/images/ppid_59c68470-image-178465504254290803.webp)
/images/ppid_a911dc6a-image-178482502578231579.webp)
/images/ppid_59c68470-image-178482505654775630.webp)
What is the story about?
For the past few years, the artificial intelligence conversation has largely revolved around one question: Which AI model is the smartest?
ChatGPT, Gemini and Claude have become household names, while companies and investors have focused on the race to build increasingly capable models.
But the next phase of the AI race may be less about the model itself and more about the physical infrastructure required to build and run it.
That infrastructure includes data centers, GPUs, computing capacity, power, cooling systems and the capital required to build them at scale.
The reason this matters is simple: countries that have access to large-scale computing capacity will have a greater ability to train, host and deploy AI systems.
India is now entering what industry leaders describe as an AI infrastructure supercycle.
The country's data center capacity has grown from around 200 megawatts seven or eight years ago to approximately 1.6 gigawatts today. Over the next five to six years, this capacity could rise to around 8 gigawatts, according to Sunil Gupta, Managing Director of Yotta Data Services.
That would represent a several-fold expansion of India's digital infrastructure in a relatively short period, requiring large investments not just in data-centre buildings, but also in GPUs, power connections, cooling systems and supporting infrastructure.
So, what is driving this rapid expansion? What does AI infrastructure actually include? What could limit India's growth? And why does it matter for the country's technology ambitions?
What exactly is AI infrastructure?
At the heart of the AI infrastructure opportunity is a simple fact: artificial intelligence requires enormous computing power.
The AI ecosystem can broadly be understood as a stack.
At the bottom is the physical infrastructure. This includes data centers, power, cooling systems and the computing equipment housed inside them.
Traditional data centers were largely built around CPU-based servers used for enterprise applications, internet services and cloud computing. AI has changed the nature of that demand.
Training and running large AI models requires GPU-based computing systems. GPUs can perform the large number of parallel calculations needed for many AI workloads, but they also consume significant amounts of power and generate considerable heat.
That increases the need for high-density data centers, specialised cooling systems and reliable power infrastructure.
Above this physical layer sits the data and model layer.
Large quantities of data are processed and used to train AI models. These models then become the foundation for applications that can be used by consumers, businesses and governments.
The final layer is the application layer. This is where users interact with AI through chatbots, enterprise software, copilots and other applications.
India has traditionally been strong in the upper layers of this stack, particularly software development, technology services and business applications.
The opportunity now is to build more of the underlying infrastructure as well.
Why is the AI infrastructure boom happening now?
Data centers have existed for decades. Yotta has been in the business for around three decades, while Sify has operated in the sector for about 25 years.
What has changed is the speed at which computing demand is growing.
According to Sharad Agarwal, CEO of Sify Data Center, the data center industry has evolved through several technology cycles.
In the early 2000s, data centers helped businesses scale the internet and enterprise networks. In the following years, they supported enterprise resource planning systems, data analytics, digital transformation and cloud computing.
The rise of AI has now created a new phase of demand.
"The time to scale has consistently changed. The time to double has consistently changed. And now, because it's changing so fast, it's doubling so fast, it is important that we have this conversation now," Agarwal said.
The underlying change is that AI requires computing capacity at a much greater scale and speed than many earlier technology cycles.
Companies need infrastructure not only to train AI models, but also to run them after they are built. This latter process, known as inference, takes place every time an AI system generates an answer, recommendation, prediction or other output.
As AI applications become more widely used, demand for inference capacity can grow rapidly.
That creates simultaneous demand for GPUs, data centers, power and cooling infrastructure.
From 200 MW to a projected 8 GW: How big is the opportunity?
India's data center industry has already expanded sharply.
Gupta said the country's data center capacity stood at only around 200 MW seven or eight years ago. Today, it is around 1.6 GW.
The next phase could be even more dramatic.
Based on the investments being planned by major companies and the demand signals emerging from the market, Gupta expects India's data center capacity to reach approximately 8 GW over the next five or six years.
The significance of that projection is not just the headline number. It points to a rapid expansion of the physical infrastructure needed to support India's digital economy and AI ambitions.
Building that capacity would require large investments in servers, GPUs, land, power connections, cooling systems and network infrastructure.
It would also require companies to make large capital commitments to a technology environment in which computing hardware is evolving rapidly.
The expansion is being driven by a combination of cloud computing, digital services and, increasingly, AI workloads.
The scale of the opportunity is also changing India's position in the global technology ecosystem.
For decades, India was primarily a consumer of technology created elsewhere. However, Gupta said the country's data center infrastructure is now being used to serve customers outside India.
"For the first time, I am seeing that, sitting in India, I am not able to give my compute, GPU compute, to Indian use cases only. Right now, US and European customers are coming and consuming my compute, which is hosted in India," he said.
That represents a significant shift.
India is not only building infrastructure for its own AI demand. It could also become a global provider of computing capacity.
Why data centers alone will not make India an AI power
Building data centers is only one part of the AI opportunity.
The infrastructure must ultimately be used to create models, software and applications.
According to Gupta, India's software and technology industry has a strong advantage at the application layer.
Indian IT companies have spent decades building software and technology solutions for global businesses. That experience could now be applied to AI.
The challenge, however, will be to adapt.
Instead of focusing only on traditional software development and coding, companies will increasingly need to build AI-based products and applications.
India also has a growing pool of companies working on AI models, enterprise applications and sector-specific use cases.
Ankush Sabharwal, CEO and CTO of CoRover.ai, argued that the country already has much of the technology and infrastructure required to build solutions that can address business and societal challenges.
The next step, therefore, is not simply to build more computing capacity. India must also convert that capacity into products and applications that can be used at scale.
What does sovereign AI actually mean?
The AI infrastructure debate is also closely linked to national sovereignty.
AI systems depend on several different elements: data, models, computing hardware, software and the infrastructure used to train and run the systems.
These elements do not necessarily have to be controlled by the same country.
For example, having a data center in India does not automatically make an AI system sovereign. Sovereignty also depends on who controls the data, the model, the chips, the software stack and the infrastructure used to train and run the system.
The debate is important because AI models can be trained on large amounts of information relating to people, businesses, industries and societies.
Gupta argued that India cannot afford to remain only a consumer of technology when its own data is being used to create valuable intellectual property.
"If AI is percolating into our personal lives and professional lives across industries, the fundamental infrastructure to build that, to host that, to train those AI models and to use those AI models for various use cases in our lives becomes fundamentally important," he said.
The argument for building AI infrastructure in India is therefore not simply about constructing more data centers.
It is also about creating the ability to train models using Indian data, build AI systems for Indian requirements and develop capabilities that are less dependent on foreign infrastructure and foreign models.
This could include AI models and agents that understand India's languages, culture, industries and public-sector requirements.
The broader objective is to ensure that India has greater control over the infrastructure and technology needed for its AI future.
Is capital the biggest bottleneck?
AI infrastructure is extremely capital-intensive.
GPU servers, high-density data centers, specialised cooling systems and power infrastructure require large upfront investments.
Gupta said Yotta has already invested around $4 billion and plans to invest another $4 billion within the current financial year.
The scale of investment required creates a financing challenge.
The technology is evolving rapidly, which means investors and lenders must assess not only the demand for computing capacity but also the risk that today's hardware could be replaced by more powerful technology within a relatively short period.
That creates challenges around:
According to Gupta, the sector requires more risk capital, including private equity and other investors with a greater appetite for technology-related risks.
The question is therefore not only whether India can attract enough capital to build infrastructure. It is also whether the financing structure can keep pace with the speed at which AI technology is changing.
As the AI infrastructure supercycle gathers momentum, access to capital could determine how quickly India can build capacity.
What does an 8 GW data-centre industry mean for India's power system?
The rapid expansion of AI infrastructure has raised concerns about electricity demand.
Large data centers consume significant amounts of power, and AI workloads are particularly energy-intensive.
However, the question is more complex than simply comparing India's total installed power generation capacity with projected data-centre demand.
For data centers, the critical requirements are reliable, uninterrupted and high-quality power at the location where the facility is being built.
That means transmission capacity, grid connectivity and the ability to secure power at the right location can be just as important as the country's total installed generation capacity.
Agarwal does not see power generation itself as a major constraint for India's data center growth.
India's installed power generation capacity is more than 550 GW, while peak demand in May was around 270 GW.
Even if data-centre capacity rises to 8 GW, Agarwal said the sector would account for a relatively small share of India's overall power ecosystem.
India is also expanding renewable energy capacity, with solar and wind accounting for a significant share of new generation capacity.
The government's growing focus on nuclear power could further add to the country's long-term energy mix.
This means the data-centre industry believes India has the overall energy potential to support a significant expansion of AI infrastructure.
The more immediate challenge could be ensuring that sufficient reliable power reaches the specific locations where large data centers are being developed.
Why water consumption could be different in India
Water consumption is another major concern globally.
Many data centers in the West use evaporative cooling systems, which can consume substantial amounts of water.
Agarwal said the Indian data-centre industry has largely adopted closed-loop water chiller systems instead.
According to him, around 70% of data-centre capacity globally uses evaporative cooling. In India, however, almost 100% of the industry uses closed-loop water chiller systems.
The difference in water consumption can be substantial, although the actual requirement depends on the cooling technology, climate, facility design and the way water use is measured.
Agarwal said evaporative cooling can require around 3 million gallons of water per megawatt every year, while closed-loop systems use less than 1,000 litres per megawatt annually.
These figures represent the industry executive's comparison of the two cooling approaches and should be understood in the context of the specific systems being compared.
The broader point is that India's choice of cooling technology could influence how quickly water becomes a constraint as data-centre capacity expands.
The industry believes that the widespread use of closed-loop systems could reduce water consumption compared with some evaporative cooling models used in other markets.
Why speed could determine who wins
The opportunity is large, but so is the competition.
AI infrastructure is being built globally, and countries are competing to attract investment, computing capacity and technology companies.
For India, the speed of execution could become a critical factor.
Agarwal identified the ability to build infrastructure quickly as one of the country's biggest challenges.
"Ability to execute and build the infrastructure in the fastest possible manner. So, I would say speed," he said.
The need for speed is particularly important because AI demand is evolving rapidly.
By the time a large infrastructure project is completed, the technology and computing requirements may have already changed.
India therefore needs to expand capacity while also ensuring that its infrastructure remains flexible enough to accommodate new generations of chips and AI systems.
This creates a difficult balancing act.
Companies need to build quickly enough to capture demand, but they also need to avoid locking themselves into infrastructure designed for technology that may become outdated.
The bigger opportunity for India
The AI infrastructure supercycle is ultimately about more than data centers.
It represents a potential shift in India's position in the global technology economy.
The country has already built significant capabilities in software, IT services and digital applications. The next step is to build the physical infrastructure that supports AI at scale.
That means data centers with high-density GPU capacity, access to reliable and increasingly renewable power, specialised cooling systems and sufficient capital.
It also means building models and applications that use Indian data and solve Indian problems.
The transition from around 200 MW of data-centre capacity seven or eight years ago to approximately 1.6 GW today, with the possibility of reaching 8 GW over the next five or six years, shows the scale and speed of the opportunity.
The next phase of the AI race may therefore not be decided only by which company builds the smartest model.
It could also be decided by which countries build enough computing capacity to train, host and run those models, while developing the talent, applications and sovereign capabilities needed to turn that infrastructure into economic value.
For India, the AI infrastructure supercycle has already begun.
The question now is whether the country can build fast enough, raise enough capital, secure reliable power and cooling, and develop enough sovereign AI capability to turn its infrastructure expansion into a long-term competitive advantage.
ChatGPT, Gemini and Claude have become household names, while companies and investors have focused on the race to build increasingly capable models.
But the next phase of the AI race may be less about the model itself and more about the physical infrastructure required to build and run it.
That infrastructure includes data centers, GPUs, computing capacity, power, cooling systems and the capital required to build them at scale.
The reason this matters is simple: countries that have access to large-scale computing capacity will have a greater ability to train, host and deploy AI systems.
India is now entering what industry leaders describe as an AI infrastructure supercycle.
The country's data center capacity has grown from around 200 megawatts seven or eight years ago to approximately 1.6 gigawatts today. Over the next five to six years, this capacity could rise to around 8 gigawatts, according to Sunil Gupta, Managing Director of Yotta Data Services.
That would represent a several-fold expansion of India's digital infrastructure in a relatively short period, requiring large investments not just in data-centre buildings, but also in GPUs, power connections, cooling systems and supporting infrastructure.
So, what is driving this rapid expansion? What does AI infrastructure actually include? What could limit India's growth? And why does it matter for the country's technology ambitions?
What exactly is AI infrastructure?
At the heart of the AI infrastructure opportunity is a simple fact: artificial intelligence requires enormous computing power.
The AI ecosystem can broadly be understood as a stack.
At the bottom is the physical infrastructure. This includes data centers, power, cooling systems and the computing equipment housed inside them.
Traditional data centers were largely built around CPU-based servers used for enterprise applications, internet services and cloud computing. AI has changed the nature of that demand.
Training and running large AI models requires GPU-based computing systems. GPUs can perform the large number of parallel calculations needed for many AI workloads, but they also consume significant amounts of power and generate considerable heat.
That increases the need for high-density data centers, specialised cooling systems and reliable power infrastructure.
Above this physical layer sits the data and model layer.
Large quantities of data are processed and used to train AI models. These models then become the foundation for applications that can be used by consumers, businesses and governments.
The final layer is the application layer. This is where users interact with AI through chatbots, enterprise software, copilots and other applications.
India has traditionally been strong in the upper layers of this stack, particularly software development, technology services and business applications.
The opportunity now is to build more of the underlying infrastructure as well.
Why is the AI infrastructure boom happening now?
Data centers have existed for decades. Yotta has been in the business for around three decades, while Sify has operated in the sector for about 25 years.
What has changed is the speed at which computing demand is growing.
According to Sharad Agarwal, CEO of Sify Data Center, the data center industry has evolved through several technology cycles.
In the early 2000s, data centers helped businesses scale the internet and enterprise networks. In the following years, they supported enterprise resource planning systems, data analytics, digital transformation and cloud computing.
The rise of AI has now created a new phase of demand.
"The time to scale has consistently changed. The time to double has consistently changed. And now, because it's changing so fast, it's doubling so fast, it is important that we have this conversation now," Agarwal said.
The underlying change is that AI requires computing capacity at a much greater scale and speed than many earlier technology cycles.
Companies need infrastructure not only to train AI models, but also to run them after they are built. This latter process, known as inference, takes place every time an AI system generates an answer, recommendation, prediction or other output.
As AI applications become more widely used, demand for inference capacity can grow rapidly.
That creates simultaneous demand for GPUs, data centers, power and cooling infrastructure.
From 200 MW to a projected 8 GW: How big is the opportunity?
India's data center industry has already expanded sharply.
Gupta said the country's data center capacity stood at only around 200 MW seven or eight years ago. Today, it is around 1.6 GW.
The next phase could be even more dramatic.
Based on the investments being planned by major companies and the demand signals emerging from the market, Gupta expects India's data center capacity to reach approximately 8 GW over the next five or six years.
The significance of that projection is not just the headline number. It points to a rapid expansion of the physical infrastructure needed to support India's digital economy and AI ambitions.
Building that capacity would require large investments in servers, GPUs, land, power connections, cooling systems and network infrastructure.
It would also require companies to make large capital commitments to a technology environment in which computing hardware is evolving rapidly.
The expansion is being driven by a combination of cloud computing, digital services and, increasingly, AI workloads.
The scale of the opportunity is also changing India's position in the global technology ecosystem.
For decades, India was primarily a consumer of technology created elsewhere. However, Gupta said the country's data center infrastructure is now being used to serve customers outside India.
"For the first time, I am seeing that, sitting in India, I am not able to give my compute, GPU compute, to Indian use cases only. Right now, US and European customers are coming and consuming my compute, which is hosted in India," he said.
That represents a significant shift.
India is not only building infrastructure for its own AI demand. It could also become a global provider of computing capacity.
Why data centers alone will not make India an AI power
Building data centers is only one part of the AI opportunity.
The infrastructure must ultimately be used to create models, software and applications.
According to Gupta, India's software and technology industry has a strong advantage at the application layer.
Indian IT companies have spent decades building software and technology solutions for global businesses. That experience could now be applied to AI.
The challenge, however, will be to adapt.
Instead of focusing only on traditional software development and coding, companies will increasingly need to build AI-based products and applications.
India also has a growing pool of companies working on AI models, enterprise applications and sector-specific use cases.
Ankush Sabharwal, CEO and CTO of CoRover.ai, argued that the country already has much of the technology and infrastructure required to build solutions that can address business and societal challenges.
The next step, therefore, is not simply to build more computing capacity. India must also convert that capacity into products and applications that can be used at scale.
What does sovereign AI actually mean?
The AI infrastructure debate is also closely linked to national sovereignty.
AI systems depend on several different elements: data, models, computing hardware, software and the infrastructure used to train and run the systems.
These elements do not necessarily have to be controlled by the same country.
For example, having a data center in India does not automatically make an AI system sovereign. Sovereignty also depends on who controls the data, the model, the chips, the software stack and the infrastructure used to train and run the system.
The debate is important because AI models can be trained on large amounts of information relating to people, businesses, industries and societies.
Gupta argued that India cannot afford to remain only a consumer of technology when its own data is being used to create valuable intellectual property.
"If AI is percolating into our personal lives and professional lives across industries, the fundamental infrastructure to build that, to host that, to train those AI models and to use those AI models for various use cases in our lives becomes fundamentally important," he said.
The argument for building AI infrastructure in India is therefore not simply about constructing more data centers.
It is also about creating the ability to train models using Indian data, build AI systems for Indian requirements and develop capabilities that are less dependent on foreign infrastructure and foreign models.
This could include AI models and agents that understand India's languages, culture, industries and public-sector requirements.
The broader objective is to ensure that India has greater control over the infrastructure and technology needed for its AI future.
Is capital the biggest bottleneck?
AI infrastructure is extremely capital-intensive.
GPU servers, high-density data centers, specialised cooling systems and power infrastructure require large upfront investments.
Gupta said Yotta has already invested around $4 billion and plans to invest another $4 billion within the current financial year.
The scale of investment required creates a financing challenge.
The technology is evolving rapidly, which means investors and lenders must assess not only the demand for computing capacity but also the risk that today's hardware could be replaced by more powerful technology within a relatively short period.
That creates challenges around:
- the cost of financing;
- the useful life of GPU hardware;
- the ability to maintain high utilisation rates;
- the stability of customer contracts; and
- the returns generated on large capital investments.
According to Gupta, the sector requires more risk capital, including private equity and other investors with a greater appetite for technology-related risks.
The question is therefore not only whether India can attract enough capital to build infrastructure. It is also whether the financing structure can keep pace with the speed at which AI technology is changing.
As the AI infrastructure supercycle gathers momentum, access to capital could determine how quickly India can build capacity.
What does an 8 GW data-centre industry mean for India's power system?
The rapid expansion of AI infrastructure has raised concerns about electricity demand.
Large data centers consume significant amounts of power, and AI workloads are particularly energy-intensive.
However, the question is more complex than simply comparing India's total installed power generation capacity with projected data-centre demand.
For data centers, the critical requirements are reliable, uninterrupted and high-quality power at the location where the facility is being built.
That means transmission capacity, grid connectivity and the ability to secure power at the right location can be just as important as the country's total installed generation capacity.
Agarwal does not see power generation itself as a major constraint for India's data center growth.
India's installed power generation capacity is more than 550 GW, while peak demand in May was around 270 GW.
Even if data-centre capacity rises to 8 GW, Agarwal said the sector would account for a relatively small share of India's overall power ecosystem.
India is also expanding renewable energy capacity, with solar and wind accounting for a significant share of new generation capacity.
The government's growing focus on nuclear power could further add to the country's long-term energy mix.
This means the data-centre industry believes India has the overall energy potential to support a significant expansion of AI infrastructure.
The more immediate challenge could be ensuring that sufficient reliable power reaches the specific locations where large data centers are being developed.
Why water consumption could be different in India
Water consumption is another major concern globally.
Many data centers in the West use evaporative cooling systems, which can consume substantial amounts of water.
Agarwal said the Indian data-centre industry has largely adopted closed-loop water chiller systems instead.
According to him, around 70% of data-centre capacity globally uses evaporative cooling. In India, however, almost 100% of the industry uses closed-loop water chiller systems.
The difference in water consumption can be substantial, although the actual requirement depends on the cooling technology, climate, facility design and the way water use is measured.
Agarwal said evaporative cooling can require around 3 million gallons of water per megawatt every year, while closed-loop systems use less than 1,000 litres per megawatt annually.
These figures represent the industry executive's comparison of the two cooling approaches and should be understood in the context of the specific systems being compared.
The broader point is that India's choice of cooling technology could influence how quickly water becomes a constraint as data-centre capacity expands.
The industry believes that the widespread use of closed-loop systems could reduce water consumption compared with some evaporative cooling models used in other markets.
Why speed could determine who wins
The opportunity is large, but so is the competition.
AI infrastructure is being built globally, and countries are competing to attract investment, computing capacity and technology companies.
For India, the speed of execution could become a critical factor.
Agarwal identified the ability to build infrastructure quickly as one of the country's biggest challenges.
"Ability to execute and build the infrastructure in the fastest possible manner. So, I would say speed," he said.
The need for speed is particularly important because AI demand is evolving rapidly.
By the time a large infrastructure project is completed, the technology and computing requirements may have already changed.
India therefore needs to expand capacity while also ensuring that its infrastructure remains flexible enough to accommodate new generations of chips and AI systems.
This creates a difficult balancing act.
Companies need to build quickly enough to capture demand, but they also need to avoid locking themselves into infrastructure designed for technology that may become outdated.
The bigger opportunity for India
The AI infrastructure supercycle is ultimately about more than data centers.
It represents a potential shift in India's position in the global technology economy.
The country has already built significant capabilities in software, IT services and digital applications. The next step is to build the physical infrastructure that supports AI at scale.
That means data centers with high-density GPU capacity, access to reliable and increasingly renewable power, specialised cooling systems and sufficient capital.
It also means building models and applications that use Indian data and solve Indian problems.
The transition from around 200 MW of data-centre capacity seven or eight years ago to approximately 1.6 GW today, with the possibility of reaching 8 GW over the next five or six years, shows the scale and speed of the opportunity.
The next phase of the AI race may therefore not be decided only by which company builds the smartest model.
It could also be decided by which countries build enough computing capacity to train, host and run those models, while developing the talent, applications and sovereign capabilities needed to turn that infrastructure into economic value.
For India, the AI infrastructure supercycle has already begun.
The question now is whether the country can build fast enough, raise enough capital, secure reliable power and cooling, and develop enough sovereign AI capability to turn its infrastructure expansion into a long-term competitive advantage.
/images/ppid_59c68470-image-178482753023531365.webp)
/images/ppid_59c68470-image-178482755840818614.webp)
/images/ppid_59c68470-image-178482752886187612.webp)
