AI data-centre boom puts power grid under pressure

1. At a Glance

• AI data centres use much more electricity than conventional facilities. Their load can swing sharply with usage, especially when running Large Language Models (LLMs). This changes how power has to be moved through the grid [4].
• Grid modernisation costs are going up. Global demand for AI hardware is making the hardware for grid supervisory control systems (SCADA/EMS) more expensive [4].
• India's data-centre capacity grew about 4×, from about 375 MW (2020) to about 1,500 MW (2025). Data-centre electricity demand is projected at 13.56 GW by 2031–32 [1].
• UPSC link: this topic connects digital infrastructure, energy security, the grid and its regulation, how costs are passed on to consumers, and the carbon footprint of AI. It is a GS-III topic covering several themes.

2. Why in the News

• July 2026: The North Eastern Regional Power Committee (NERPC) held a special meeting with grid operators and Power Grid Corporation of India Ltd. (PGCIL). Maintenance contracts for the region's integrated grid were about to expire, so system upgrades had to be discussed [4].
• PGCIL told the meeting that the estimated cost of upgrading SCADA/EMS at the region's State Load Despatch Centres (SLDCs) had risen from ₹521 crore to about ₹688 crore. The stated reason was AI-driven demand pushing up hardware prices [4].
• 11 October 2026: The Hindu reported (Aroon Deep and Saptaparno Ghosh, Chennai edition, p. 17) that electricity officials across India are dealing with the ways AI infrastructure is changing transmission [4].

3. Background & Evolution

• Older cost pressures: grid upgrade costs were already rising because manufacturers are phasing out older components and specialists who can maintain legacy systems are getting scarce [4].
• Newer pressure: the fast growth of data-centre projects adds a further load on both generation and transmission infrastructure [4].
• Milestones:
• 2020: India's data-centre capacity was about 375 MW [1].
• 2023–24: the Economic Survey described India's grid as one of the largest unified electricity grids in the world [3].
• 2025: data-centre capacity reached about 1,500 MW [1].
• January 2026: the transmission network crossed 5 lakh circuit km (ckm), with 1,407 GVA of transformation capacity [2].

• July 2026: NERPC's special meeting on the SCADA/EMS upgrade, where costs had risen ₹521 crore → ₹688 crore [4].

• Related global work: the IEA "Energy and AI" report is hosted on NITI Aayog's ICEMF portal [5].

4. Core Static Facts

5. Multi-Dimensional Analysis

Economic

• AI demand for chips and servers raises prices for grid-control hardware. The NER SCADA/EMS estimate rose by about ₹167 crore [4].
• Costs pass through transmission charges, so AI growth can raise tariffs for all grid users, not only data centres [4].
• Data-centre power demand is projected at 13.56 GW by 2031–32, which needs early generation and transmission planning [1].

Scientific / Technological

• AI workloads, especially LLM training and inference, create sharp, fast load swings. This makes frequency and voltage control harder [4].
• Modern SCADA/EMS is needed for real-time balancing. Old legacy systems are hard to maintain [4].

Environmental

• A large new power load on a grid that still relies on fossil fuels can raise emissions unless renewable energy (RE) procurement and storage keep pace (analytical inference).
• Link: the IEA Energy and AI report frames AI as both a source of energy demand and a tool for optimising energy systems [5].

Administrative / Federal

• SLDCs are state-level. RPCs and PGCIL are regional and central. Upgrades need Centre–state coordination, as the NERPC forum showed [4].
• The North-East region has small state utilities, so a cost rise weighs on them more heavily [4] (inference on burden).

Geopolitical / Strategic

• Grid-control hardware competes with AI buyers worldwide for the same semiconductor and electronics supply chain. This raises supply-chain-security and cyber-security concerns for critical infrastructure (analytical).
• Data-localisation and digital-sovereignty goals encourage building data centres in India, which in turn needs reliable domestic power (analytical).

6. Recent Developments (last 12–18 months)

• 2025: India's data-centre capacity reached about 1,500 MW [1].
• January 2026: the national transmission network crossed 5 lakh ckm with 1,407 GVA [2].
• 2026: National Electricity Plan targets for 2032 were set out: 6.48 lakh ckm, 2,345 GVA, 168 GW inter-regional capacity [2].
• July 2026: NERPC's special meeting with PGCIL on the SCADA/EMS upgrade, with costs at ₹688 crore [4].
• October 2026: national reporting that AI is reshaping the way power is moved through the grid [4].

7. Prelims Hooks

• India's data-centre capacity grew from about 375 MW in 2020 to about 1,500 MW by 2025 [1].
• Data-centre electricity demand is projected at 13.56 GW by 2031–32 [1].
• India's national grid is the world's largest synchronous national grid [2].
• The transmission network crossed 5 lakh ckm by January 2026 [2].
• Transformation capacity is 1,407 GVA, with a target of 2,345 GVA by 2032 [2].
• The NEP target for transmission length is 6.48 lakh ckm by 2032 [2].
• Inter-regional transfer capacity is set to rise from 120 GW to 168 GW by 2032 [2].
• SCADA = Supervisory Control and Data Acquisition. EMS = Energy Management System [4].
• The NER SCADA/EMS upgrade cost rose from ₹521 crore to about ₹688 crore because of AI-driven hardware demand [4].
• NERPC convened the July 2026 meeting. PGCIL gave the cost estimates [4].
• Grid investments are recovered through transmission charges [4].
• SCADA/EMS upgrades were for State Load Despatch Centres (SLDCs), not RLDCs [4].

8. Is the Grid Really Under Pressure? The Case That It Isn't

• The strongest objection: data centres are still a small load
• India's data centres had about 1.57 GW of installed power capacity in August 2026 [6].
• Worldwide, data centres used about 415 TWh in 2024. That is only about 1.5% of all electricity used in the world [5].

• So a sceptic can fairly say that the AI load alone cannot yet strain a grid with 120 GW of inter-regional transfer capacity (the power that can be moved between regions) [2].

• What the objection gets right: the North-East cost rise is a price story, not a load story

• PIB says the SCADA/EMS cost went from ₹521 crore to ₹688 crore because hardware prices rose. It does not say the region's electricity demand went up [4].
• AI firms around the world are buying the same chips and servers. That pushes up prices for everyone, including grid operators [4].

• So the headline "grid under pressure" mixes two separate problems: costlier equipment today, and a bigger load in the future.

• What the objection misses: speed and growth

• The 13.56 GW projected for 2031–32 is about nine times the 2025 capacity [1].
• The government's own plan is for capacity to reach nearly 8 GW by 2030 [6].

• Building a transmission line takes years. Planners must act on the growth that is coming, not only on today's small share.

• Fair conclusion for the answer sheet: the load problem is a matter of planning for the future. The cost problem has already arrived. Do not mix up the two.

9. Why the North-East Pays for an AI Boom It Does Not Host

• The cost is shared by every consumer, not only by the cause
• Grid investments are recovered through transmission charges (the fee every grid user pays for moving power over the lines) [4].
• So the extra ₹167 crore for the North-East upgrade goes into the bills of that region's consumers [4].

• The price rise came from global AI hardware demand, not from data centres in the region [4]. The people who pay are not the people who caused the cost.

• Small utilities feel it more

• A 32% jump on one control-system contract is a large shock for small state utilities. A large state would absorb the same amount more easily [4].

• This is a cost-socialisation problem (a cost created by a few users is spread over everyone).

• Why this matters for Mains

• Sample question 3 in this note asks "who should bear the cost". The honest answer separates two things:
  • a global price shock (fair to share it, because no single user caused it), and
  • new connection load from big data centres (fairer to charge the user who adds it).

10. Data Centres Can Bend Their Load — If the Rules Ask Them To

• Today's planning treats AI data centres as rigid loads
• Grid planners usually treat large AI data centres as "always-on" loads that cannot change [10].

• In several countries, this has caused multi-year delays in connecting new data centres to the grid [10].

• Research shows the load can actually move

• Software can slow some AI jobs when the grid is stressed and still meet the service promises (SLAs) made to customers [10].

• In this way, a data centre can work like a flexible grid asset (a user that reduces demand when the grid needs relief) instead of a burden [10].

• Why this fits India's problem

• India's officials already warn that heatwaves, volatile demand and renewable variability are straining grids [8].

• A new load that can drop during an evening heatwave peak helps. A rigid load makes things worse.

• What should follow

• CEA and state regulators should make flexibility a condition of connection: big data centres that agree to cut load on a grid signal could be connected faster, following the model described in [10].
• SLDCs should get this ability in the SCADA/EMS upgrade itself, so that the new control rooms can see and signal large data-centre loads in real time [4].

11. Green Pledges Are Not the Same as Green Hours

• Companies promise renewable power
• Yotta aims to source over 80% of its power from renewables within three to five years [9].
• CtrlS has set a goal of 100% renewable power by 2030 [9].

• The government says the extra demand is to be met mainly through renewable energy [6].

• But a data centre runs all day and all night

• Solar power is produced only in daylight. A data centre needs power at 9 p.m. as much as at noon (analytical).
• A pledge counted over a whole year can still leave the night-time load running on coal-based power from the grid (analytical).

• Evening peaks are already the hardest hours for the grid during heatwaves [8].

• What should follow

• Regulators and buyers should ask for round-the-clock matching: renewable power plus storage that covers the hours the data centre actually runs, not only a total for the year (analytical).
• This links to the note's reading on Battery Energy Storage Systems and pumped hydro.

12. The New Control Rooms Are Also New Targets

• SCADA is the brain of the grid, so it is attractive to attackers

• SCADA/EMS lets an SLDC see and control the grid in real time [4]. Whoever controls it can switch power flows.

• The cyber shield covers the central layer clearly

• Grid-India has set up a Security Operation Centre (SOC) (a 24×7 cyber-watch room) [7].
• The government statement says this SOC watches the RLDCs and the NLDC [7].

• The upgrade in the news is at state-level SLDCs, which states run [4]. The statement does not say these are covered by the same SOC [7].

• The hardware comes from the same global market as AI servers

• Grid buyers now compete with AI buyers for the same chips and electronics [4]. The pressure to buy whatever is available, quickly and cheaply, can weaken checks on where the parts come from (analytical).

• The good news and its limit

• The government says no cyber incident so far has affected the operations of the Indian grid [7].

• But the North-East is about to replace its control systems [4]. A replacement is exactly when security must be built in, not added later.

• What should follow: the SLDC upgrade contracts should require the new systems to link to SOC-style monitoring, so that the state layer gets the same cyber-watch as the regional and national layers [7].

13. Anchors for Answers

• Data: India's data-centre installed power capacity was 1.57 GW in August 2026, with a plan for nearly 8 GW by 2030 [6]
• Data: Global data-centre electricity use was about 415 TWh in 2024, expected to exceed 945 TWh by 2030, rising from about 1.5% to about 3% of world demand [5]
• Data: The North-East SCADA/EMS upgrade cost rose from ₹521 crore to ₹688 crore (about 32%), recovered through transmission charges [4]
• Report/Committee: IEA, Energy and AI (hosted on NITI Aayog ICEMF) [5]; CEA National Electricity Plan (Transmission), with 2032 targets [2]
• Law/Case: Electricity Act, 2003: SLDC (s.31), RLDC (s.27), NLDC (s.26); electricity in the Concurrent List, Entry 38 (static)
• Comparison: Abroad, treating AI data centres as rigid "always-on" loads led to multi-year connection delays. Flexible-load software lets them cut demand during grid stress [10]
• Scheme: Grid-India's Security Operation Centre for cyber-monitoring of RLDCs and NLDC [7]

14. Mains Relevance

• GS-III: Infrastructure (Energy); Science & Technology (developments in IT/AI); Indian economy (investment, growth); Environment (energy transition).
• GS-II (secondary): Centre–state relations in a concurrent subject (electricity is in the Concurrent List, Entry 38, List III — static knowledge).
• Sample question stems: 1. The rise of AI data centres presents India with both an economic opportunity and a grid-management challenge. Examine. (15 marks) 2. Discuss how the volatile load profiles of AI workloads affect power-system stability. What institutional and technological reforms are needed in India's load-despatch architecture? (15 marks) 3. Who should bear the cost of grid upgrades driven by data-centre growth? Analyse in the context of transmission-tariff design and consumer equity. (10 marks)

15. Related Topics to Study Next

• Electricity Act, 2003 and the Electricity (Amendment) Bill: the legal basis of load-despatch centres and RPCs.
• One Nation One Grid / Green Energy Corridors: expanding transmission for renewables and new loads.
• National Electricity Plan (Transmission), CEA: the source of the 2032 transmission targets.
• IndiaAI Mission and the GPU compute programme: the policy driver of AI compute demand.
• Data localisation under the DPDP Act, 2023: drives domestic data-centre construction.
• Battery Energy Storage Systems and pumped hydro: smooth out volatile loads.
• India Semiconductor Mission: the hardware supply chain shared by AI and grid control systems.
• Cyber-security of critical infrastructure (CERT-In, NCIIPC): SCADA systems are high-value targets.

16. NCERT Concepts to Revise

• Data centre (Geography): The article's core subject: AI data centres draw heavy, fluctuating power and strain the grid.
• Energy security (Economy): Reliable, affordable power for data centres and the wider economy depends on a stable, well-upgraded grid.
• Infrastructure (Economy): Power transmission and grid control systems are core infrastructure that must expand to meet new AI demand.
• Energy transition (Economy): Adding big new loads to a fossil-heavy grid raises emissions unless renewables and storage keep pace.
• Centre-state relations (Polity): Grid upgrades involve state load centres, regional committees and the central PGCIL, needing Centre-state coordination.
• Concurrent List (Polity): Electricity is a Concurrent List subject, so both Parliament and states can legislate on it.

17. Common Errors / Trap Areas

• SLDC vs RLDC vs NLDC: the upgrade in the news was at State Load Despatch Centres [4]. Statutory sections (s.31 vs s.27 vs s.26) are often confused (static).
• NERPC vs NERLDC: the RPC is a coordination and planning forum. The RLDC is the operational despatch body. The meeting was convened by NERPC [4].
• PGCIL vs Grid-India: PGCIL is the transmission company that gave the cost estimate [4]. Grid-India (formerly POSOCO) runs the load-despatch centres (static).
• Units: data-centre capacity is in MW/GW, transformation capacity in GVA, and line length in ckm. Do not mix them [1][2].
• Year confusion: the 13.56 GW projection is for 2031–32. The NEP transmission targets are for 2032 [1][2].