Sept 21 AI Infra Shift: MeitY Wants Microsoft Compute at Power Plants

For AI practitioners, the biggest constraint on model training capacity may not be GPUs but power delivery. MeitY's suggestion to co-locate data centres with generation hubs is an early indicator that India plans to solve AI compute's energy problem at the grid architecture level. This could influence where future AI training clusters are built across the country.

On September 21, 2026, at the Microsoft AI Infrastructure Summit India in New Delhi, S Krishnan, Secretary of India's Ministry of Electronics and Information Technology (MeitY), urged Microsoft to rethink where its future Indian data centre capacity is sited. Rather than concentrating solely around cable landing stations, Krishnan argued, the company should increasingly place capacity closer to power generation hubs. His core point was economic and systemic: moving large amounts of electricity over long distances—'wheeling' it across the grid—entails avoidable costs. Co-locating demand with generation, he said, would be 'much more economical' for India, for the ecosystem, and for Microsoft itself.

“

MeitY's suggestion to co-locate data centres with generation hubs is an early indicator that India plans to solve AI compute's energy problem at the grid architecture level.

This intervention matters because it signals the next phase of India's data centre buildout. The current model has favoured coastal metros such as Mumbai and Chennai, where submarine cable landing stations provide international connectivity. But India's power geography is different: much of its scaling generation fleet, especially renewable capacity, is located in interior states and solar/wind-rich zones. Krishnan's suggestion is an attempt to align digital infrastructure with the energy transition rather than fight the grid's geography.

Krishnan also framed the policy environment as unusually favourable. He cited taxation policies, ease of doing business measures, and changes under the new telecom law. Building a data centre and connecting it to the grid, he said, is now easier in India than in many countries, including in Europe. Electricity permissions and access to other resources had also become easier. This is a direct pitch to Microsoft and other hyperscalers: India wants capital-intensive digital infrastructure and is prepared to reduce regulatory friction.

For Microsoft, the request lands during a significant expansion. Krishnan said the company's buildout in India is running ahead of schedule. He also stressed that the requirement for such infrastructure is not only an Indian need but a global one. The subtext is that India's buildout can be a template for how hyperscale AI infrastructure integrates with national grids.

The implications for data centre economics are substantial. Power can account for half or more of operating costs, and transmission and distribution charges add a layer that co-location could reduce. If hyperscalers site large training clusters near renewable parks, they may capture lower effective energy costs and support grid investment. But the trade-offs are real: cable landing stations exist for a reason—international fibre connectivity, low-latency access to global networks, and mature markets for land, cooling, and talent. Moving capacity inland requires robust fibre backhaul, water access for cooling, and grid interconnection that can deliver reliable power, not just abundant generation.

Krishnan's mention of efficient use of power and water indicates the government is not only focused on location but on resource intensity. That is notable because data centre water consumption has become a sensitive issue in water-stressed regions. A policy push that encourages co-location must also address cooling water availability and environmental clearances at generation hubs.