In July 2026, the North Eastern Regional Power Committee (NERPC) convened a special meeting with grid operators and Power Grid Corporation of India to address an urgent issue. Maintenance contracts for the region's integrated electricity grid were nearing expiry, and the systems needed upgrading. The financial implications were significant.
The reason: AI-driven demand was making hardware used in supervisory systems more expensive. That is a jump of roughly ₹167 crore, or about 32%, driven primarily by the surge in data centre infrastructure.
This isn't just about replacing old servers. The Hindu's report notes that as manufacturers phase out older components and specialists capable of maintaining legacy systems become harder to find, replacement and maintenance costs can rise sharply. The rapid expansion of data-centre projects is adding a new layer of pressure on power generation and transmission infrastructure. PGCIL's filings don't explicitly break down the ₹688 crore figure, but the message from the NERPC meeting is clear: the hardware needed to run modern grid control systems is getting pricier, and AI is a key driver.
Data centres aren't like typical factories or offices. They consume substantially more power than conventional facilities, and their demand patterns can fluctuate sharply depending on usage, especially large language models (LLMs), server distribution, and the time zones of their customers.
A committee constituted by Grid-India was tasked with studying these demand patterns.
This creates a specific problem for generators. At a meeting of the Eastern Regional Power Committee's Technical Coordination Committee last December, an Odisha representative flagged the potential "ramping impact" on States when industries such as data centres undertake "bulk collective purchase" of electricity. Such fluctuations can strain slow-ramping generators, which cannot adjust output quickly enough to match sudden changes in demand. This mismatch exposes generators to penalties under the Deviation Settlement Mechanism (DSM) for deviations beyond their control. In simple terms, if a data centre suddenly draws a lot more power than scheduled, other plants on the grid have to scramble to fill the gap, and if they can't move fast enough, they get penalized.
The challenges don't stop at demand patterns. At meetings of the Northern Regional Power Committee's Operation Coordination Sub-Committee last year, officials repeatedly discussed monitoring data centres' compliance with grid-connection requirements before supplying them with electricity.
Here's what that means: if a grid disturbance occurs, data centres may automatically switch to their own backup power supplies. This would remove a large block of demand from the grid within moments. The grid would then have more generation than demand, which can disturb frequency and voltage stability. The 12 June incident in the Western Region, where a faulty bus tripped two interconnecting transformers and shed about 32 MW of data-centre load, is a real-world example of how vulnerable the system can be.
The physical infrastructure is also under strain. Alok Kumar, Director-General of the All-India Discoms Association (AIDA), told The Hindu that incremental demand from data centres ranging between one and two gigawatts per year will not make a huge difference nationally. However, this demand is concentrated in certain cities: Hyderabad, Mumbai, Chennai, Visakhapatnam, and the Delhi NCR region. State transmission utilities would need to expand capacity quickly in these locations, but "it could be a challenge in certain urban areas because of right of way (RoW)-related issues," Kumar said.
Maharashtra, home to a large share of India's data centres, illustrates the problem. Minutes of a Western Regional Power Committee meeting last year recorded discussions on the June 12 grid incident. At a January meeting, an official from Maharashtra State Electricity Distribution Company Limited warned that the proposed splitting ("line-in line-out") of the Talegaon-Kharghar transmission line could leave the already stressed corridor vulnerable to severe overloading if another line tripped. The corridor serves an area experiencing growing data-centre development around Mumbai. The official suggested upgrading the line to 2,100 MVA.
In the Southern region, the Southern Regional Power Committee has been preparing for a surge in demand from data-centre projects in Visakhapatnam. Officials discussed the prospect of gigawatt-scale demand emerging in the region over the coming months. At an April meeting, the SRPC's Protection Coordination Sub-Committee examined the implications of data centres for transmission infrastructure. An official noted that "inputs were sought from Grid-India to assess reserve requirements and system planning implications."
All this infrastructure costs money, and there's a live economic question: should data-centre operators or all consumers of a distribution company pay for the transmission they need? Kumar emphasized that data-centre operators must bear the costs of additional investments required to extend transmission infrastructure to their facilities. Such expenditure, he argued, should not be passed on to all consumers of the same distribution company. He pointed to Central Electricity Regulatory Commission (CERC) regulations that mandate this principle.
The regulatory framework supports this view.
Additionally, connectivity bank guarantees ensure applicants fund the cost of their own connection. If commissioning is delayed, the connectivity grantee pays Rs.3,000/MW/month from the connectivity start date, and this amount is reimbursed to other drawees, ensuring the delay doesn't fall on general consumers.
Regional power committees across the country are grappling with these issues. The NERPC meeting highlighted the cost pressure on SCADA/EMS upgrades. The Eastern committee flagged ramping impacts. The Northern committee is focused on compliance monitoring and load rejection risks. The Southern committee is planning for gigawatt-scale demand. The Western committee is dealing with corridor congestion and reliability.
Documents showed that Noida had 125 MW of operational data-centre capacity, with more than four times that capacity in the pipeline. This concentration explains why regional planning is becoming so critical. The grid can handle the load, but only if the right infrastructure is built in the right places, and if the costs are allocated fairly. As AI adoption accelerates, the tension between data centre growth and grid stability will only intensify. The question isn't whether India can power its AI ambitions—it's who will pay to keep the lights on when the machines wake up.
