electronics & Information Technology / Press Information Bureau
IndiaAI Mission / Principal Scientific Adviser, Government of IndiaMinistry of New and Renewable Energy, Government of IndiaInternational Energy Agency — Energy and AI and Electricity 2026JLL India — India Data Centre 2026 Mid-Year Report
