Data Centre Evaluation Framework India NITI Aayog: AI Buyer Guide

What evidence supports a provider’s claims about power availability, cooling efficiency and future expansion capacity? A useful starting point is NITI Aayog’s Data Centre Evaluation Framework, released in October 2026. The framework outlines the documentation states should collect before approving a data centre. This guide interprets those requirements from a buyer’s perspective, helping organisations understand what evidence to request before selecting a provider.

Designed for CTOs, infrastructure heads and procurement teams evaluating data centre providers for GPU rack deployments, this guide translates the framework’s power, water and capacity requirements into a practical document checklist and explains what to look for in each record. It focuses on evaluating provider claims rather than facility design. For the technical requirements of an AI-ready facility, refer to our guide to what makes a data centre GPU-ready for AI workloads.

What NITI Aayog’s Data Centre Evaluation Framework Requires States to Verify

NITI Aayog’s Frontier Tech Hub released the Data Centre Evaluation Framework on 14 October 2026, developed with inputs from industry, the Ministry of Electronics and Information Technology (MeitY), the Ministry of Power and the Department for Promotion of Industry and Internal Trade (DPIIT). It is built on five pillars: Power and Energy; Water and Cooling; Siting and Environmental Sustainability; Economic and Strategic Value; and Policy, Governance and Implementation.

The framework is guidance for state governments, not a new regulation. Each pillar closes with recommended minimum approval documents, the core records a state should obtain before approval to confirm resource readiness, safeguards and measurable commitments. Those documents, along with the framework’s evaluation criteria, are what this guide borrows.

Two proposals point towards verification. The framework asks the Centre to set up a central audit mechanism, with MeitY as nodal ministry, to periodically audit or inspect data centres’ claims, and it asks the Bureau of Energy Efficiency (BEE), the Bureau of Indian Standards (BIS) and MeitY to standardise how PUE, WUE and CUE are measured, reported and verified. These are power, water and carbon usage effectiveness, ratios that compare a facility’s total electricity, water and emissions with what its IT equipment alone consumes.

MeitY told a parliamentary committee that it maintains no official data on data centre capacity. On the industry side, Uptime Institute’s 2025 survey found that collection and reporting of key sustainability metrics did not improve in 2025.

The framework also asks states to classify each proposal as conventional colocation, AI, high-performance computing (HPC) or GPU compute, or hybrid, and to assess power, cooling and infrastructure against that facility type. The PIB release describes this as distinguishing colocation from active compute services. A buyer can ask the same question of any hall on offer: was it planned and approved as GPU compute, or as conventional colocation?

Figure 1. Each common provider claim maps to something the framework asks states to check: a power claim to the Power Readiness and Ramp-up Plan, an efficiency claim to the design and operating PUE criteria, a water claim to the phase-wise water-supply plan and an expansion claim to year-wise energisation.

Which Power Documents Show a Provider Can Energise Your Racks?

The central record is a Power Readiness and Ramp-up Plan, validated jointly by the developer, the distribution company (DISCOM) and the state transmission utility (STU). The framework expects it to cover total campus demand, built capacity, expected utilisation, sanctioned MW, year-wise energisation, substation and transmission augmentation, lead times and the party responsible for each upgrade. Can the provider share its own version of this plan, or the utility sanction behind it?

The framework goes further than a headline MW figure. It asks states to record firm MW, the connection point and the sanction stage, and to align each capacity milestone with confirmed substation and transmission lead times. It calls for DISCOM and STU load-flow studies covering feeder limits and N-1 or N-2 contingency, that is, whether supply holds if one or two network elements are out of service, and for verification of independent upstream sources where a site claims dual feeds.

The framework adds three further points. For facilities above 20 MW, it asks for automated demand-response readiness and integration with State Load Despatch Centre (SLDC) telemetry and control. It recommends approving load enhancement only after the supporting assets are commissioned. And it counts renewable or storage-backed reductions in grid draw only where hourly or time-block matching, contracted quantum, commissioning and metering are confirmed.

How Should a Buyer Read PUE and WUE Claims?

The framework separates modelled design PUE from the operating PUE the developer commits to, and asks for the assumptions behind both for climate, rack density, cooling, uninterruptible power supply (UPS) and backup configuration. A buyer can ask a provider for the same three items. Is the quoted figure a design value or a measured operating value, over what period, and where do the meters sit?

The definitions come from international standards. PUE is defined in ISO/IEC 30134-2, whose second edition was published in January 2026 and which covers how to measure, calculate and report PUE. The standard also defines designed PUE, a projected value set by the data centre’s design targets. Water usage effectiveness is defined in ISO/IEC 30134-9:2022, which quantifies a data centre’s water consumption during its use phase.

The framework’s cross-pillar condition asks states to assess design-stage PUE, WUE and CUE against published BIS standards and to require post-commissioning measurement and reporting using the same boundaries, the pairing shown in Figure 2.

Figure 2. The framework pairs design-stage PUE with post-commissioning measurement and an operating PUE report on the same boundaries. A buyer can check whether a provider’s design and operating figures follow the same pairing.

For context on what operators report, Uptime Institute’s 2025 survey put the average PUE at 1.54, with facilities commissioned in the previous five years at 1.48 and those above 20 MW around 1.44. These are global averages from Uptime Institute’s 2025 survey; Uptime notes that improvement is constrained by legacy infrastructure and region-specific barriers to efficient cooling.

The framework applies the same approach to water. It asks for fit-for-use water specified separately for facility-level and rack-level cooling, distinguishing air cooling, water cooling, rear-door heat exchangers, direct liquid cooling and immersion. It asks for peak-summer demand and WUE modelled at local conditions, including approximately 40°C where applicable, and for a phase-wise water-supply plan covering peak demand, contracted supply, recycled sources, on-site storage and metering. Recycled or treated supply is deducted only when it is contracted and connected.

How cooling choice affects is covered in our post on Varuna and water-conscious AI infrastructure.

Testing Capacity and Expansion Claims Against Your Growth Plan

The PIB release puts India’s data centre capacity at about 375 MW in 2020 and approximately 1.5 GW as of October 2026, while MeitY, as MediaNama reports, has told a parliamentary committee that it maintains no official capacity data. The table below maps each common claim to the framework requirement and the evidence to ask for.

Sources: NITI Aayog Data Centre Evaluation Framework (14 Oct 2026); PIB release (15 Oct 2026).

The middle column shows what the framework asks a state to assess or obtain. Can the provider answer the right-hand column with hall-level, measured records, or only with design targets and campus totals?

How the checklist applies depends on the deployment:

• Inference pilot on a few racks. A buyer can ask whether the hall’s power is energised today and on what boundary the quoted PUE is measured.
• Training cluster growing in phases. A buyer can ask for year-wise energisation against the cluster roadmap, since the framework recommends load enhancement only after supporting assets are commissioned, and how demand response applies if the campus is above 20 MW.
• Regulated workload, for example banking, financial services and insurance (BFSI) under the Digital Personal Data Protection Act (DPDP Act). The framework’s Condition 14, as reported by MediaNama, asks states to map the facility and IT owner, workload controller, Data Fiduciary or Processor and compute controller. A buyer can ask the provider where it would sit in that map for the contract.

If the constraint you are hitting is rack power or cooling headroom rather than documentation, our post on signs your AI infrastructure has outgrown traditional colocation covers that decision.

Limits of this approach. The framework advises states; it does not bind providers, and its central audit mechanism is a proposal. It asks BEE, BIS and MeitY to standardise PUE, WUE and CUE boundaries. The Uptime averages above are global rather than Indian. Which of these records will be checked on a site visit, and which are written into the contract?

FAQs

Is India’s national data centre policy the same as the NITI Aayog framework?
No, they are separate. The government has said a dedicated data centre policy is coming to keep the sector from raising household power bills, without giving its provisions or a publication date. Will any of its provisions apply to a contract signed before it is published?

Which Indian states already have data centre policies?
Reports that states including Uttar Pradesh, Gujarat and Karnataka already run their own data centre policies. The NITI Aayog framework describes itself as guidance and does not prescribe a uniform approach.

Will WUE be reported by water source?
The framework lists source-wise WUE reporting among the tasks it proposes at the central level. It is a proposal rather than a current requirement. Does the provider already split WUE by water source, or report one site figure?

Choosing a Provider on Evidence, Not Headlines

The framework was written for states, but its records map onto provider claims: sanctioned MW and a dated energisation plan for power, measured values on stated boundaries for efficiency and water, and utility milestones for expansion.