Data-centre load strains the grid and raises the question of who pays
AI-driven hardware demand pushed the estimated cost of upgrading one region's grid control systems from ₹521 crore to about ₹688 crore, and regional power committees have been flagging data-centre load for over a year. The live economic question is whether operators or all consumers of a distribution company pay for the transmission they need.
What happened
In July the North Eastern Regional Power Committee convened grid operators and Power Grid Corporation of India to discuss upgrading the region's integrated grid systems as maintenance contracts neared expiry. Power Grid told the meeting that the estimated cost of upgrading the Supervisory Control and Data Acquisition and Energy Management System at the region's State Load Despatch Centres had risen from ₹521 crore to about ₹688 crore, because AI-driven demand was making the hardware used in supervisory systems more expensive. Similar concerns have surfaced across regional power committees over the past year.
Context
Data centres consume substantially more power than conventional facilities and their demand can fluctuate sharply with usage, server distribution and the time zones of their customers. The Central Electricity Authority's National Power Committee noted in February that "data centres typically exhibit high constant loads, with possible ramping characteristics depending on global time zones and server distribution", and a committee constituted by Grid-India was tasked with studying their demand patterns. An Odisha representative had earlier flagged the "ramping impact" on States when data centres undertake bulk collective purchase of electricity.
Why it matters
Three distinct problems are being conflated in public discussion and the committee records separate them. The first is aggregate demand, and on that the industry view is relaxed: Alok Kumar, Director-General of the All-India Discoms Association, said incremental demand of one to two gigawatts a year "will not make a huge difference". The second is volatility rather than volume. Sharp swings strain slow-ramping generators that cannot adjust output fast enough, exposing them to penalties under the Deviation Settlement Mechanism for deviations outside their control, and the Northern Regional Load Despatch Centre warned in February that data centres could switch to backup power rapidly, creating a "credible risk of abrupt load rejection". The third is local transmission, which is where the constraint actually binds — Kumar's point was that State transmission utilities would have to upgrade quickly where demand concentrates, and that right-of-way issues make this hard in urban areas. The cost-allocation question follows directly from the third. Kumar argued that operators must bear the cost of extending transmission to their own facilities rather than having it spread across every consumer of the same distribution company, pointing to Central Electricity Regulatory Commission regulations that mandate the principle.
Impact for India
- Grid control systems: the upgrade estimate for one region's load despatch centres rose from ₹521 crore to about ₹688 crore on AI-driven hardware costs alone.
- Geographic concentration: demand is concentrated in Hyderabad, Mumbai, Chennai, Visakhapatnam and the Delhi NCR region, so the strain is local rather than national. Noida had 125 MW of operational data-centre capacity with more than four times that in the pipeline.
- Generators: sharp load swings can expose slow-ramping plants to Deviation Settlement Mechanism penalties for deviations they cannot control.
- Tariff incidence: transmission investment is recovered through transmission charges, but the cost of extending infrastructure to a specific facility should fall on its operator under CERC regulations rather than on all consumers of that distribution company.
- Reliability: a grid incident on June 12 in the western region saw a faulty bus trip two interconnecting transformers and shed about 32 MW of data-centre load, and an official warned that splitting the Talegaon-Kharghar line could leave an already stressed corridor vulnerable, suggesting an upgrade to 2,100 MVA.
Key data
Concepts
Theory lens
Marginal cost pricing of a network service — efficiency requires that a user facing a cost it alone causes should pay that cost, otherwise the price signal is wrong for both parties and other users subsidise the decision. The committee records turn an engineering issue into exactly this question: aggregate demand is affordable, but a single large load that forces a specific transmission upgrade imposes a cost traceable to it, which is why the CERC principle on operator-borne extension costs is the economically live part.
Stakeholders
Practice Question — from the story
With reference to data-centre electricity demand and India's grid, consider the following statements:
- The discoms association considers incremental data-centre demand unmanageable in aggregate, at one to two gigawatts a year.
- The main concern raised in regional power committee records is the total volume of electricity data centres consume, rather than how sharply their load varies.
Which of the statements given above is/are correct?
Reveal answer & explanation
Answer: Neither 1 nor 2
- Statement 1: Incorrect because the association's Director-General said demand of one to two gigawatts a year "will not make a huge difference" — the aggregate view is that it is manageable.
- Statement 2: Incorrect because the records separate volume from volatility and locate the problem in the latter and in local transmission: sharp swings strain slow-ramping generators and expose them to Deviation Settlement Mechanism penalties, while right-of-way issues slow urban upgrades.
Therefore, the correct answer is Neither 1 nor 2.
Concept check — test your understanding
A single large consumer requires a dedicated transmission upgrade. On standard pricing principles, who should bear that cost?
Reveal answer & explanation
Answer: D
- Efficiency requires the party causing a cost to face it, otherwise its decision is made on the wrong price and other users subsidise it. This is the CERC principle cited: operators must bear the cost of extending transmission to their own facilities.
- Option A is how shared network costs are recovered through transmission charges generally, which is exactly why the distinction for facility-specific assets matters.
- Option B is tempting because grid stability genuinely is a system-wide concern, but that does not make a single site's connection a system cost.
- Option C has some intuitive appeal through the revenue link, yet generators do not control where a large load chooses to locate.