AI Hyperscale Data Centre Power Outage Underwriting India 2026: BI Triggers, Heat Stress, and Reliance on Discom

Indian data centre capacity crossed 1,800 MW of IT load by early 2026, with the active pipeline expected to take cumulative IT load past 3,000 MW by 2028. The 2024 to 2026 growth has been driven by hyperscale builds: AWS in Mumbai and Hyderabad regions, Microsoft Azure in Pune and Hyderabad regions, Google Cloud in Mumbai and Delhi regions, plus the Indian colocation operators (CtrlS, Yotta, Sify, NxtraData, NTT Global Data Centers) building hyperscale-grade facilities to host both Indian and global cloud workloads. The shift from traditional enterprise hosting to AI training and inference workloads has changed the power, cooling, and risk profile materially.

AI workloads are power-intensive in ways that traditional enterprise computing was not. A single AI training rack with high-end GPUs (NVIDIA H100, GB200, AMD MI300 series) can draw 30 to 130 kW per rack compared to 5 to 15 kW per rack for traditional enterprise compute. The increased rack density produces higher heat output, requiring more aggressive cooling (liquid cooling becoming standard for AI workloads), more reliable power supply, and tighter operational margins. A power-quality incident that traditional enterprise hosting could absorb may produce equipment damage at AI hyperscale densities.

The loss patterns have started accumulating. Through 2024 to 2025 the Indian data centre market saw multiple notable power-outage incidents: a Maharashtra grid-side outage in 2024 that produced UPS-system stress at multiple Mumbai data centres with several hours of degraded operation, a Karnataka discom load-shedding event in 2025 that produced extended backup-generation operation at Bengaluru data centres with subsequent fuel-logistics challenges, a Pune data centre fire in 2024 originating in switchgear that produced 28 hours of partial-facility outage, and a Hyderabad data centre incident in 2025 where heat-driven derating during a summer peak reduced effective capacity for over a week. Each incident produced BI claims and triggered tenant SLA payments that flowed into contingent BI considerations.

The insurance market response has tightened underwriting on hyperscale and AI-focused data centres in 2026, with rate increases of 20 to 35 percent at the April 2026 treaty renewal on the most exposed installations, capacity tightening on AI-specialised facilities, and policy wording updates addressing power-quality, heat-derating, and tenant-SLA scenarios that earlier wordings did not address cleanly.

The Indian underwriting market for data centre placements in 2026 spans domestic insurers (Tata AIG, ICICI Lombard, HDFC ERGO, Bajaj Allianz, the four public-sector insurers) supported by GIC Re domestic reinsurance and offshore reinsurance via GIFT City IIO and direct Lloyd's India access. The largest hyperscale placements typically require layered structures across multiple primary insurers and reinsurance markets.

This guide covers the 2026 underwriting framework for AI hyperscale data centre power outage exposure, structured around the Indian operator set, the discom-side outage patterns by state, the backup generation and fuel logistics realities, the heat-driven derating exposure, the DPDP Act data-loss penalty interface, the contingent BI from tenant SLA failure, and the INR-denominated treaty capacity available.

Understanding the operator set is the foundation of the underwriting framework because risk profiles differ significantly across operator types.

Indian colocation operators with hyperscale-grade builds

CtrlS Datacenters: The Hyderabad-headquartered operator runs hyperscale-grade Tier-IV-equivalent facilities across Mumbai, Bengaluru, Chennai, Hyderabad, Noida, and Kolkata. The 2024 to 2026 build pipeline added capacity at multiple sites with single-facility IT load up to 80 to 120 MW at the largest builds. CtrlS hosts both Indian enterprise and global hyperscale workloads.

Yotta Infrastructure: The Hiranandani-group operator runs hyperscale facilities at Panvel near Mumbai and Greater Noida, with single-facility IT load up to 150 MW at the largest builds. The Yotta facilities have been positioned for both colocation and managed-services hyperscale workloads.

Sify Technologies: The Sify data centre business operates facilities across Mumbai, Chennai, Bengaluru, Hyderabad, Noida, and Kolkata, with capacity additions through 2024 to 2026. Sify has expanded its AI-workload-optimised capacity through dedicated builds.

NxtraData (Bharti Airtel): The Bharti Airtel data centre subsidiary runs facilities across Mumbai, Bengaluru, Chennai, Hyderabad, and Noida, with a 2024 to 2026 expansion programme focused on hyperscale and AI workloads.

NTT Global Data Centers: The NTT Communications-owned operator runs facilities across Mumbai, Bengaluru, Chennai, Delhi, and Noida.

Other operators: Additional operators with hyperscale-grade Indian presence include ST Telemedia Global Data Centres, Iron Mountain Data Centers, Equinix, Princeton Digital Group, and CapitaLand AscendasReit.

Cloud hyperscale operators with India regions

AWS: Amazon Web Services operates the Mumbai (ap-south-1) and Hyderabad (ap-south-2) regions, with significant additional capacity build through 2024 to 2026. The capacity is hosted across multiple Availability Zones with diverse infrastructure providers.

Microsoft Azure: Microsoft Azure operates the Central India (Pune) and South India (Chennai) regions, plus the West India (Mumbai) region launched through 2024 to 2025, with significant AI-workload-specific capacity build.

Google Cloud: Google Cloud operates the Mumbai (asia-south1) and Delhi (asia-south2) regions, with capacity expansion through 2024 to 2026.

Oracle Cloud Infrastructure: OCI operates the Mumbai and Hyderabad regions.

Insurance market implications

The colocation operator's insurance programme typically covers the facility infrastructure (building, power systems, cooling systems, network infrastructure) with tenant equipment frequently excluded or covered under separate tenant-side programmes. The hyperscale cloud operator's insurance programme typically covers both the facility and the tenant equipment (the cloud's own servers) as the cloud operator owns both.

The BI exposure differs accordingly. The colocation operator's BI exposure is principally the contracted colocation revenue plus tenant SLA payment obligations under colocation agreements. The hyperscale cloud operator's BI exposure includes the much larger revenue from the cloud services delivered through the facility, which can be orders of magnitude larger than the colocation revenue equivalent.

The 2026 underwriting practice differentiates between colocation BI and cloud-service BI, with the cloud-service BI typically not placed in the Indian insurance market but in the cloud operator's global insurance programme. The Indian market focuses on the facility insurance and the colocation-revenue BI element.

The Indian distribution-company (discom) supply reliability is a fundamental underwriting consideration for data centre placements. The reliability differs materially across states and within states, with significant implications for backup-generation reliance and BI exposure.

Maharashtra discom environment

Maharashtra hosts the largest concentration of Indian data centre capacity primarily in Mumbai (Powai, Andheri, Belapur) and increasingly at Panvel through the Yotta and other builds. The discoms involved include Adani Electricity (Mumbai island and suburbs), Tata Power (Mumbai island), Brihanmumbai Electric Supply and Transport (BEST) (central Mumbai), and Maharashtra State Electricity Distribution Company (MSEDCL, also called Mahavitaran) for areas outside Mumbai including Panvel.

The Mumbai discom environment has improved materially through 2020 to 2026 with reduced AT&C losses and improved supply continuity, but the system experiences periodic stress events. The 2020 Mumbai grid outage produced widespread disruption that informed subsequent data centre risk management. The 2024 to 2025 period saw discrete outage events including transmission-side disturbances and substation incidents, with most events resolved within hours but with the underlying exposure visible.

Karnataka discom environment

Karnataka hosts substantial data centre capacity primarily in Bengaluru. The discoms involved include Bangalore Electricity Supply Company (BESCOM) for Bengaluru urban areas and Bangalore Mangalore Electricity Supply Company (BMESCOM) and others for surrounding regions. The Karnataka discom environment has been more stressed than Maharashtra in 2024 to 2025 with load-shedding events during peak demand periods affecting parts of Bengaluru, summer-season grid stress producing voltage instability, and several incidents that required data centre operators to operate on backup generation for extended periods.

The Karnataka government and the discoms have invested through 2024 to 2025 in transmission and distribution infrastructure improvements, but underwriting continues to factor the Karnataka discom environment into BI assumptions.

Telangana and Andhra Pradesh

Telangana hosts substantial data centre capacity in Hyderabad served by Telangana State Southern Power Distribution Company (TSSPDCL). Andhra Pradesh hosts emerging data centre capacity served by Andhra Pradesh Southern Power Distribution Company (APSPDCL) and others. Both states have demonstrated improving reliability through 2024 to 2025 but with periodic stress events particularly during summer peak demand.

National Capital Region

The NCR data centre market is served by BSES (Tata Power for parts of Delhi), BSES Rajdhani and BSES Yamuna, plus Noida Power Company (in Noida sector areas) and Uttar Pradesh Power Corporation (in Greater Noida and Yamuna Expressway). The reliability has improved through 2020 to 2026 but with monsoon-season disruptions and summer peak stress producing periodic events.

Tamil Nadu and Chennai

Chennai data centres are served by Tamil Nadu Generation and Distribution Corporation (TANGEDCO). Tamil Nadu has demonstrated improving reliability with the state's renewable generation reducing fuel-cost stress, but with cyclone exposure and monsoon-season disruptions producing periodic events.

Underwriting implications

The discom environment affects underwriting in several ways. First, the assumed backup-generation runtime during a typical outage event scales with discom reliability, with the most reliable states requiring lower backup-generation assumption than the most stressed states. Second, the BI exposure during an extended outage is materially higher in states with stressed discom environments where extended backup-generation operation is more likely. Third, the fuel-logistics exposure (the risk that backup generation cannot be sustained because fuel cannot be replenished in time) is higher in states where outage duration risk is higher.

The 2026 underwriting practice incorporates state-specific discom reliability assumptions, with rate differentials reflecting the state-level exposure profile. The differential between the most reliable states (Gujarat, parts of Maharashtra including Mumbai) and the most stressed states can be 15 to 30 percent on the BI rate component.

Data centres maintain backup generation to cover grid outages, but the backup architecture has practical limits that produce material BI exposure beyond simple equipment failure scenarios.

Backup generation architecture

Indian hyperscale data centres typically operate N+1 or 2N backup generation architectures using diesel generators, with capacities scaling with the facility IT load. A 100 MW IT load facility typically maintains 130 to 200 MW of backup generation capacity to cover N+1 redundancy plus auxiliary loads (cooling, lighting, control systems).

The generators are typically sourced from major OEMs (Caterpillar, Cummins, MTU, Kohler, MWM) with on-site fuel storage typically sized for 24 to 72 hours of operation at full load. The exact sizing reflects the operator's assessment of outage duration risk, with most Indian hyperscale builds in 2024 to 2026 targeting 48 to 72 hours of on-site fuel storage.

UPS and battery support

Uninterruptible power supply (UPS) systems bridge the brief gap between grid outage and backup generator stabilisation, typically 30 seconds to several minutes. UPS systems use lead-acid batteries (older facilities), lithium-ion batteries (newer hyperscale builds), or flywheel systems (specialised applications).

The UPS systems are themselves subject to failure scenarios including battery degradation, control system faults, and capacity exhaustion in extended events. UPS-related incidents are a recurring source of data centre BI claim activity.

Fuel logistics during extended outages

The fuel logistics during extended grid outages is the practical exposure that operators sometimes underestimate. The on-site fuel storage provides 48 to 72 hours of operation; an outage exceeding the storage requires fuel replenishment from external suppliers. The fuel logistics chain includes:

1. Supplier arrangements: Data centres typically have contracts with multiple fuel suppliers (Indian Oil, Bharat Petroleum, Hindustan Petroleum, Reliance) for emergency fuel delivery.
2. Tanker capacity: Diesel tankers typically carry 10 to 28 kilolitres. A 100 MW backup generation load consuming approximately 25 to 35 kilolitres per hour requires tanker arrivals on a continuous schedule to maintain operations.
3. Road logistics: The tanker delivery depends on road access to the facility, which can be disrupted during the underlying event causing the grid outage (cyclone, flood, monsoon disruption).
4. Permit and regulatory: Some fuel transportation requires permits that may not be available on short notice during emergency events.
5. Concurrent demand: A regional grid outage produces concurrent fuel demand from many facilities, with fuel suppliers facing capacity constraints.

The Bengaluru 2025 incident

The Karnataka load-shedding events in 2025 produced an instructive case study. Several Bengaluru data centres operated on backup generation for extended periods (24 to 48 hours at the worst-affected facilities). Fuel logistics held up at most facilities but with significant operational stress, with operators dedicating senior staff to fuel coordination and with some near-miss situations where fuel deliveries arrived with limited margin. The incident drove subsequent improvements in fuel-logistics planning and supplier diversification across the affected operators.

Underwriting implications

The backup generation and fuel logistics exposure produces several underwriting considerations. First, the assumed sustained backup-generation runtime is a key input, with the realistic assumption typically capping at the on-site fuel storage capacity plus a conservative fuel-replenishment factor. Second, the fuel logistics arrangements are a survey-time consideration, with the underwriting reflecting the documented fuel supplier arrangements, the tested logistics protocols, and the operator's track record in fuel-logistics events.

The BI cover wording must respond to extended-outage scenarios including the fuel-logistics failure scenario. The 2026 market practice typically extends BI cover to backup-generation-related events including fuel exhaustion, generator failure during extended operation, and consequential damage from fuel-logistics challenges.

Liquid cooling and AI workloads

The shift to liquid cooling for AI workloads introduces additional considerations. Liquid cooling systems are more sensitive to power interruptions than air cooling, with potential coolant flow disruption producing equipment damage in seconds rather than minutes. The backup architecture for AI hyperscale facilities must address both compute power and cooling power continuity, with reduced tolerance for backup transition delays.

Indian data centre operations face a structural heat-stress challenge that affects effective capacity and the BI exposure during summer peak periods. The 2024 to 2026 underwriting position has incorporated heat-derating considerations more explicitly than earlier frameworks.

The heat-derating mechanism

Data centre IT equipment operates within specified temperature envelopes, with performance degrading and damage risk increasing as inlet temperatures rise. The cooling system removes heat from the IT load and rejects it to the outside environment. The cooling capacity depends on the temperature differential between the data centre cool side and the outside environment.

At extreme outside temperatures (typically above 40 degrees Celsius dry-bulb, with humidity considerations on top), the cooling system effectiveness degrades. Air-cooled chillers lose efficiency, evaporative cooling systems lose capacity in high-humidity conditions, and the data centre must derate IT load to maintain temperatures within specification.

Indian climate conditions produce summer peak temperatures exceeding 40 degrees Celsius across most data centre locations (Mumbai, Pune, Bengaluru, Hyderabad, NCR, Chennai). The 2024 summer in particular produced extreme heat events in NCR and Hyderabad with temperatures exceeding 45 degrees Celsius for extended periods.

The 2025 Hyderabad incident

A Hyderabad data centre incident in summer 2025 illustrated the heat-derating exposure. During a sustained period of extreme heat with temperatures above 44 degrees Celsius and high humidity, the facility's air-cooled chillers operated at reduced capacity, requiring the operator to derate IT load by approximately 20 percent for over a week to maintain temperature specification within tenant agreements. The derated capacity reduced contracted services delivery and triggered tenant SLA payments.

Underwriting implications

The heat-derating exposure produces several underwriting considerations.

1. Cooling architecture review: The cooling system design, including chiller capacity, evaporative cooling provisions, and redundancy, is a key underwriting consideration. Facilities with marginal cooling capacity for Indian summer conditions are higher risk.
2. Liquid cooling for AI workloads: AI workload facilities increasingly use liquid cooling for the high-density racks. Liquid cooling provides better thermal management at high rack densities and is more tolerant of high outside ambient conditions.
3. BI cover for derated operation: The standard property-and-BI cover wording may not respond cleanly to derated operation scenarios where the facility is functional but operating at reduced capacity. The 2026 market practice has tightened the wording to explicitly address derated operation BI.
4. Tenant SLA exposure: Derated operation can trigger tenant SLA payments where the colocation agreement specifies capacity commitments. This is a contingent BI consideration that requires explicit treatment.

Forward outlook

Climate change projections suggest that Indian summer extremes will continue to intensify through the 2020s, with implications for data centre cooling architecture design and ongoing BI exposure. Operators planning new builds in 2026 are increasingly designing for higher peak ambient temperatures than earlier facilities, with corresponding capital cost implications.

The insurance market response is starting to differentiate facilities by cooling architecture, with newer facilities designed for AI workloads and extreme ambient conditions securing better rate terms than legacy facilities with cooling designed for earlier climate norms.

Beyond the operational property and BI exposure, AI hyperscale data centres carry significant regulatory and contractual exposure that intersects with the insurance programme.

Digital Personal Data Protection Act 2023 (DPDP Act)

The DPDP Act 2023 establishes data fiduciary accountability for personal data processing in India. Data centre operators may carry data-fiduciary or data-processor obligations depending on their role: a colocation operator hosting tenant equipment is typically a data-processor for the tenant's personal data; a managed-services operator providing data processing services is often a data-fiduciary.

The DPDP Act establishes administrative penalties for non-compliance including:

1. INR 250 crore for failure of data fiduciary to take reasonable security safeguards (Section 33).
2. INR 200 crore for failure to notify the Data Protection Board of a personal data breach.
3. INR 50 to 150 crore for various other compliance failures.

The penalties can be triggered by data-loss scenarios arising from data centre incidents including power-outage-induced data corruption, cooling-failure-induced equipment damage with data loss, and security incidents at the facility.

Insurance interface with DPDP Act penalties

The insurance treatment of DPDP Act penalties varies by product. Property insurance typically does not cover regulatory penalties. Cyber insurance may cover regulatory penalties subject to sub-limits and specific conditions including the absence of intentional wrongdoing. Professional indemnity or directors' and officers' liability may respond to specific scenarios involving claims against named persons or the operating entity.

The 2026 market practice for hyperscale data centre placements typically addresses DPDP Act exposure through the cyber insurance element of the operator's broader programme rather than the property and BI cover. Operators should ensure that the cyber programme scope includes data-loss scenarios arising from physical-event causation, not only cyber-attack causation.

Tenant SLA failure and contingent BI

Colocation operators contract with tenants under master service agreements that specify service level commitments including uptime guarantees, capacity commitments, and performance commitments. Failure to meet the SLAs triggers service credits or service charges that flow into the operator's financial position.

The SLA structures typically include:

1. Uptime SLAs: typically 99.9 percent to 99.999 percent uptime (the latter being the Tier IV-equivalent reference) with service credits for downtime.
2. Capacity SLAs: commitments on power and cooling capacity availability, with credits for capacity shortfall.
3. Performance SLAs: commitments on network connectivity, latency, and other technical parameters.
4. Recovery SLAs: commitments on time to restore service after an incident.

The SLA-triggered payments can be material in absolute terms during extended events. A multi-day outage at a major hyperscale facility can produce SLA payments exceeding INR 100 crore depending on the tenant mix and SLA terms.

Insurance interface with SLA payments

The insurance interface with SLA payments is contested. Standard BI cover responds to the operator's revenue loss, which is conceptually distinct from the SLA payment exposure (which is the operator's contractual liability to tenants). The 2026 market practice typically treats SLA payments as a contingent BI exposure with explicit cover wording, with sub-limits scaled to realistic worst-case SLA payment exposure.

The wording for SLA payment cover requires careful negotiation. The standard issues include: whether the cover responds only when there is underlying property damage or also for non-damage outage events, how SLA payments are quantified for cover purposes (gross or net of any operator defences), the sub-limit calibration, and the coordination with the operator's tenant insurance requirements.

Tenant insurance requirements

Colocation master service agreements typically require tenants to maintain their own insurance covering tenant equipment, tenant-side BI, and tenant-side cyber exposure. The tenant insurance requirements interact with the operator's insurance programme through subrogation considerations and through claim-time coordination.

The 2026 best practice for hyperscale operators is to maintain explicit alignment between the operator insurance programme and the tenant insurance requirements specified in the colocation MSA, with periodic review of the alignment as tenant mix evolves.

The placement of hyperscale data centre cover at the scale required by the Indian pipeline depends on the available INR-denominated treaty capacity supported by offshore reinsurance.

Indian primary insurer capacity

The major Indian non-life insurers writing hyperscale data centre cover in 2026 include Tata AIG, ICICI Lombard, HDFC ERGO, Bajaj Allianz, the four public-sector insurers, and selective participation from Kotak Mahindra General, SBI General, and IFFCO Tokio. The individual primary capacity for a single hyperscale facility typically runs INR 300 to 800 crore depending on the operator's loss history, the facility characteristics, and the placement structure.

For major hyperscale facilities (typically 80 MW IT load or larger), the placement is layered across multiple primary insurers in co-insurance with the upper layers supported by offshore reinsurance.

GIC Re domestic reinsurance support

GIC Re provides treaty and facultative reinsurance support to Indian primary insurers on data centre placements. The treaty support has expanded through 2024 to 2026 as the data centre pipeline has activated, but GIC Re's retention on individual large placements remains constrained.

GIFT City IIO offshore reinsurance

The primary capacity source for the upper layers on major hyperscale placements is offshore reinsurance via GIFT City International Insurance Offices. The major participants include Munich Re, Swiss Re, Hannover Re, SCOR, and various Lloyd's syndicates with data centre underwriting expertise.

The offshore reinsurance pricing for Indian data centre cover has tightened through 2024 to 2026, with rate increases of 20 to 35 percent at the April 2026 treaty renewal on the most exposed installations. The pricing reflects both global data centre underwriting positions and the specific Indian considerations including discom reliability variability, heat-derating exposure, and the AI-workload-specific risk profile.

Direct Lloyd's India access

The Lloyd's India direct entry through 2024 to 2026 has opened additional capacity for large Indian commercial placements including hyperscale data centres. The direct Lloyd's access provides an alternative path to international capacity that complements the IIO route.

Pricing range in 2026

The pricing range for hyperscale data centre property and BI cover in 2026 spans:

1. Hyperscale property cover: 0.30 to 0.65 percent rate on line for the property element, with the lower end for top-tier operators with strong loss history and the upper end for newer facilities or operators with elevated exposure.
2. Hyperscale BI cover: 0.45 to 0.90 percent rate on line for the BI element, depending on the indemnity period (typically 18 to 30 months) and the tenant mix.
3. AI-workload-specific facilities: typically command a 15 to 30 percent premium loading on base rates reflecting the higher rack density and cooling sensitivity.
4. Contingent BI for tenant SLA: 0.80 to 2.20 percent rate on line on the contingent sum insured, reflecting the limited capacity and concentrated exposure.

Placement timeline

Major hyperscale data centre placements in 2026 typically require 6 to 12 months of preparation including engineering surveys, market engagement, treaty support negotiations, and wording finalisation. The placement timeline has lengthened from the historical 3 to 6 month norm reflecting the additional underwriting rigour and the multi-market structures required.

What operators should do

The operator posture for clean hyperscale data centre placement in 2026 includes: early broker engagement (6 to 12 months pre-renewal for major facilities), detailed submission with infrastructure documentation, demonstrated operational discipline with documented evidence on backup-generation maintenance, fuel-logistics arrangements, UPS battery testing, cooling system performance, incident response protocols, and tenant SLA structures. Operators that present this evidence cleanly typically secure both better terms and improved capacity access compared to peers presenting thinner submissions.

The forward view

The Indian hyperscale data centre pipeline through 2026 to 2030 will continue to require expanding insurance capacity. The market response to the pipeline activation has been broadly constructive, with insurers and reinsurers building Indian data centre expertise and with capacity availability sufficient for well-prepared placements. The capacity is not, however, infinite. Operators planning multi-facility builds should treat insurance capacity as a material project consideration alongside power supply, cooling architecture, and tenant demand.

The AI hyperscale segment specifically is in early stages of insurance market maturation. The first major AI hyperscale claim in India, when it occurs, will likely reshape both the pricing and the capacity availability. The operators that have built relationships with insurance markets through 2024 to 2026 are positioned to manage the transition; the operators treating insurance as a marginal placement will face capacity and pricing challenges when the market reprices.