India’s artificial intelligence ambitions are driving a new wave of data centre investment. But as AI workloads become more power-intensive, the next constraint on expansion may not be land, capital or computing capacity. It could be the availability of reliable electricity.
For years, data centre location decisions have centred on connectivity, land availability, latency and proximity to customers. AI is changing that equation by making power availability a strategic consideration in infrastructure planning. The question for operators is no longer simply how quickly a facility can be built, but how quickly it can secure dependable electricity at the scale required.
“‘Time-to-power’ is becoming as important as time-to-market,” says Bragadesh Damodaran, Vice President – Global Industries Platform Leader, Capgemini.
The Capgemini Research Institute’s AI meets the grid report highlights the scale of the challenge. AI workloads can consume five to ten times more power than traditional computing, while 83% of data centre executives expect AI-driven facilities to significantly increase regional electricity demand over the next three to five years.
For India, this could have implications for where the next generation of AI infrastructure is built. Locations with reliable grid access, spare capacity, resilient electricity supply and faster interconnection timelines may become more attractive than those selected primarily for digital connectivity.
“In many cases, access to dependable electricity may become the primary determinant of where the next generation of AI infrastructure is built,” Damodaran says.
When energy becomes a constraint on AI growth
The pressure is not simply about rising electricity consumption. It is about whether power infrastructure can expand at the pace required by AI.
According to the research, 77% of electricity executives believe data centre energy demand will grow faster than supply, while 68% anticipate shortages.
Damodaran argues that the challenge is less about a shortage of investment appetite or demand for AI and more about the readiness of the energy ecosystem.
India has been expanding its renewable energy capacity and modernising its grids. However, AI-scale infrastructure requires reliable, continuous power rather than installed generation capacity alone.
The distinction matters because the availability of electricity depends on more than generation. Grid capacity, transmission infrastructure, resilience and the time required to connect new facilities all influence whether a proposed data centre can operate at its intended scale.
“The winners in the AI economy will be regions that can align power planning, grid modernisation, transmission expansion, and data centre growth,” Damodaran points out.
Energy readiness could therefore become a stronger competitive differentiator than computing availability itself. For infrastructure developers and policymakers, this means treating electricity planning as an integral part of digital infrastructure strategy rather than a parallel exercise.
The problem of ‘phantom demand’
Planning for this expansion is complicated by uncertainty over how much proposed demand will actually materialise.
The Capgemini research finds that 67% of electricity organisations believe some data centre load requests are speculative, with 19% estimating that some requested capacity may never be realised.
This creates a difficult planning problem. Data centre developers may reserve electricity capacity years before a facility becomes operational and, in some cases, submit requests to multiple utilities. If those requests are treated as firm demand, utilities risk overestimating requirements, directing investment towards the wrong locations and building infrastructure that remains underutilised.
At the same time, underestimating genuine demand could delay projects and constrain AI infrastructure growth.
“Phantom demand is one of the most underappreciated challenges facing utilities globally,” Damodaran says.
His proposed response is a shift from reactive planning towards demand intelligence. Utilities and regulators could use milestone-based capacity allocation, readiness scoring and stronger disclosure requirements to distinguish committed projects from speculative requests. AI-based forecasting could further improve visibility into the likelihood and timing of future demand.
The objective is not to slow infrastructure investment but to ensure scarce grid capacity is allocated to projects most likely to proceed.
AI could also help utilities manage the grid
AI is contributing to rising electricity demand, but it could also help utilities manage increasingly complex power systems.
The research indicates that more than 60% of utilities expect AI to deliver meaningful improvements in reliability, productivity, outage prevention and operational efficiency.
The most immediate opportunities include load forecasting, predictive maintenance, fault detection, renewable energy integration and real-time grid optimisation. AI can help utilities anticipate demand surges, identify equipment failures before they occur and optimise power flows as electricity demand becomes more variable.
However, adoption remains uneven. According to the research, only around 45% of utilities currently use AI for grid optimisation at scale.
For Damodaran, this highlights the difference between the potential of AI and its operational maturity.
“AI will not replace grid infrastructure investment, but it can significantly improve how existing assets are utilised and managed,” he says.
Realising those benefits will require strong data foundations, governance, cybersecurity and integration with operational processes. AI can improve how the grid is managed, but it cannot substitute for the physical infrastructure needed to generate, transmit and distribute electricity.
Data centres become active energy participants
A further shift is emerging in the relationship between data centres and the electricity system.
Traditionally, data centres have operated primarily as consumers, drawing power from the grid to maintain continuous digital services. As constraints intensify, operators are increasingly exploring on-site generation, battery storage, microgrids, demand response and flexible workload management.
The Capgemini research finds that 29% of data centres already deploy some form of on-site or behind-the-meter power generation, while another 39% plan to add it within the next one to two years.
“We are already seeing the beginning of that transformation,” Damodaran says. “In the AI era, [data centres] are increasingly becoming active energy participants.”
For India, this could create a more collaborative energy ecosystem in which data centre operators work with utilities and regulators to improve reliability, manage local constraints and explore opportunities for storage and grid flexibility.
It could also influence future data centres and energy policy. Alongside location, connectivity and sustainability, operators may need to plan for how their facilities generate, store and manage electricity — and how their energy strategies interact with the wider grid.
Damodaran sees the future data centre functioning as both a digital asset and an energy asset.
Power planning becomes AI infrastructure planning
The implications extend beyond data centre operators. Utilities must improve demand visibility, regulators need mechanisms to distinguish genuine projects from speculative capacity requests, and infrastructure planners must coordinate electricity generation, transmission and digital expansion more closely.
For India, the challenge is to ensure that growing AI investment is matched by energy infrastructure capable of supporting it reliably and sustainably.
AI may be accelerating demand for computing, but the physical conditions required to run that computing are becoming equally strategic. Locations that can provide dependable power, timely grid connections and resilient energy systems are likely to gain an advantage as data centre capacity expands.
The next phase of India’s AI infrastructure race may therefore be decided not only by who can build the largest facility or deploy the most advanced computing hardware but also by who can secure and manage the power needed to keep it running.
As Damodaran’s assessment suggests, the future of AI infrastructure will depend as much on the grid as on the data centre itself.