Karnataka ties data centre incentives to power and water efficiency
Karnataka's new data centre policy targets 1 GW of capacity by 2031 while paying for efficiency against power and water intensity ratios, as AI racks draw up to ten times the power of conventional ones.
What happened
Karnataka's Cabinet approved a Sustainable Data Centre Policy 2026-2031 in September, targeting cumulative data centre capacity of 1 GW by 2031 and paying for the adoption of renewable energy and treated water. The State has 32 data centres, one in Mangaluru, one in Bidadi and the rest in and around Bengaluru. Official data puts the combined capacity of 18 of them at 292.27 MW, with estimates for all of them at around 400 MW. In March 2026 IT&BT Minister Priyank Kharge had called data centres a "necessary evil" while announcing plans for the policy.
Context
What changed is the power draw per rack. "A traditional, smaller-scale data centre, used for web and standard IT applications, typically requires about 5 to 10 kW per rack. For an AI hyperscale data centre, the numbers can be 30 to 100 kW per rack," said Rishika Ranga of CSTEP. More power means more heat, and cooling can consume large quantities of water where it is water-based. Water data is available for only 12 of the 32 centres with the Bengaluru water board; the rest potentially rely on private tankers. Those 12, mostly using air-cooled and dry-cooling designs, are estimated to consume up to 1,964 KLD, or 1.964 million litres per day, at peak capacity — equivalent to the daily demand of around 3,600 households of four people.
Why it matters
The policy regulates intensity rather than absolute use, which is the standard way to handle a sector you want to grow and discipline at the same time. Two ratios carry the incentives. Power Usage Effectiveness is total energy consumption divided by the energy the IT equipment uses, and Water Usage Effectiveness is water consumed per kilowatt-hour of IT energy; for both, lower is better. A PUE of 1.00 to 1.20 earns a one-time ₹20 crore, disbursed in the fourth year after the figure holds for three consecutive years, while a WUE of 0.7-1.3 L/kWh sustained for three years earns full reimbursement of water charges up to ₹30 lakh a year for two years. Renewable energy above 50% of consumption earns ₹0.50 per unit for five years, capped at ₹1.25 crore, and treated water procurement is reimbursed up to ₹8 lakh a year. Crucially the policy mandates plant-wise PUE and WUE reporting, without which none of it is enforceable. The limitation is built in: an intensity target permits total consumption to rise, and the State is targeting 1 GW against about 400 MW today.
Impact for India
- Data centre operators: A PUE of 1.00 to 1.20 earns a one-time ₹20 crore, and renewable use above 50% of consumption earns ₹0.50 per unit for five years, capped at ₹1.25 crore.
- Bengaluru's water supply: Only 12 of 32 centres report water use to the board, and those 12 alone draw up to 1.964 million litres per day at peak, about the demand of 3,600 households of four.
- Water-efficient designs: A WUE of 0.7-1.3 L/kWh held for three consecutive years earns full reimbursement of water charges up to ₹30 lakh a year, and treated water procurement up to ₹8 lakh a year.
- Power planning: The State will map water and energy availability across existing and proposed locations as a public guide for investors.
- The grid: With AI racks drawing 30 to 100 kW against 5 to 10 kW for conventional ones, the 1 GW target by 2031 is a materially different load from today's roughly 400 MW.
- Transparency: Plant-wise PUE and WUE reporting becomes mandatory, which is the precondition for any of the incentives being verifiable.
Key data
Concepts
Theory lens
Performance standards tied to incentive payments — instead of capping a facility's absolute resource use, the regulator defines an efficiency ratio and pays for performance against it, so scale is not penalised while inefficiency earns nothing. Karnataka applies this twice: a Power Usage Effectiveness of 1.00 to 1.20 earns a one-time ₹20 crore, and a Water Usage Effectiveness of 0.7-1.3 L/kWh sustained three years earns water-charge reimbursement. Mandatory plant-wise reporting is what makes either verifiable, which matters when only 12 centres currently report water use.
Global context
Data centres have been contested well beyond Karnataka, with locals protesting their environmental impact and governments rethinking policy as AI workloads raise power draw. The shift is physical rather than regulatory in origin: the specialised chips used to train and run large AI models consume far more electricity than conventional servers, taking a rack from about 5 to 10 kW to between 30 and 100 kW, with the additional heat then translating into cooling demand that can be met with water.
Stakeholders
Practice Question — from the story
With reference to Karnataka's Sustainable Data Centre Policy, consider the following statements:
- The policy targets a cumulative data centre capacity of 1 GW by 2031.
- Power Usage Effectiveness measures the water consumed per kilowatt-hour of energy used by IT equipment.
Which of the statements given above is/are correct?
Reveal answer & explanation
Answer: 1 only
- Statement 1: Correct. The policy targets cumulative capacity of 1 GW by 2031, against an estimated 400 MW in the State at present.
- Statement 2: Incorrect. That describes Water Usage Effectiveness. Power Usage Effectiveness is the ratio of a facility's total energy consumption to the energy consumed by its IT equipment.
Therefore, the correct answer is 1 only.
Concept check — test your understanding
Karnataka rewards data centres on Power Usage Effectiveness, a ratio of total energy use to the energy used by IT equipment, rather than capping total electricity consumption. What is the main advantage of regulating a ratio instead of an absolute level?
Reveal answer & explanation
Answer: B
- An intensity ratio separates efficiency from size. A hyperscale facility drawing 30 to 100 kW per rack can still score well if its overhead is low, while a small but wasteful one earns nothing, which is why the payment attaches to a PUE of 1.00 to 1.20 rather than to being small.
- Option A has the mechanics backwards. Computing the ratio requires metering both total consumption and IT-equipment consumption, so it needs more measurement, not less.
- Option C names the real weakness of intensity targets rather than an advantage. The State is targeting 1 GW by 2031 against about 400 MW now, so total use is expected to rise even as each facility gets more efficient.
- Option D describes firm-specific targets, whereas a single ratio threshold holds everyone to the same standard, which is what makes plants comparable.