The world's AI buildout
needs firm 24/7 power
that grids cannot
deliver today.
AI data centers are being built at a pace grids were never designed for. Operators currently rely on hybrid stacks of solar, wind, and short-to-medium duration storage. However, between weather intermittency and the duration limits of today’s local technology, a critical gap remains. The grid was not designed for this density, and 100% renewable mandates from global hyperscalers stay physically unmet.
Kyros Infra closes this gap. We connect US-based long-duration resiliency technology with the Indian infrastructure leaders ready to secure true 24/7 firm power.
70%
Max renewable supply achievable with solar and short-to-medium duration storage
8-12 hours
What current storage bridges, when AI workloads need 24/7 resilient firm power
5GW+
India's projected data center capacity by 2030, from 1.6GW today
100%
Renewable mandate from global hyperscalers, not yet deliverable in India's current infrastructure
Kyros Infra facilitates market entry for the next generation of energy storage. We are not a utility, developer, or fund. We are a trusted operating bridge between US LDES companies ready to scale and Indian infrastructure operators with the land, mandates, and capital to deploy.
The Reality
The power exists.
Firm, always-on,
100%
renewable
power
does not.
India’s data center operators are fully aware, and they are scaling a comprehensive hybrid energy stack — grid power, captive solar, wind PPAs, rooftop solar, and short-duration BESS. Leading firms have already reached nearly 50 percent renewable penetration, while others are aggressively pushing for a 70 percent target. This momentum signifies deep industry progress that is being validated by the market.
But every serious operator hits the same ceiling. Solar peaks at noon and drops to zero by 9pm. BESS bridges two to four hours. The grid was engineered for industrial loads, not for 30-100kW-per-rack AI density running continuously. Operators can reach 70-80% clean energy. The final 20-30% – overnight hours, monsoon weeks, demand spikes – breaks uptime SLAs and impacts hyperscaler contracts.
Long-duration energy storage – thermal batteries, iron-flow systems, pumped-heat technology – closes this gap. It is commercially proven in the US. It is not yet deployed in India at scale. That gap is the commercial opportunity Kyros Infra facilitates.
20%
The firm power gap that solar + BESS cannot close - the hours every day when Indian AI data centers are exposed to grid instability or unclean power
8-24hr
Discharge duration LDES technologies deliver - vs. the 2-4hr ceiling of every lithium-ion BESS currently deployed in India
₹1.4T
Estimated firm power infrastructure gap across India's data center hubs - the addressable LDES deployment market by 2030
Three Gaps
The operators have power. They do not have firm power. And they cannot build the bridge alone.
Kyros Infra operates to close these gaps.
Gap 01 / The Density-Power Mismatch
The grid was not built for what AI needs
India’s legacy operators run at 5–8 kW per rack; today’s AI workloads demand 40–120 kW, with a roadmap toward 250 kW. Renewables disappear by evening or during weather drops. While current storage can provide 8-12 hours, they cannot bridge a multi-day outage. For a facility that must run at 100% carbon-free full load, a multi-day drought is not a buffer, but It is a broken SLA and a stranded hyperscaler client.
Gap 02 / The Technology Access Gap
The solution exists — but not in India yet
Long-duration storage is no longer a future bet. Multiple players are deploying it commercially in the US right now. Not pilots. Operational assets delivering 8-24+ hours of discharge, zero degradation over 20,000 cycles, zero rare-earth dependency. The technology that closes India’s firm power gap is ready. The market entry pathway is not.
Gap 03 / The Trust and Access Gap
Localized Proxy for India’s Strategic AI Infrastructure
India’s infrastructure ecosystem is not moved by outreach. The developers and operators controlling strategic land banks and grid priority operate on a foundation of long-term local integration. This level of access cannot be manufactured from a pitch deck. We act as your localized operational proxy, bridging the gap between US technical standards and Indian infrastructure realities.
The Data
Why current power stacks fail to provide 24/7
green energy
For AI data centers with 100% renewable mandates, standard batteries (4-hour) cannot bridge multi-day power drops in solar or wind. While diurnal storage covers a single night, only resiliency technology (24-100+ hours) closes the 30% physical gap to provide true 24/7 firm power.
Case Study: Project Economics
100MW AI Data Center,India | $650M CAPEX | 25-Year Project Life
Technology Comparison
Why BESS alone will not
win this race
Four metrics that determine which storage technology survives a 25-year infrastructure contract in India. The numbers are not close after year seven.
BESS — Lithium-Ion
Current dominant storage in India
LDES — Thermal / Flow / Pumped Heat
What KyrosInfra facilitates into India
Sources: NREL Long-Duration Storage Report 2024, BloombergNEF LCOS Outlook, ESS Tech published cycle data, IRENA Battery Cost Report. Figures illustrative — actual performance varies by technology and deployment conditions.
Who We Bridge
Two sides of one
critical transaction.
Kyros Infra focuses on the players positioned to close the energy infrastructure gap defining the AI buildout, wherever that gap exists.
US Technology Seller Categories
LDES innovators seeking structured India market entry
Thermal Storage Innovators
Utilize graphite and carbon at extreme temperatures. 8-24hr discharge. No rare-earth exposure. 20,000+ cycle lifespan.
Long-Duration Flow Battery Firms
25-year storage from iron, salt, and water. No lithium. No China supply chain. Grid-scale economics, zero degradation.
Grid-Scale Pumped Heat Systems
Electricity stored as heat and cold in molten salt. Designed to replace fossil peaker plants with 24/7 firm output at utility scale.
Multi-Technology Integrators
Gravity, hydrogen, and battery dispatch under one AI-managed layer. Covers FTM utility and BTM C&I clients simultaneously.
India Infrastructure Buyers Categories
Operators who need firm power to fulfill contractual mandates
Pure-Play Data Center Operators
Uptime SLAs contractually exposed to grid instability. Need BTM firm power at the rack level, not contingency plans.
Global Hyperscaler India Teams
100% renewable mandates that India’s current hybrid stack cannot honor without long-duration BTM storage at campus level.
Tier-1 Real Estate & Industrial Developers
Strategic land parcels near Tier-1 hubs that become stranded assets without a bankable firm power roadmap attached.
C&I IPPs, Utilities & Tier-1 EPC Firms
Seeking non-lithium LDES to win 24/7 PPA contracts and validated hardware specs.
The Bridge
The definitive gateway for US energy innovators to integrate
with the Indian operators holding the capital and land to deploy.
Our Role
Defined Mandate.
Targeted Impact.
Kyros Infra is not a consultant you hire for a report. We work towards execution – the specific commercial, regulatory, and relational steps that turn a US technology pitch into a signed BTM energy partnership in India.
What US Companies Get
A sovereign path into a ₹1.4T market
-
EPC Partner
Identification
- — Tier-1 Engineering, Procurement & Construction firms qualified for thermal, mechanical, or flow-chemistry storage at industrial scale
-
PPA
Structuring
- — 15–25 year Power Purchase Agreement navigation in India's C&I sector, bypassing Discom delays by 18–24 months
-
BTM
Strategy
- — Determining the model which maximizes return and regulatory simplicity for your specific technology class
-
Institutional
Access
- — Introductions to developers and operators
The Bridge
KyrosInfra
Market Entry Architecture
BTM · C&I
EPC Partner Facilitation
Engineering · Procurement · Construction
PPA Structuring
15–25yr Power Contracts
LCOS Positioning
vs. conventional storage at grid scale
Strategic Advisory
Indo-US Corridor
WHAT INDIAN OPERATORS GET
The firm power answer your tenants demand
-
LCOS
Advantage
- — Solution that beats short-to-medium duration storage, without degradation cycles or replacement risk in your financial model
-
24/7 Uptime
Assurance
- — Always-on power that lets you actually honor 100% renewable SLAs hyperscalers require contractually
-
Supply Chain
Sovereignty
- — Zero rare-earth dependency, insulated from lithium price volatility and single-country geopolitical supply risk
Why Kyros Infra
Kyros Infra is the convergence of two decades of work that produced the exact profile this transaction requires. Dhiraj Seth built his operational foundation as Project Manager at Fidelity Investments ($1.5M+ resource portfolios, 20+ person teams) and Lead Systems Analyst at Wellington Management, the institutional fluency US enterprise executives recognize instantly. That foundation carried into twelve years building strategic relationships with India’s largest developers through Kreative Access, and now into Kyros Infra, where both skill sets converge on a single transaction: closing the gap between US energy technology and the infrastructure operators positioned to deploy it.
Fidelity Investments
Project Manager, Boston MA - 5+ years. $1.5M+ portfolios. Agile at institutional scale.
Wellington Management
Lead Systems Analyst, Boston MA - institutional investor platforms.
Kreative Access
Program Manager, Enterprise Delivery - 12+ years. Lodha, Kalpataru, Dosti, PepsiCo, Viatris.
PMP Certified
PMI. MS Computer Science. BE Computer Engineering.
Dhiraj Seth
Kyros Infra
Two decades building institutional trust globally, from managing multi-million dollar technology programs at Fidelity and Wellington, to producing strategic narratives for India’s largest real estate groups.
What started as creative content briefs with major real estate developers sparked a deep awareness of how land, capital, and power are converging to define the AI infrastructure decade. Kyros Infra is not a pivot. It is the logical destination of everything that came before it.
Dhiraj Seth, Senior Program Manager — Kyros Infra
Intelligence
Independent market briefs focused on the Indo-US energy corridor. This is our open-access analysis on the convergence of land, capital, and long-duration storage. Built for the operators and investors seeking a strategic roadmap of the Indian infrastructure landscape. No paywalls.
The 2030 Speed-to-Compute Arbitrage
India will reach 9GW of data center capacity by 2030. The bottleneck is not fiber, land, or capital – it is firm power. This brief maps why the first BTM LDES deployments will own the decade’s most valuable compute contracts.
The 2030 Speed-to-Compute Arbitrage
Why firm power — not land, capital, or fiber — is the binding constraint on India's AI data center decade, and what closing that gap is worth commercially.
India will add roughly 7.6GW of new data center capacity between now and 2030. The investment is committed, the land is being acquired, and the hyperscalers are signing letters of intent. But across every serious project, the same question is stalling final contracts: where does the 24/7 firm renewable power come from?
India's operators are not without power. They run real hybrid stacks — captive solar, wind PPAs, grid supply, and short-duration BESS. Nxtra has reached 49% renewable. Yotta targets 70%. This is genuine progress. But there is a ceiling: current solutions can deliver approximately 70–80% clean energy. The remaining 20–30% — the overnight hours, monsoon weeks, and unexpected demand spikes — is what breaks uptime SLAs and stalls hyperscaler contracts worth hundreds of crores.
Long-duration energy storage (LDES) — thermal batteries, iron-flow systems, pumped-heat technology — is the only class of technology that closes this gap. It is commercially operational in the US. It is not yet deployed in India at scale. The operators who secure LDES-backed firm power first will win the decade's most valuable compute leases.
Between March 2025 and April 2026, Indian data center operators announced approximately 3.5GW of planned new capacity. Andhra Pradesh and Telangana account for over 2GW, driven by AI-centric mega campuses in Visakhapatnam and Hyderabad. Maharashtra remains the most competitive market by project count. Chennai and Noida continue to attract hyperscale and enterprise buildouts.
Grid India's estimate puts India's total data center capacity at 1.2GW today, rising to 8–10GW by 2030. IEEFA places the 2030 figure at 9GW. Either way, the directional reality is the same: a 6x to 8x increase in eight years, concentrated in a handful of urban hubs whose grid infrastructure was designed for industrial loads, not for continuous 30–100kW-per-rack AI compute.
The core issue is not total power availability. India has 520GW of installed generation capacity as of January 2026, with over half now from non-fossil sources. The issue is power density and continuity at the point of consumption.
Traditional data centers run at 5–8kW per rack. AI training and inference workloads require 30–100kW per rack, with some next-generation GPU configurations projected to reach 250kW per rack by 2028. This is not a marginal increase — it represents a 10x to 50x jump in power requirements per unit of floor space, delivered continuously, with near-zero tolerance for interruption.
The grid was not designed for this concentration of demand. And solar — which now represents a significant share of India's renewable capacity — peaks at midday and produces nothing after sunset. For a data center running AI training jobs at 3am, solar is not a solution. It is a timing mismatch.
The hybrid approach being deployed across India's data center sector — grid power supplemented by solar PPAs and short-duration BESS — is a genuine and significant improvement over grid-only supply. But it has a structural ceiling.
Short-duration BESS (the lithium-ion systems currently being deployed) can bridge 2–4 hours of storage. This is sufficient for grid frequency regulation and short outage backup. It is not sufficient for the 8–16 hours of overnight darkness that defines the gap between solar generation and AI compute demand.
CEEW's February 2026 analysis confirms that achieving 24/7 operations on 100% renewables “remains technically challenging” under current approaches, with approximately 80% renewable supply being the realistic ceiling using geographically diversified solar-wind portfolios combined with battery storage. The remaining 20% requires something different.
That something different is LDES.
Long-duration energy storage technologies — specifically thermal batteries (Antora Energy, Fourth Power), iron-flow batteries (ESS Tech, Form Energy), and pumped-heat systems (Malta Inc.) — are designed to discharge power for 8 to 24+ hours continuously. They do not degrade meaningfully over 20,000 charge cycles. They do not rely on rare-earth minerals or the lithium supply chain running through China. And critically for India, they are designed to operate efficiently in high-ambient temperature environments, unlike lithium-ion systems that require active cooling to function in Mumbai or Chennai conditions.
These are not experimental technologies. Form Energy signed commercial contracts with Crusoe and Google in early 2026 for iron-air battery systems providing multiple days of electricity. Wood Mackenzie reported that LDES installations globally hit 15GWh in 2025 — a 50% increase year over year — with AI data centers emerging as the fastest-growing application category.
The technology is proven. The challenge is deployment in India's specific regulatory, EPC, and commercial environment.
The operators who solve the firm power equation first will have a decisive commercial advantage. Hyperscalers — AWS, Google, Microsoft — operate under 100% renewable energy mandates. They cannot sign long-term leases in facilities that cannot deliver on this commitment. Indian operators who can genuinely offer 24/7 firm renewable supply will unlock lease rates and contract durations that operators relying on hybrid stacks simply cannot access.
The addressable opportunity: India's 9GW data center build-out, with approximately 20% of power requirements representing the LDES deployment market, translates to roughly 1.8GW of long-duration storage required by 2030. At current LDES deployment costs, this represents a multi-billion dollar market that is effectively uncontested in India today.
Kyros Infra exists at the exact intersection this report describes. We facilitate the commercial, regulatory, and relational pathway that connects US LDES technology companies — whose products are ready to deploy — with Indian infrastructure operators and data center developers who have the land, the capital, and the contractual mandate to deploy firm power solutions now.
KyrosInfra Intelligence — February 2026 — www.KyrosInfra.com
Sources: IEEFA (June 2025), CEEW (February 2026), Grid India (2026), Wood Mackenzie LDES Report (2026), S&P Global Market Intelligence (2025), IEA Data Centre Report (2025).
The Regulatory Playbook for US Energy Tech in India
A practical breakdown of PPA structures, EPC qualifications, and BTM vs. FTM paths a US LDES company must navigate to reach commercial operation in India’s C&I market before 2028.
The Regulatory Playbook for US Energy Tech in India
A practical breakdown of the PPA structures, EPC qualifications, BTM vs. FTM deployment paths, and commercial mechanics a US LDES company must navigate to reach operational status in India's C&I market before 2028. Written for operators, not lawyers.
India is not a difficult market to understand. It is a difficult market to navigate without the right relationships and the right commercial structures in place before you start. US energy storage companies that have approached India through standard channels — cold outreach to procurement teams, responses to public tenders, distributor agreements — have consistently found that the market does not respond to those approaches at the speed or depth required for a large-scale infrastructure deployment.
The companies that succeed in India's energy infrastructure market do so through a combination of four things: a trusted local partner who can facilitate introductions at the right level, a clear understanding of whether the deployment is Behind-the-Meter (BTM) or Front-of-the-Meter (FTM), a structured approach to Power Purchase Agreements that reflects India's specific regulatory environment, and an EPC partner who has the technical capacity to handle the specific hardware requirements of the technology being deployed.
This report covers each of these four areas in practical terms.
The single most important strategic decision for a US LDES company entering India is whether to pursue Behind-the-Meter or Front-of-the-Meter deployment. The two paths have very different regulatory requirements, commercial structures, customer relationships, and timelines.
Behind-the-Meter (BTM) means the storage system is installed at the data center or industrial facility itself, serving that specific customer's power needs directly, without flowing through the public grid. For AI data center applications, BTM is typically the faster and more commercially straightforward path. The customer is a single identifiable entity with a clear power need. The regulatory approvals are simpler. The commercial structure — a direct Energy Service Agreement or an enhanced PPA with the facility operator — does not require Discom involvement. The technology can be sized specifically for the facility's load profile.
Front-of-the-Meter (FTM) means the storage system connects directly to the state grid, selling power to the grid or to multiple offtakers through a licensed power trading mechanism. FTM projects are more complex, involve CERC (Central Electricity Regulatory Commission) or SERC (State Electricity Regulatory Commission) approvals, require grid connection agreements with the relevant TRANSCO, and operate under the framework of the Electricity Act 2003 and its amendments. FTM is appropriate for utility-scale deployments targeting C&I Independent Power Producers, large conglomerate utilities like Adani Green or Reliance New Energy, or projects that want to serve multiple data center customers from a single large asset.
Power Purchase Agreements in India's C&I sector operate under a different framework than the long-term utility PPAs that US companies are familiar with from the American market. Key distinctions:
Open Access: Under India's open access framework, large consumers (typically above 1MW) can procure power directly from a generator rather than through the state Discom. This is the mechanism through which most serious data center operators in India are structuring their renewable PPAs today — bypassing Discom procurement and negotiating directly with independent power producers. For an LDES company, open access provides the commercial pathway to sign directly with a data center operator or industrial facility for BTM storage services.
Round-the-Clock (RTC) PPAs: The emerging commercial structure that LDES enables is the Round-the-Clock PPA — a contract where the power supplier guarantees continuous, 24/7 supply at a fixed tariff, combining renewable generation with storage to back the commitment. SECI (Solar Energy Corporation of India) has begun tendering RTC renewable PPAs, and private RTC contracts are increasingly being negotiated between renewable developers and hyperscalers. LDES is the enabling technology that makes RTC commitments credible. Tariffs for firm and dispatchable renewable energy are currently at Rs 4.98–4.99 per unit, below the Central Energy Authority's median natural gas tariff of Rs 5.4 — making LDES-backed RTC power commercially competitive today.
Duration: Indian infrastructure PPAs typically run 15–25 years. This is both an advantage (long-term revenue certainty) and a negotiation complexity (the terms must be right from the start). Getting a first PPA signed in India generally requires 9–18 months of commercial negotiation, regulatory approvals, and counterparty due diligence. Having a trusted local partner who can compress this timeline by facilitating introductions and managing the relationship process is the difference between 9 months and 24 months.
Engineering, Procurement, and Construction firms are the operational delivery mechanism for any LDES project in India. A US LDES company cannot deploy hardware in India without an Indian EPC partner who has the technical capability, supply chain relationships, and regulatory credentials to manage the installation.
India's Tier-1 EPC landscape for energy infrastructure is dominated by firms like Tata Projects, Larsen & Toubro, Thermax, and Sterling & Wilson. These are sophisticated, large-scale operators with experience in utility and industrial power projects. However, they require significant technical validation before they will commit to handling a new technology class — particularly one like thermal storage or iron-flow chemistry that they have not previously deployed in India.
The practical implication for a US LDES company: you need to provide detailed technical documentation, climate adaptation specifications for Indian conditions, and ideally a reference installation from another market before a Tier-1 EPC firm will engage seriously. Mid-tier EPC firms specializing in renewable energy — including several firms active in solar and wind project delivery — are often more willing to engage on new technology classes and can be the right starting point for a first BTM deployment.
For a US LDES company targeting its first commercial deployment in India, a realistic timeline from initial market entry to signed commercial contract looks like this:
Months 1–3: Market entry architecture. Defining the target segment (BTM data center vs. FTM utility), identifying the anchor customer or partner, establishing the legal entity structure in India (typically a wholly-owned subsidiary under India's FDI framework for the energy sector, which permits 100% FDI under the automatic route).
Months 3–6: Technical validation and partner engagement. Providing technical documentation to prospective EPC partners. Completing climate adaptation assessment. Beginning commercial negotiations with the anchor customer or offtaker. Engaging with state-level regulatory bodies to understand the grid connection and open access approval pathway for the target geography.
Months 6–12: Commercial term sheet and regulatory applications. Negotiating the term sheet for the BTM Energy Service Agreement or FTM PPA. Filing for open access approvals with the relevant State Discom. Initiating EPC partner qualification process. Beginning the grid connection application if FTM.
Months 12–18: Final contract execution and project commissioning preparation. Signed commercial contract, EPC agreement, and equipment import/supply chain in place. First physical deployment begins.
Kyros Infra is not a regulatory consultant. We do not file paperwork or provide legal advice. What we do is compress the relationship timeline that sits upstream of every regulatory and commercial step described in this report.
The introductions to the right data center operators, the right EPC firms, and the right power producers — introductions that come from twelve years of institutional relationships built through Kreative Access's work with Lodha, Kalpataru, Hiranandani, and others — are what turn an 18-month process into a 9-month process.
If you are a US LDES company evaluating India, the conversation starts at www.KyrosInfra.com. A Strategic Briefing is a 45-minute session to map the specific commercial pathway for your technology in India's current market environment.
KyrosInfra Intelligence — April 2026 — www.KyrosInfra.com
Sources: CERC Regulatory Framework for Energy Storage (2024), SECI RTC Tender Documentation, IEEFA India Data Centre Report (June 2025), Ministry of Power Open Access Rules (2022), India FDI Policy (2024), CEEW India Data Centre Infrastructure Report (February 2026).
The next generation of infrastructure execution in India
The future of India’s AI infrastructure built not by isolated players, but by those who control the Land-to-Power, as an integrated stack.
The Land-to-Power Stack
Why India's AI infrastructure race will be won or lost on sequencing, and what it means for developers, energy firms, and capital allocators entering the market now.
India's AI infrastructure race is not constrained by ambition, capital, or even technology. It is constrained by sequencing failure. The industry continues to treat land acquisition, power provisioning, and compute deployment as parallel workstreams. In reality, they form a tightly coupled stack. Misalignment across this stack is now the single biggest cause of delay in data center execution.
Most data center strategies in India still follow a legacy development model:
- Secure land
- Design capacity
- Apply for grid power
- Backfill energy solutions later
This linear approach is structurally broken in the AI era.
The next generation of infrastructure execution in India must operate as an integrated stack:
- Land Layer: Zoning, aggregation, and proximity to transmission corridors
- Power Layer: Grid access, captive generation, and storage integration
- Compute Layer: Hyperscaler and enterprise demand alignment
- Capital Layer: Structured financing tied to execution certainty
Each layer is interdependent. Failure to synchronize them results in multi-year delays.
- Land acquired without transmission feasibility studies
- Power approvals delayed due to lack of integrated planning
- Compute contracts signed before energy certainty
- Capital deployed into partially de-risked assets
A new class of players is beginning to emerge, defined not by ownership of assets, but by control over sequencing.
- Pre-aligned land with power pathways
- Energy-first site selection
- Integrated execution planning across stakeholders
These players are not developers in the traditional sense. They are stack orchestrators.
Market entry strategies that focus solely on technology or regulatory navigation are insufficient.
Success in India will require:
- Early alignment with land and developer ecosystems
- Integration into project timelines before site finalization
- Participation in structuring, not just supplying
Kyros Infra operates at the intersection of land intelligence, energy structuring, and global technology integration.
By aligning these layers early, the platform reduces execution risk and accelerates speed-to-compute.
Over the next 24 months, the market will bifurcate:
- Projects that remain stuck in sequential execution
- Projects that move forward through integrated stack alignment
The gap between the two will define who captures India's AI infrastructure upside.
The constraint is no longer identifying opportunity.
The constraint is executing across interdependent systems.
KyrosInfra Intelligence — May 2026 — www.KyrosInfra.com
The Monsoon Stress Test
India’s AI data center operators enter monsoon season 2026 having made significant renewable energy commitments to hyperscaler clients and having never been tested on those commitments at full AI compute density during an extended weather event.
The Monsoon Stress Test
India's AI data center operators enter monsoon season 2026 having made significant renewable energy commitments to hyperscaler clients — and having never been tested on those commitments at full AI compute density during an extended weather event. This brief maps what the southwest monsoon does to each layer of India's current clean energy stack, quantifies the storage duration gap it creates, and explains why the next 120 days are the sector's most consequential operational test since hyperscaler demand arrived in the country.
India's southwest monsoon runs June through September, covering every primary data center geography in the country: Mumbai, Pune, Hyderabad, Chennai, Bangalore, and the rapidly expanding NCR corridor. This is not a regional event operators can route around. It is a nationwide disruption to the clean energy stack the entire sector has built its renewable commitments upon.
Solar generation drops 40 to 70 percent from peak output levels during active monsoon periods. Mumbai receives approximately 2,200mm of rainfall concentrated in these four months. The relevant condition is not rainfall itself — it is the persistent, multi-day cloud cover that accompanies low-pressure systems. These are not the afternoon clouds that short-duration BESS is designed to bridge. They are 48 to 72-hour windows of near-zero solar irradiation that occur four to six times each monsoon season across Maharashtra, Telangana, and Karnataka — India's three most active data center markets.
Wind generation provides no reliable backstop for most of India's data center hubs during monsoon. Tamil Nadu's coastal resources do peak during southwest monsoon. But the inland markets — Hyderabad, Pune, Greater Noida — see variable, non-dispatchable wind patterns that cannot substitute for the solar capacity being suppressed by the same weather system.
Grid stability degrades across the exact hubs where data centers are concentrated. Monsoon simultaneously drives peak residential and commercial cooling loads. Distribution infrastructure under seasonal stress generates elevated frequency disturbances that trigger UPS events even in well-protected, well-capitalised facilities.
The core calculation for any AI data center operator during monsoon is straightforward. An extended low-pressure system over Maharashtra or Telangana suppresses solar generation for 48 to 72 hours. The lithium-ion BESS systems comprising India's current storage stack bridge 2 to 4 hours of supply gaps under favourable charge conditions — and they require solar generation to recharge. When solar drops to near zero for 72 hours, BESS depletes and cannot recover until the weather clears.
This is not a problem that more BESS resolves. Stacking additional lithium-ion capacity addresses duration at enormous capital cost while introducing proportionally larger fire risk in high-ambient Mumbai and Chennai conditions — and still failing to close a multi-day gap. BESS is a sprint technology. The monsoon duration gap requires a fundamentally different technology class.
Short-duration BESS (Li-ion): 2–4 hour bridge. Exhausted by hour 4 of a 72-hour event.
Extended BESS stack (stacked Li-ion): 6–8 hour bridge. Exhausted by hour 8. Cannot self-recharge while solar is suppressed.
Long-Duration Thermal or Flow Storage: 8 to 100+ hour discharge duration. Closes the monsoon gap entirely.
Pumped Hydro (where geographically available): 8–24 hour bridge. Closes the gap but is geographically constrained to specific hill state topographies not adjacent to India's primary data center markets.
India's data center operators publish their renewable energy penetration statistics as annual averages. These figures represent genuine, hard-won progress. But annual averages mask seasonal variance in ways that hyperscaler due diligence teams are beginning to ask about directly.
A facility that delivers 80% renewable energy across a full calendar year may deliver 35 to 40% renewable energy during the six worst weeks of monsoon. For a hyperscaler client whose 100% renewable mandate carries no seasonal carve-out, the annual average is contractually irrelevant. The worst-case period defines the SLA exposure.
The facilities approaching high renewable penetration have done so serving traditional enterprise IT loads — more predictable, less dense, more tolerant of brief interruptions. AI training and inference workloads running at 40 to 100kW per rack, continuously, with zero tolerance for unplanned interruption, represent a structurally different operational profile. The monitoring instruments, the SLA definitions, and the contractual remedies written into hyperscaler leases reflect this.
Long-duration energy storage resolves the monsoon arithmetic that short-duration BESS structurally cannot. Thermal storage systems and iron-flow batteries charge during the high-irradiation periods that bookend active monsoon events — India's intense April and May pre-monsoon peak, the inter-monsoon breaks, the clear-weather windows between low-pressure systems — and discharge for 8 to 24+ hours continuously at consistent output through the extended generation gaps.
The India climate fit is not incidental. Thermal storage thrives in high-ambient temperature environments. The same Mumbai and Chennai conditions that require active HVAC for lithium-ion systems — and impose a 15 to 20% parasitic cooling load on net efficiency — are operationally neutral for graphite-based thermal batteries. There is no flammable chemistry. No fire risk at 40°C. No cooling overhead degrading the energy balance during the exact periods when grid stability is most pressured.
Iron-flow and pumped-heat storage systems do not accumulate degradation through repeated deep-discharge cycles — precisely the operating profile that monsoon season demands. A system fully discharged during a 72-hour weather event and fully recharged during the subsequent clear-weather window performs identically in its first monsoon season and its twenty-fifth. The monsoon cycle does not cost LDES anything. It costs BESS its remaining calendar life.
LDES systems sized for monsoon resilience deliver more than seasonal weather coverage. They provide the always-on firm power foundation that 100% renewable SLAs require across every day of the year — resolving the overnight solar gap, the morning wind lull, and the demand spike that no combination of renewable generation can address without duration storage behind it.
The operators who can demonstrate genuine monsoon resilience have a categorically different infrastructure product than those who cannot. This distinction is beginning to surface in the commercial questions hyperscaler procurement teams are asking before committing to long-term Indian leases.
The questions are specific: What is your worst-case renewable percentage during the June to September window? What is your demonstrated BESS bridge duration during a multi-day weather event? What is your grid draw exposure when solar is near zero for 72 hours? What remedies apply when renewable SLA thresholds are not met during monsoon season?
These are not theoretical questions. They are the product of hyperscaler teams observing India's renewable energy claims closely enough to understand where the seasonal exposure lives — and building contract language that protects them from it. The operators who can answer these questions with an operational LDES deployment — not a future roadmap — will unlock contract tenors, lease rates, and hyperscaler relationships that define India's compute economics for the next decade.
India's infrastructure races consistently produce one or two reference deployments that set commercial terms for everyone who follows. The operator who achieves and demonstrates genuine monsoon-proof firm power first does not merely win one contract. They become the benchmark against which every subsequent hyperscaler procurement decision in India is measured. The window to be that reference deployment is not unlimited.
The monsoon stress test is the natural validation event for the Kyros Infra thesis. The operators who understand — before hyperscalers formalise the audit — that the June to September window is a live exposure event, and who begin structuring LDES deployments now to be operational before monsoon season 2027, are the exact counterparts Kyros Infra is in active conversation with.
The technology that closes the monsoon gap is commercially proven in the United States and operationally ready for India deployment. The 12 to 18-month commercial and regulatory pathway to first deployment means the decisions made in the second half of 2026 determine whether a facility is genuinely monsoon-proof by 2027 — or still managing the exposure with grid draws and contractual disclaimers.
Kyros Infra facilitates the specific commercial and relational pathway connecting US LDES companies with Indian infrastructure operators who understand what is at stake. A Strategic Briefing is a 45-minute session to map whether your position — technology seller or infrastructure operator — has a viable path to deployment before the next monsoon season resets the market. The conversation starts at www.KyrosInfra.com.
KyrosInfra Intelligence — June 2026 — www.KyrosInfra.com
Sources: IMD Southwest Monsoon Data (2024–2025), CEEW India Data Centre Infrastructure Report (February 2026), IEEFA India Data Centre Report (June 2025), Grid India Power Sector Report (2026), Wood Mackenzie LDES Outlook (2026), NREL Long-Duration Storage Report (2024), BloombergNEF LCOS Outlook. Figures illustrative — actual performance varies by technology class, site geography, and deployment configuration.
India's Hyperscaler Bottleneck
India’s hyperscale build-out is no longer gated by whether power exists in a given region. It is gated by how long it takes to get connected to it. This brief maps the shift from aggregate power availability to grid energization timelines as the defining site-selection variable, and explains why the operators who can decouple from that timeline are already pulling ahead of the ones still waiting in line.
The Time-to-Power Race
India's hyperscale build-out is no longer gated by whether power exists in a given region. It is gated by how long it takes to get connected to it. This brief maps the shift from aggregate power availability to grid energization timelines as the defining site-selection variable, and explains why the operators who can decouple from that timeline are already pulling ahead of the ones still waiting in line.
For the past two years, the India data center conversation has centered on a single question: is there enough power. That question is now answered. India's electricity generation base is expanding fast enough to support the sector's growth on paper.
The question that actually determines who wins the next round of hyperscaler contracts is different: how long does it take to get energized at the specific site you have committed capital to.
The old data center site selection model ran on three questions: is there land, is there power in the region, is there fiber. All three could be answered yes for dozens of sites across Maharashtra, Telangana, and the NCR corridor.
The model breaks down at the moment of energization. A site with confirmed regional power availability can still sit for 18 to 36 months waiting on a DISCOM connection, a transmission upgrade, or a substation slot, while a competing site with a cleaner interconnection path goes live in a fraction of that time.
Industry analysis published this year makes the point directly: hyperscale growth through 2030 will be shaped less by how much power India generates and more by how quickly specific locations can be connected and energized. That is a locational, project-by-project constraint, not a national one.
Three forces are compounding the backlog at the exact moment demand is accelerating:
- Mega-campus scale. Single-site capacity requirements now routinely exceed 1GW. A request of that size does not slot into existing substation capacity. It requires new transmission planning, which runs on a multi-year cycle regardless of how fast the developer wants to move.
- Speculative reservations. As land-banking accelerates, DISCOMs are fielding interconnection requests for sites that may never break ground, congesting the queue for operators with real capital committed and real timelines to hit.
- Concentration in a small number of hubs. Mumbai, Pune, Hyderabad, and Chennai account for the overwhelming majority of demand, which means the queue congestion is not distributed evenly across India's grid. It is stacked in exactly the four places every hyperscaler wants to be.
None of this shows up in a regional power availability study. It only shows up when a specific project tries to move from land acquisition to energized capacity, and finds the timeline has quietly become the binding constraint.
The queue congestion problem is not unique to India, and the regulatory response building in the US is worth tracking closely. Utilities managing overloaded interconnection queues have started shifting cost exposure onto the customers making the requests, requiring minimum committed usage regardless of whether that capacity is ever fully utilized. The stated purpose is to filter out speculative requests and protect the timeline for developers who are actually building.
India's DISCOMs are watching the same dynamic. As gigawatt-scale campuses multiply, the incentive to apply similar discipline to India's interconnection queue grows. Operators who have reserved capacity without committed capital behind it should not assume that position stays free indefinitely.
Two categories of operator are emerging from this constraint, and the gap between them is becoming the clearest predictor of who signs the next round of hyperscaler leases.
Grid-sequential operators follow the traditional path: secure land, apply for connection, wait for energization, then begin operations. Their timeline is set by the DISCOM queue and the transmission planning cycle. It is not a timeline they control.
Power-independent operators deploy firm, dispatchable capacity on site (captive generation paired with duration storage) that lets them begin operations on a partial grid allocation, or with materially reduced grid dependency, while the full interconnection process runs in parallel rather than as a precondition.
Firm, on-site duration storage changes the sequencing math in a specific way. A facility does not need full grid energization to go live if it can run critical load on dispatchable storage charged from whatever partial power access it already has, whether captive generation, a preliminary grid allocation, or renewables paired with duration.
This matters most in the exact hubs where the queue is worst. A Mumbai or Hyderabad site facing a 24-month wait for full transmission capacity can bring AI compute online on a fraction of that timeline if the power architecture is designed for firm output from day one, rather than treating storage as a resilience layer added after grid connection is secured.
The strategic implication is a sequencing reversal. Instead of land, then power, then storage as an afterthought, the fastest-moving operators are designing the storage and generation stack first, and letting full grid interconnection catch up in parallel rather than gating the schedule.
Hyperscaler procurement teams are beginning to ask a new question in site diligence, alongside renewable percentage and uptime history: what is your actual time to full energized capacity, and how much of your operational timeline depends on a DISCOM queue you do not control.
- Operators who can answer with a firm-power architecture and a queue-independent go-live date will price and contract faster.
- Operators still quoting regional power availability as their answer will find that number increasingly irrelevant to the diligence conversation.
- Capital allocators evaluating India data center opportunities should treat interconnection queue position as a distinct, quantifiable risk factor, separate from land and separate from renewable sourcing.
This is the thesis Kyros Infra has been building toward since the first report in this series. Land-to-power sequencing and monsoon resilience are both, at their core, expressions of the same underlying problem: operators who treat power as something they wait for lose to operators who treat power as something they architect.
The technology that enables a queue-independent timeline (firm, dispatchable, long-duration storage) is commercially proven in the United States and ready for India deployment now. The decisions made in the second half of 2026 determine which operators are still waiting on a connection date in 2028, and which ones are already running.
Kyros Infra facilitates the specific commercial and relational pathway connecting US LDES companies with Indian infrastructure operators who are ready to build the power stack first. A Strategic Briefing is a 45-minute session to map your actual time-to-power exposure and whether a queue-independent path is viable for your site. The conversation starts at www.KyrosInfra.com.
KyrosInfra Intelligence — July 2026 — www.KyrosInfra.com
Sources: Frost & Sullivan Grid Connectivity and Hyperscale Data Center Growth Analysis (May 2026), Mordor Intelligence India Data Center Power Market Report (2026), IEEFA India Data Centre Sector Report (April 2026), JLL India Data Center Capacity Outlook (March 2026), S&P Global Energy Interconnection and Utility Tariff Analysis (2025-2026). Figures illustrative. Actual timelines vary by site, DISCOM jurisdiction, and transmission planning status.
Ensuring Sustainability with India's AI DC Boom
While AI data center boom in India is working on answering the power question, it is now running into a harder one: whether that power, and the water used to cool it, can be sourced without contradicting the renewable and resilience claims operators are making to the hyperscalers signing their leases.
The Sustainability Underwrite
While AI data center boom in India is working on answering the power question, it is now running into a harder one: whether that power, and the water used to cool it, can be sourced without contradicting the renewable and resilience claims operators are making to the hyperscalers.Sustainability has quietly moved from a marketing line to a financing and procurement test, and most operators are not yet built to pass it.
India's data center market is on a clear growth path, from roughly ten billion dollars today toward twenty two billion by 2030, with installed capacity moving from under two gigawatts to four to five gigawatts over the same period. That growth story is well told. What is less well told is the constraint sitting underneath it: the same 24x7, high-density load that makes AI infrastructure valuable also makes it one of the most resource-intensive tenants a state grid and a local watershed can host.
The operators who treat sustainability as a compliance checkbox to satisfy after the fact are going to find themselves explaining a widening gap between what they claimed and what they consumed. The operators who architect for it from day one are the ones who will keep clearing hyperscaler diligence and ESG-linked financing screens as those screens get stricter through 2027 and beyond.
India currently hosts under 300 data centers consuming around half a percent of national electricity generation. That figure is projected to climb toward three percent by 2030 as AI adoption accelerates, a six-fold increase layered onto a grid that is simultaneously trying to absorb gigawatt-scale hyperscale campuses in a handful of concentrated hubs.
Water tells a similar story. India's data centers already draw an estimated 150 billion litres a year, a figure that industry analysts expect to more than double by 2030. A single conventional 100 megawatt facility can consume close to two million litres a day. Scale that to the 8 to 10 additional gigawatts of capacity forecast to come online by 2030, concentrated overwhelmingly in Mumbai, Hyderabad, Pune, and Chennai, and the water math becomes as binding a constraint as the power math.
Every major India data center announcement now carries a renewable energy commitment. Reliance, Google, and Adani have all attached clean power targets to their build-outs, and Adani has framed its entire 5 gigawatt expansion as renewable-powered. These commitments are real and directionally correct. They are also incomplete on their own.
AI workloads need continuous, high-density compute with effectively zero tolerance for interruption. Solar and wind are variable by nature and cannot deliver that profile alone. Industry voices in the sector have been candid about the resulting gap: batteries are improving and increasingly viable for short-duration backup and load balancing, but long-duration, uninterrupted operation still leans on natural gas, thermal power, or diesel generation sitting behind the renewable share on the label.
That reliance on fossil backup for the hours renewables cannot cover is the quiet contradiction inside most current sustainability claims. A facility can be genuinely majority-renewable on an annual energy accounting basis and still be running diesel gensets every time the sun or wind profile dips, which is exactly when grid stress and local air quality impact are highest.
Cooling technology choice is where the energy and water constraints collide directly, and there is no cooling method that avoids both at once.
- Water-based and evaporative cooling is more energy efficient but depends on a reliable water supply, which is precisely what is scarce in the hub cities carrying most of the demand.
- Air-based cooling reduces water draw but increases energy consumption, which pushes more load onto a grid already straining to serve gigawatt-scale campuses.
- Liquid cooling, now being adopted by operators including CtrlS at its Kolkata campus, offers a genuine way out of the trade-off for high-density AI racks, cutting both water use and energy draw relative to legacy systems, though it requires upfront capital and design commitment most conventional builds were not engineered for.
The state-level policy response is starting to catch up. Karnataka is evaluating incentives for next-generation cooling technology that reduces water consumption and improves energy efficiency as part of its forthcoming data center policy. That is the right instinct, applied so far in only a handful of states.
India does not yet have a national data center policy setting common sustainability standards. States have moved individually, and unevenly: only a minority of state data center policies currently contain any sustainability parameters at all, covering renewable energy use, water recycling, rainwater harvesting, or responsible cooling practices.
Until a national framework arrives, the underwriting standard is being set informally by the parties with the most leverage: the hyperscalers signing anchor leases and the capital providers financing the build-out. Their diligence checklists are becoming the de facto national policy, and they are converging on the same questions: what is your actual renewable share on an hourly basis, what is your water usage effectiveness, and what happens to your uptime commitment when the renewable share dips.
This is the same underlying constraint Kyros Infra has been tracking across land-to-power sequencing, monsoon resilience, and grid queue timelines: operators who treat power as something they wait for or patch around lose ground to operators who treat power as something they architect.
Long-duration energy storage paired with renewable generation directly addresses the gap the industry itself has identified. It lets a facility bank solar and wind output and dispatch it on demand through the hours renewables cannot cover, removing the need for diesel or gas as the default answer to intermittency. That converts an annual renewable percentage into an hourly one, which is exactly the shift hyperscaler diligence is now asking for.
The same firm-power architecture also indirectly eases the water constraint. Facilities that are not built around emergency diesel backup can be designed around efficient cooling systems, including liquid cooling, from the outset rather than retrofitted later, because the power source behind them is already reliable enough to support a tighter, more capital-intensive cooling design.
- Hyperscaler procurement diligence will keep moving from annual renewable percentage toward hourly, verifiable clean power delivery, the same shift already underway in more mature markets.
- Water usage effectiveness will become as standard a diligence metric as power usage effectiveness, particularly for any site in Mumbai, Hyderabad, Pune, or Chennai.
- Operators still leaning on diesel backup as their uninterrupted-power answer will find that dependency increasingly visible and increasingly costly to explain, both to regulators and to capital providers.
- A national data center policy setting sustainability floors is likely within this window. Operators who are already ahead of where that floor will land face a lighter retrofit bill than operators who are not.
Sustainability in India's AI data center sector is not a separate track from the power and grid challenges covered in this series. It is the same problem viewed from a different angle: operators who build firm, dispatchable capacity into their power stack from the start are the ones able to hit renewable delivery targets, avoid diesel dependency, and justify the efficient, low-water cooling systems the next wave of due diligence will expect.
KyrosInfra Intelligence — August 2026 — www.KyrosInfra.com
Sources: Observer Research Foundation, "India's AI Data Centre Expansion: Balancing Growth with Sustainability" (July 2026); Council on Energy, Environment and Water (CEEW) India Data Centre Study; International Energy Agency grid connectivity analysis; Mongabay India data centre water reporting (May 2026); The Energy Pioneer, "The Race To Host AI: Data Centres In Water-Scarce India" (2026); India Briefing data center cooling market analysis (2026); ResearchAndMarkets India Data Center Market Report (2026). Figures illustrative. Actual water and energy intensity vary by site, cooling technology, and climate zone.
The Reliability Premium
Why Elon Musk had to slow down the fastest AI data center build in history, and what the reversal means for every operator underwriting firm power commitments globally.
The Reliability Premium
Why Elon Musk had to slow down the fastest AI data center build in history, and what the reversal means for every operator underwriting firm power commitments globally.
For two years, the dominant narrative in AI infrastructure has been velocity. Musk set that narrative himself: Colossus, the Memphis campus built for SpaceX's AI compute business, went from groundbreaking to operational in roughly 122 days, a timeline Nvidia's Jensen Huang publicly called superhuman. It became the build-speed benchmark every hyperscaler and developer measured itself against.
That same campus has now forced Musk into the opposite move. Following reliability failures and a construction-related outage at the Memphis site that knocked Grok offline and disrupted third-party compute customers including Anthropic and Google, Musk pulled more than a dozen data center leaders and replaced them with engineers from SpaceX's rocket and Starlink divisions. Their mandate: install the backup power, cooling redundancy, and pre-launch testing the original build sequence skipped.
Colossus was built to win the speed race. Compressing the timeline between land acquisition and hyperscaler revenue is the same instinct driving every developer racing to lock a compute tenant before a competing site does, and Musk executed it better than anyone. What the 122-day timeline did not include was the redundancy layer: backup power depth, cooling margin, and commissioning testing that do not show up in a groundbreaking-to-live headline but determine whether a facility holds up once it is running continuous, full-density AI compute rather than a demo workload.
Musk acknowledged the gap directly, writing on X that SpaceX was taking corrective action to ensure it did not happen again. The response was not a pause on AI ambition. It was a personnel and process overhaul: engineers from rocket production and Starbase operations moved into a division that had been run by data center specialists, tasked specifically with building in the reliability layer the first build sequence left out. Expansion into new sites, including a planned Texas campus, has been pushed back with no confirmed date.
SPCX stock dropped roughly 4 percent when the reliability story broke, then recovered within a day as investors reframed it: a company correcting a sequencing error before it compounds is a materially better risk profile than one papering over it. Goldman Sachs has SpaceX's AI-related revenue climbing from roughly 3.2 billion dollars in 2025 to 322 billion by 2030, and that model depends entirely on the compute infrastructure actually staying up.
That is the diligence lens hyperscalers and capital allocators are increasingly applying to every AI infrastructure asset, not just Musk's. A facility that hits its construction deadline but cannot sustain uptime under full compute load is not ahead of a facility that took longer and built the redundancy in from day one.
Data center sector globally has spent 2026 competing on the same axis Musk just got a public lesson in: who can get land, power, and a signed hyperscaler lease fastest. Focused on India, this report series has already flagged the risk in that framing, from the land-to-power sequencing failures covered in Report 03, to the monsoon duration gap in Report 04, to the grid queue timelines in Report 05. Colossus is the clearest evidence yet that the same failure mode is not theoretical.
The operators building right now are making the same sequencing decision under the same competitive pressure to move fast. The diligence questions hyperscalers are already asking, covered in Report 06, about hourly renewable delivery and water usage effectiveness are about to be joined by a more direct one: what is your backup power and cooling redundancy specification, and was it designed in before construction or added after an outage forced the issue, the way it was at Colossus.
Long-duration energy storage is the mechanism that lets an operator answer that question correctly on day one instead of retrofitting it the way Musk's team is doing now. A facility architected around firm, dispatchable power from the start does not need a team of rocket engineers to arrive later and rebuild the redundancy layer. The redundancy is the power architecture, not an emergency addition bolted onto it after a public outage.
This is the same underlying thesis this series has tracked from a different angle in every report to date: operators who treat power as something they architect from the outset outperform operators who treat it as something they wait for or patch around. Colossus is the highest-visibility proof point of that thesis to date.
- Hyperscaler site diligence will begin asking for backup power and cooling redundancy specifications as a named line item, not an assumed feature of any serious build, the same standard Musk's outage just forced onto SpaceX's own compute customers.
- Build-speed benchmarks will stop being treated as a standalone competitive advantage once separated from a demonstrated uptime record under full compute load.
- Capital allocators funding India's next wave of data center capacity should treat a documented firm-power design, not a construction timeline, as the underwriting signal that predicts which facilities avoid a costly mid-build correction.
- Operators who can show a firm-power architecture designed in from the land-to-power planning stage will be positioned as the reference deployments the next round of hyperscaler leases gets measured against, the same dynamic identified in Report 04.
Colossus is the clearest external validation yet of the sequencing argument this series has made since Report 01. Speed-to-compute is not won by the fastest construction timeline, not even Musk's 122-day record. It is won by the operator whose power architecture does not need to be rebuilt six months after going live. Long-duration storage, deployed from the design stage rather than added after a failure, is what makes that distinction real rather than theoretical.
KyrosInfra Intelligence — September 2026 — www.KyrosInfra.com
Sources: The Information, SpaceX data center infrastructure reporting (September 2026); Reuters, SPCX market coverage (September 2026); Goldman Sachs, SpaceX AI infrastructure revenue projections (June 2026); Wood Mackenzie LDES Outlook (2026); NREL Long-Duration Storage Report (2024). Figures illustrative.
Managing the power dependencies you cannot contract away
In AI data center programs, the milestone that gets celebrated is rarely the one that sets the go-live date. A signed supply agreement or an interconnection approval confirms a place in line. It does not say when the facility will actually be energized.
Approved Is Not Energized
Why a signed supply agreement does not set the go-live date, and how AI data center programs should manage the power dependencies they cannot contract away.
In AI data center programs, the milestone that gets celebrated is rarely the one that sets the go-live date. A signed supply agreement or an interconnection approval confirms a place in line. It does not say when the facility will actually be energized, and energization is the date that starts revenue.
Programs that manage the distance between those two dates on purpose finish closer to plan. Programs that treat approval as the finish line find out late, and the cost of finding out late lands on the capital side of the program, not only on the schedule.
In November 2025, Bloomberg reported that two large data centers in Santa Clara, California were finished and empty. Digital Realty's SJC37 and Stack Infrastructure's SVY02A were each designed for about 48 megawatts of critical load, close to 100 megawatts together. Both had supply agreements with Silicon Valley Power, the city-owned utility, and both were waiting on upgrades to the local network.
The utility said a 450 million dollar system upgrade was on schedule for completion in 2028, and that it was sequencing power delivery among customers as new substations and transmission lines came online. Stack said its agreement reflected new power arriving through 2027. Digital Realty's project had received planning permission in 2019, roughly six years before the report. Both companies said they were coordinating with the utility on phased power delivery as upgrades progressed.
No one in this account made an obvious mistake. The buildings were complete, the agreements were in place, and the utility was investing. The constraint was the utility's own capacity timeline, which the developers could not buy, contract, or escalate their way around.
Vendors and utilities sit side by side on a program schedule, but they behave differently. OEMs and EPC contractors work under contracts that carry penalties, replacement rights, and enforceable dates. Where a program has that leverage, the contract absorbs part of the delivery risk automatically.
Power and water utilities are regulated public infrastructure. There is no penalty to claim and no one to replace. Where leverage does not exist, the program has to size its schedule buffer to the actual risk and track the leading indicators itself, before the formal update arrives. Report 05 treated time-to-power as a site selection metric. This report picks up after the site is chosen, when the same dependency has to be managed through to energization.
Approval confirms a place in the queue. Between approval and the day a site can actually connect, several technical steps can remain, depending on the electric utility handling the interconnection:
- Relay coordination and protection settings.
- Revenue metering installation and calibration.
- Real-time telemetry to the utility's control center.
- The synchronization test that proves the site can connect without tripping anything.
Not every project hits every step, but most utilities have some version of this list. Each step runs on the utility's own clock, and none of it appears on the queue milestone that everyone is watching.
Programs bring in bridge power when the grid connection will take years longer than the project can wait. The capex is justified by the value of getting power sooner than the grid would deliver it, for as long as the program expects to run the bridge. That expected duration is usually tied to interconnection approval.
If approval is not clearance to go live, the bridge runs longer than planned. For a project without bridge power, a delay in the gap is paid once, in time. For a project running bridge power, the same delay is paid twice: the grid is not live, and the bridge spend that was supposed to have ended keeps running for every extra week. That second cost never appears on a standard project schedule. It is an ongoing capital drain that exists only because the bridge is still on.
- Programs should separate utility dependencies from vendor dependencies. Vendors are managed through contract terms. Utilities are managed through leading indicators and a named owner.
- The utility's commissioning and paralleling checklist should be obtained at the start of the program, not after approval, and treated as scope rather than paperwork.
- Schedule buffers should be re-checked at each utility milestone, such as study results, a signed agreement, or a checklist item closing, instead of being set once at kickoff.
- Every open item should carry a weekly cost: the bridge spend that continues if it slips, or the carrying cost of a finished building waiting for power.
- One owner should be named for the utility interface. The date belongs to the utility, but responsibility for watching it should never be open.
Approval is the milestone before the one that decides the date. Santa Clara shows that the distance between them can run for years, and that no contract closes it. Kyros Infra's position is that the gap should be run as a managed dependency from the first day of the program, with its own checklist, owner, and weekly cost. Programs that do this carry less surprise into the capital conversation, because the cost of delay is already priced when the date moves.
KyrosInfra Intelligence — October 2026 — www.KyrosInfra.com
Sources: Bloomberg, Data Centers in Nvidia's Hometown Stand Empty Awaiting Power (November 2025); Data Center Dynamics, Silicon Valley data centers stand empty awaiting power connections (November 2025); Tom's Hardware, Silicon Valley data centers could sit empty for years due to lack of power (November 2025).
Start Here
We are facilitating a limited
number of partnerships for 2026.
A Strategic Briefing is a 45-minute working session to assess whether there is a real commercial pathway for your technology or infrastructure position in the global AI energy infrastructure market, and map the specific steps to move it forward. We will tell you directly if we cannot help. We only take engagements where we are confident we can advance the transaction.