India’s data center industry is entering a cycle where adding capacity alone no longer defines readiness. AI is changing how operators approach compute, power, cooling, connectivity and location. Higher rack densities are forcing facilities to reconsider thermal systems, electrical distribution and structural requirements. AI workloads are also spreading across hyperscalers, colocation facilities, sovereign environments, enterprises and edge infrastructure. At the ET National Data Centers Summit 2026 in New Delhi, executives from Digital Edge India, NTT GDC, Equinix, Prasa and IBM India outlined those forces. Their comments point to a market where adaptability joins capacity across India’s rapidly expanding digital economy.

AI Infrastructure Is Redefining Capacity Planning

Sairam Prasad, chief executive officer of Digital Edge India, sees the current AI cycle as the beginning of a much larger infrastructure shift. “AI is just at the start point. What we are seeing today is just the tip of the iceberg — ChatGPT and a little bit of agentic AI. It will be embedded into many devices and applications in the future,” he said. His outlook extends toward sovereign AI, enterprise AI, agentic systems and adoption across sectors. He expects GPU infrastructure to expand alongside those applications, creating demand for accelerated-computing facilities. That shift makes future capacity planning increasingly dependent on whether facilities can support the higher-density power, cooling and structural requirements associated with AI workloads.

“AI capacity is likely to grow nearly 20 times from current levels. While overall data center capacity could expand five to seven times, AI infrastructure is expected to grow at a much faster rate, potentially by 20 to 25 times,” he added. The distinction matters because overall capacity may lack the power, cooling or structural flexibility needed for dense AI systems. Higher rack densities influence electrical architecture, thermal design, floor loading and equipment layout. “We need to ensure that it is modular and scalable. We should be able to handle the loads of the racks coming onto the floor, while ensuring that the floor capacity, cooling and power density can keep pace,” Prasad noted. Operators need to anticipate hardware generations that could demand different facility conditions. AI is turning future-proofing into a central engineering requirement.

Liquid Cooling Is Moving Into Core Design

Vimal Kaw, country managing director, India, NTT GDC, said customer requirements are becoming more differentiated across hyperscalers, emerging cloud providers and enterprises. Cooling represents one of the clearest physical consequences because newer AI systems generate heat at densities that challenge traditional air-cooled approaches. “Air cooling is not completely out at this point of time. We still have servers working on air, but the new generation will require us to change our data centers to accommodate liquid cooling,” Kaw noted. The message is not that air cooling disappears, but that facilities must support multiple thermal architectures. That requirement affects racks, piping, maintenance access and equipment deployment.

Kaw also expects electrical architecture to become a major redesign area as rack densities rise. “Whoever is building the next-generation data centres in this country is keeping AI workloads in mind from the design stage. At the same time, we are still accommodating conventional infrastructure,” he added. That combination creates a transitional model in which liquid-cooled AI systems may operate alongside conventional air-cooled servers. Power delivery must adapt as larger loads move from the grid through facility distribution systems and into individual racks. However, higher density also changes floor loading, equipment access, cable pathways and mechanical clearances. Next-generation facilities will need coordinated electrical, mechanical, structural and computing design.

AI Is Rewriting the Geography of Compute

Manoj Paul, managing director, India, Equinix, argued that AI will change where computing happens without making major cities irrelevant. “Cloud service providers may have their storage and compute outside the cities, but they have their interconnection nodes well within the city, where they are closer to the networks and the end users. The same thing is going to happen with AI.For compute, workloads could move outside cities, while inferencing stays closer to customers,” Paul said. His view separates intensive compute from low-latency locations. Large training environments may favor places with suitable power and expansion conditions, while inference can remain closer to customers and digital ecosystems. India could develop a distributed infrastructure pattern linking remote compute facilities with urban interconnection points.

For banks, exchanges and other digitally intensive businesses, proximity to cloud providers, telecom networks, partners and customers remains strategically relevant. Paul believes policymakers should account for that relationship when shaping infrastructure development. “Policymakers should not focus only on far-off locations. Policies must also encourage the development of data centres in cities and major towns,” he noted. Meanwhile, AI site strategy is becoming a network question rather than a simple land and power calculation. A single workload can span training facilities, inference nodes, cloud platforms, enterprise systems and edge locations. That makes urban connectivity an infrastructure asset that can complement large-scale compute deployed farther from population centers.

Modular Construction Could Accelerate AI Deployment

Jay Burse, chief vision officer at Prasa, said rapidly changing facility requirements are putting pressure on conventional construction methods. Data center projects now need to coordinate dense computing with power, cooling, structural loading and mechanical requirements from the beginning. “We really have to overhaul the whole traditional way of constructing. How much of the work that needs to be deployed on-site can be done in factories, where there is a more controlled environment and you can assure the deployment quality and the speed at which it is getting made?” Burse questioned. His argument favors moving complex work into controlled factories before components reach an operating site. That model could make deployment more repeatable while reducing specialized on-site construction.

The longer-term direction could produce more plug-and-play facilities, with prefabricated power, cooling and AI infrastructure assembled and connected on site. Burse cautioned that India still needs manufacturing capacity, supply chains and certification ecosystems capable of supporting that model at greater scale. “We are moving towards the direction,” he said, while emphasizing the industrial requirements behind modular deployment. Modularity could influence procurement, quality assurance, sequencing and deployment speed. It could help facilities respond to changing AI hardware without making every expansion bespoke. Ultimately, construction itself is becoming part of the data center technology strategy as operators seek greater speed and adaptability.

Distributed AI Expands the Infrastructure Stack

Sunil Yalagoor, technical sales leader-infrastructure at IBM India, expects enterprise AI to operate across several infrastructure environments rather than one dominant facility type. Enterprises may combine hyperscalers, colocation facilities, sovereign environments, on-premise systems and edge infrastructure according to workload requirements. “AI will be a distributed kind of architecture,” Yalagoor said. This model separates large-scale model training from locations where organizations apply AI to customer interactions, industrial processes and operational decisions. Large GPU facilities can handle intensive training, while inferencing can move closer to users, machines and relevant data. The resulting architecture depends on reliable movement among compute, networks and data.

Yalagoor used industrial inspection to show why edge infrastructure can matter alongside centralized AI compute. “For example, a factory may want to check whether a product is defective or not. There, AI at the edge is important because it needs to be responsive and much quicker. Hyperscalers will play a role in training the models, while colocation or sovereign cloud environments will be used for inferencing,” Yalagoor added. Data pipelines, security and connectivity therefore become core AI infrastructure functions. An enterprise can have powerful models yet still struggle if data cannot move reliably between environments. Paul also warned enterprises against repeating coordination problems associated with early cloud adoption. “Enterprises should know the end goal. When everything is accessible through AI, organizations should make sure on how they will achieve and orchestrate it. Connectivity to multiple clouds and nodes will all have to fall in place,” he added.

India’s Data Center Blueprint Is Becoming More Flexible

Paul’s warning expands the data center beyond server housing, because facilities can link cloud platforms, telecom networks, enterprises, edge devices and specialized compute. Network resilience, cross-connects, latency and changing traffic patterns can influence AI performance alongside processor capacity. Interoperability matters when enterprises use different providers across training, inference and data operations. A facility that cannot connect efficiently with surrounding digital ecosystems may offer substantial compute while delivering limited strategic value. As AI workloads distribute across locations, infrastructure decisions will increasingly need to account for how those locations operate together. That makes connectivity an architectural requirement rather than an afterthought.

Burse also challenged the industry’s traditional reliance on power consumption and facility scale as the main indicators of progress. “By 2030, I think the metric should be how well AI has penetrated and how well it has been adopted. Large-scale warehouses may be hundreds of kilometres away, but eventually, the adoption of AI is going to be with individuals. They do not care how many gigawatts are being consumed; what matters is whether it is bringing value to them and improving their lives,” Burse said. His argument tests infrastructure investment against practical value. India’s challenge is more complex than building additional server halls. The industry must support higher density, mixed cooling, adaptable power, modular construction and distributed workloads across increasingly connected operating environments. Taken together, the summit’s views suggest that adaptability will become as important as capacity in India’s next phase of data center development.