Capital Stack and Exit for Indian Data Centres: Underwriting, Tax and the Development Spread

The short answer. An Indian data centre development is built at a yield on cost in the mid-teens and sold at a materially lower stabilised yield. The difference between the two, applied to stabilised earnings, is the development spread, and it is the return the developer is actually pursuing. Most of the risk to that spread is concentrated in the period before energisation.

This post sets out how Indian data centre developments are financed, what lenders and buyers test, how the development spread arises and where it is lost, what the 2026 tax provisions changed, and which exit routes are available. It is written for the developer assembling the capital, the credit fund underwriting the construction, and the institutional buyer pricing the asset at stabilisation.

The development spread exists because a completed, let, operating asset is a different instrument from a construction project, and is priced by different capital at a different required return. The developer's business is converting the first into the second. Everything in the preceding eleven posts is either a mechanism for protecting that conversion or a way of losing it.

1. Sources and uses #

Model assumption — 20 MW IT block at ₹62 crore per MW

Two features of the uses column matter for the financing. The electrical and mechanical packages together are the majority of the draw and are spent over a compressed period in the middle of the programme, which concentrates the drawdown profile. And a substantial part of that spend is committed to long-lead plant ordered before the connection agreement is executed, as set out in Post 3, which means capital is at risk against a utility process that has not yet completed.

The financing structure follows from the asset's two lives. During construction and ramp the asset produces no earnings and the debt is sized against the sponsor's completion support and the contracted tenancy. At stabilisation it produces predictable, contracted, escalating cashflow and can support considerably more leverage on considerably better terms. A refinancing at stabilisation is therefore part of the plan rather than a response to conditions, and the construction facility is structured with that refinancing in view.

1.1 The construction facility #

The senior line in the sources column is a construction facility. It makes a commitment available over a defined period, releases it in tranches against certified progress, and converts to a term facility once completion conditions are met.

The availability period is commonly negotiated against the construction sequence, while the binding constraint is the connection sequence, which runs to between twenty-six and thirty-four months on the critical path derived in Post 3. A period set against the building programme alone can expire with the works complete and the facility unenergised.

Conditions precedent operate at three levels: to signing, to the first drawing, and to each drawing after it. The first-drawing list shapes the equity requirement, and its failure mode is specific to this asset class. A list requiring an executed connection agreement sits against a procurement programme that must order transformers and switchgear before that agreement exists, for the reasons in Post 1 §2.1, so the sponsor funds those deposits from equity and peak equity arrives earlier than the sources table implies.

Interest accrues from the first drawing while the asset earns nothing. Where it is funded from the commitment rather than paid in cash it consumes headroom that would otherwise cover construction cost, so a delay raises the interest to be funded and reduces the contingency left to fund an overrun.

1.2 Completion tests and the release of sponsor support #

Completion is a sequence of separate events, and the facility documents distinguish between them because each releases a different obligation.

The interval between commercial operation and financial completion is the period the sponsor funds. Financial completion requires operating history at a coverage level, which cannot exist at energisation, so completion support runs through the early ramp and falls away only once the ratio has been demonstrated rather than projected. That support takes four forms, and the form determines what the sponsor's balance sheet stands behind: a guarantee of the debt to commercial operation, a cost overrun undertaking, an equity contribution undertaking, and a debt service undertaking through the ramp.

The diligence question is which certificate the documents name for each test, and whether any depends on the act of a third party. Energisation does, which is why it belongs in an extension mechanism rather than an event of default.

1.3 The refinancing at stabilisation #

A stabilised data centre supports more debt on longer terms than the same asset under construction, and the refinancing that captures the difference is where the development spread is first realised in cash.

Two tests limit the proceeds, and the binding one differs by asset. A loan-to-value test measures the new debt against a valuation produced by the method in section 4, and so moves with the exit yield. A coverage test measures cashflow against debt service, so it moves with contracted rent and with the period over which the debt amortises. A compression in the exit yield relaxes the first test and leaves the second untouched, which is why a sponsor relying on yield compression to fund a distribution should establish which of the two governs before committing to it.

The refinancing requires a defined set of evidence: an operating record over a stated period, occupancy at or near the level the model calls stabilised, a weighted average lease expiry beyond the proposed tenor, a valuation from a valuer the lender accepts, and a clean compliance record under the construction facility. Each is assembled during construction and operation rather than at the point of application.

1.4 The mezzanine tranche and the intercreditor position #

The sources column carries a mezzanine line because senior debt sized against completion support and contracted tenancy stops short of the total cost, and the sponsor is unwilling to fund the remainder from equity. Mezzanine occupies that interval. It is priced above the senior facility because it is repaid after it, and it is sized against the gap rather than against the asset.

The instrument takes one of three forms, and the form determines where the claim sits rather than how it is priced.

The third form is the one most often mispriced by a sponsor. A holding company loan is structurally subordinated, meaning that it is serviced only from cash the senior covenant package permits to leave the borrower, so the distribution lock-up described in section 2.1 determines whether it is serviced at all. A lock-up that traps cash inside a performing borrower is nonetheless a payment default at the holding company.

The intercreditor agreement converts the ranking into an operating rule, and it is the document that decides what the junior creditor can actually do when the project underperforms.

Two consequences follow for the sponsor's plan. Where the coupon accrues rather than being paid in cash, the balance compounds through construction and ramp, so the refinancing described in section 1.3 has to clear a junior claim larger than the sources table shows. And the release provision means the junior's security is discharged on an enforcement sale whether or not the sale realises enough to reach it, which is why a junior creditor negotiates the valuation and process requirements attached to that provision rather than the ranking itself.

Where the mezzanine is a preferred instrument, it can sit in the equity column of a lender's presentation while ranking ahead of ordinary equity in the distribution waterfall. The diligence question is which column the instrument occupies in the loan-to-value test and which position it occupies in the waterfall, because the two are set by different documents and the answer is frequently inconsistent between them.

A contractual ranking is a contract between creditors. It is given effect in an insolvency only so far as the statutory process permits, and in India that process runs under the Insolvency and Bankruptcy Code, 2016. The intercreditor position therefore governs conduct outside insolvency with certainty and governs the outcome inside it conditionally, which is the reason a junior creditor's protection is built from payment blockage and standstill periods rather than from the ranking clause alone.

1.5 Equity structuring and the shareholders agreement #

Sponsor equity is a single figure in the sources table and several instruments in the documents. The split matters because the instrument determines the ranking on a liquidation, the form in which a return is paid, and how the holding is treated when the asset is sold.

Equity is subscribed in some combination of ordinary shares, an instrument convertible into ordinary shares, and shareholder debt subordinated to the senior facility. Where a subscriber is a foreign investor, the choice of instrument is constrained by the rules applying to foreign investment as well as by commercial preference, and the applicable requirement belongs in the term sheet rather than in the documentation phase.

The shareholders agreement governs the relationship between the subscribers. In a development it is tested during construction rather than at exit, because the events it addresses are cost overruns and funding calls rather than sale mechanics.

The funding clause carries the most weight in a development, because the event it addresses is the event the programme makes likely. A cost overrun requires a further contribution at a point when the asset produces no earnings, and the agreement decides whether a shareholder that will not fund is diluted, lent to, or left in place. A lender takes an equity contribution undertaking from a named party for that reason, and where the undertaking is given by a sponsor whose co-investors are under no obligation to fund alongside it, the undertaking rests on one balance sheet rather than on the shareholding as a whole.

The alignment question is the one to run at documentation. The finance documents impose obligations on the borrower, and the shareholders agreement determines who can cause the borrower to perform them. Where an act required by a covenant sits on the reserved matters list, the covenant is performable only with the consent of whoever holds that reserved matter, and a lender that has not read the reserved matters list holds a covenant of unknown reliability.

1.6 Minority protections, drag and tag rights #

A minority shareholder cannot direct the company, so its position is held through consent rights over defined acts and through rights that operate on an exit. Consent rights protect the value of the holding during the hold. Exit rights determine whether the holding can be realised, and on whose timetable.

The drag-along is the provision a buyer of a stabilised asset reads first, because it determines whether the seller can deliver what is being offered. A buyer paying a stabilised yield for a whole asset wants one counterparty, one warranty schedule and one covenant standing behind them. Where the drag cannot be exercised, the transaction becomes a sale of the majority stake with a minority left in place, which is the partial stake route in section 7 and is priced on a different basis.

The conditions attached to a drag are where the negotiation sits. A threshold holding fixes who may exercise it. A minimum price, or a valuation mechanism that produces one, protects the dragged holder against a sale at a level it would not have accepted. A requirement that every holder receives the same form of consideration prevents a structure in which the majority takes cash and the minority takes paper. A cap on the warranties a dragged holder gives, with liability limited to its share of the proceeds, determines whether the dragged holder can complete at all, because a minority holder cannot support a warranty package sized against the whole of the equity value.

Enforceability runs through two documents. The restriction binds the shareholders as a contract between them. It binds the company, and therefore the register of members, only where it has been carried into the articles of association. Where the articles are silent, a transfer made in breach of the agreement can still be registered, and the injured shareholder is left with a claim in damages rather than with the shares.

An exercise of the drag is itself a change of control, so it engages every consent listed in section 7.3 and every change-of-control provision in the finance documents. The order to run is the consent position first and the drag second, because a drag exercised into a transaction that cannot complete for want of a licensee's consent leaves the dragged holder committed and the buyer released.

2. What lenders test #

Energisation certainty warrants particular attention because it is the test most specific to this asset class in this market. A lender is underwriting a facility that cannot generate revenue until a state transmission utility energises it, on a timeline the sponsor does not control and cannot accelerate with money.

What satisfies the test is documentary rather than narrative: an executed connection agreement, a paid deposit, purchase orders for transformers and switchgear, and a right-of-way position that has been secured rather than assumed. The India Data Centre Review 2026 states the underwriting rule directly, which is to underwrite commercial operation from power evidence rather than from the announcement date.

The Review's own falsification table is usable as a diligence checklist.

Source: IDCR 2026, Chapter 14.

2.1 The covenant package #

The tests above are applied once, at credit approval. The covenants carry them into the documents for the life of the facility.

The treatment of the ramp is the most heavily negotiated part of the package. A facility at first-year occupancy produces earnings below its debt service, as shown in Post 1 §6.1, so the coverage ratio either is not tested until a stated date, or is calculated against contracted rather than occupied capacity, or is supported by a reserve the sponsor funds. Which of the three applies determines who carries ramp risk, and it is a more consequential negotiation than the level at which the ratio is set.

The numerator is where the ratio is most often overstated in this asset class. Where energy is recovered from tenants at landed cost, the recovered amount appears in revenue and in cost in equal measure, and including the receipt without deducting the payment inflates the ratio by the whole energy line, whose accounting treatment is set out in Post 1 §3.2.

Three points establish whether a coverage covenant means what a model assumes: the cash flow definition on which the ratio is calculated and whether recovered energy sits inside it; whether the denominator carries scheduled amortisation or interest alone; and whether the test looks backward, forward, or both.

2.2 Security and its enforceability in India #

The security package follows the standard project finance form. What it delivers on enforcement is governed by what a purchaser could operate, rather than by what the components would realise.

Two features of the asset class limit what the package is worth. The plant has little break-up value against its installed cost, because transformers, switchgear and cooling plant are specified for a particular point of connection and a particular building. The value is the operating facility, which is a bundle of consents held by the borrower.

The connection agreement, the sanctioned load registered against a named consumer, any open access approval and any group captive qualification attach to the borrower rather than to the land. Where the connection agreement does not permit assignment without the licensee's consent, a lender enforcing the mortgage acquires a building without a supply, and so does a purchaser at enforcement. The mechanism that addresses this is a direct agreement with the licensee recording the lender's right to step in and to procure a transfer to a nominee, and whether one exists should be established in diligence rather than assumed.

Enforcement through the share pledge preserves the consents, because the borrower survives and only its ownership changes. It substitutes a different exposure, in that every change-of-control condition set out in section 7.3 is triggered by the enforcement itself.

2.3 The financial model and the cases run on it #

The tests set out above are applied to the outputs of a financial model, and the model is the object on which the credit decision is actually taken. Its structure follows a common form, and so do its defects.

Three cases are run on that structure, and each answers a different question.

The lender case matters procedurally as well as analytically, because it is locked at financial close and used thereafter for every forward-looking test. The projected coverage ratio, the loan life ratio and any cash sweep calculation are run against it, so an assumption conceded during negotiation is conceded for the life of the facility rather than for the credit approval alone.

What separates the sponsor's case from the lender's is a short list, and each item on it is the subject of an earlier post.

A case that moves one of these in isolation understates the exposure, because the variables are not independent of one another. A delayed energisation extends interest during construction, postpones the start of the ramp, and pushes the first renewal beyond the tenor, so a case that moves the energisation date should move the dependent assumptions with it. The correlation runs in one direction only: the events that delay energisation are the events that delay the ramp.

The reproducibility requirement is the part sponsors most often underestimate. A borrower demonstrates covenant compliance by re-running the agreed case with actual figures substituted for forecasts, so a model that cannot be operated by whoever holds it at the fourth test date has become a compliance problem rather than a modelling one. Post 1 §6.2 sets out the reconstruction procedure for the operating model, and the same discipline applies to the financing model built on top of it.

2.4 Sensitivity, break-even and switching values in credit analysis #

Sensitivity analysis in a credit context differs from the ordering exercise in Post 1 §7, which ranks variables by their effect on return. A lender runs the same mechanics against a threshold, and the output is a distance to a covenant rather than a change in the internal rate of return.

The sensitivity. One input is moved, everything else is held, and the resulting ratio is recorded. The output is an ordering of inputs by effect, which is a statement about where diligence money should be spent rather than a statement about the asset.

The break-even. The value of an input at which a stated outcome occurs, being the ratio equal to its covenant level or the facility failing to repay within the tenor. It is expressed in the unit of the input, which makes it precise and makes it incomparable across inputs.

The switching value. The proportional movement in an input required to reach that break-even, expressed against the base assumption. This is the credit-useful form, because a rent assumption and an energisation date cannot be compared in their own units and can be compared as the proportion by which each may move before a covenant is breached.

The leverage the sources table implies can be read against the two denominators a lender uses, and the two behave differently for the reason given in section 1.3.

Model assumption — leverage implied by the sources and uses table

The first pair is the measure a refinancing is sized on where the coverage test binds. The second pair is the measure it is sized on where the loan-to-value test binds. A compression in the exit yield moves the second pair and leaves the first untouched, which is the mechanism of section 1.3 expressed as a ratio rather than as a description.

A downside case combines movements rather than testing them singly, and the combination that matters in this asset class is a late energisation together with a slow ramp, because the two share a cause. Where a model treats them as independent, the probability attached to the combined outcome is understated by the product of two probabilities that are not in fact independent, and the case the committee believes it has stressed is milder than the case the asset can produce.

2.5 The independent technical adviser and the scope of its report #

A credit committee cannot assess an electrical design, and the sponsor's own engineers are not independent of the sponsor. The independent technical adviser closes that gap. It is an engineering firm appointed by the lenders, paid by the borrower, and reporting to the lenders, and its work divides into a report before financial close and a certification function afterwards.

After financial close the adviser's function changes from assessment to certification, and each certificate releases something.

The adviser's two most consequential outputs are its contingency view and its programme view, because the facility's contingency line and its availability period are set from them. An adviser appointed late, after the design has been frozen and the construction contracts signed, can report a scope gap between packages and cannot cause it to be closed, and the gap then falls into the sponsor's cost overrun undertaking rather than into any contractor's scope.

A buyer at exit needs the same comfort, and its own lender will appoint its own adviser. Whether the original report can be relied on by a transferee is a matter of the reliance letter rather than of the report, and a seller that did not secure extended reliance at the outset presents the buyer with a technical diligence exercise that has to be run from the beginning, on the seller's timetable.

2.6 The model audit #

The model audit is a separate appointment, usually to an accounting firm, and its subject is the financial model rather than the asset. Its purpose is narrower than the name suggests: it establishes that the model computes what the finance documents define, rather than that the assumptions fed into it are correct.

The third row is the row that justifies the appointment. A covenant is tested on the definition in the agreement, so a model computing a coverage ratio on any other definition produces a number that is not the covenant, and the divergence surfaces at the first test date. A model audit moves that discovery forward to a point at which the definition can still be negotiated.

Five defects recur in models built for this asset class, and each is visible to an auditor who knows where to look.

The output is an opinion addressed to the lenders and a model that becomes the agreed base case annexed to the credit agreement. From that point the model is a contractual document, and a change to it is an amendment rather than a revision.

2.7 Insurance during construction and operation #

Insurance appears three times in the finance documents: as a condition precedent, as a continuing covenant, and as an assigned asset in the security package. The programme is negotiated once, at the outset, against a risk profile that changes materially at energisation.

Field note. Delay in start-up cover responds to a delay caused by insured physical damage. A delay caused by a licensee not energising the connection, or by an approval that has not issued, involves no physical damage and falls outside the policy. The risk with the highest probability of postponing revenue in this asset class is therefore the risk the construction insurance package does not reach, and a sponsor that has bought delay cover has bought nothing against the delay described in Post 3.

The operational indemnity period is the term most worth negotiating, because it runs against reinstatement rather than against repair. Reinstating a failed EHV transformer is governed by a manufacturing lead time set out in Post 3, and an indemnity period negotiated against the time to rebuild a building is short against the time to replace the plant inside it.

A lender requires a defined set of endorsements, and each addresses a way in which a policy can fail the lender rather than the insured.

Placement follows the local market. A risk located in India is written by an insurer regulated by the Insurance Regulatory and Development Authority of India, with reinsurance placed offshore, so a lender seeking the covenant strength of an international reinsurer looks to the reinsurance arrangements and to the broker's undertaking rather than to a direct offshore policy. That structure also fixes the claims path, which runs through the local insurer whatever the reinsurance says.

Underinsurance is the defect most likely to be discovered at a claim rather than at renewal. The sum insured is set at placement against a cost estimate, and the electrical and mechanical packages are substantially imported, so a currency movement between placement and claim can leave the sum insured below replacement cost even where nothing about the asset has changed. The remedy is a periodic reinstatement valuation rather than an indexation clause, because an index tracks a general price level and not the plant.

At exit, cover is replaced at completion and the diligence item is the claims history rather than the policy. Section H of the data room carries it, and a declined claim or an uninsured loss is a finding that runs to the maintenance record rather than to the insurance programme.

2.8 Interest rate and currency hedging #

Two market exposures sit inside a construction financing, and both convert a price movement into a coverage movement rather than into a cost the sponsor can absorb out of margin.

Interest rate. A floating rate facility drawn across a construction period converts a rate movement into an increase in interest during construction, and that increase consumes the same headroom a cost overrun consumes, for the reason set out in section 1.1. The hedge is either a swap fixing a proportion of the projected balance across the drawdown profile, or a cap bought for a premium.

Currency. Revenue is denominated in rupees, and two exposures run against that.

The first is the capital exposure, and it is specific to this asset class. A substantial part of the electrical and mechanical package is imported and ordered before the connection agreement is executed, then paid in instalments against a delivery schedule, so a movement between the order date and the payment dates changes the cost line directly. A forward series covering the payment schedule converts an unknown cost into a known one. The diligence question is whether the forward series follows the delivery schedule or the original programme, because a delayed delivery leaves forwards maturing against payments that have not fallen due.

The second is the debt exposure. Where the facility is denominated in a foreign currency and the revenue is not, a currency movement restates the liability while leaving the asset where it was, so the coverage ratio and the loan-to-value ratio both move on a price in which the borrower does not participate. Indian regulation of foreign currency borrowing carries its own hedging expectations under the Reserve Bank of India's external commercial borrowing framework, and the applicable requirement belongs in the term sheet rather than in documentation.

A natural hedge exists only to the extent that a lease is denominated in the currency of the exposure, and the denomination of each lease should be established from the lease rather than inferred from the currency in which market rent is quoted. The match is partial in every case, because the operating cost base and the debt service are not denominated in the same proportions as the revenue.

The exposure that cannot be hedged is the one embedded in the exit yield. A foreign buyer converting rupee earnings into its own currency prices a country and currency premium, which is the second component described in section 4.1, and no instrument available to the seller removes it. What moves that component is the composition of the buyer pool, which is the subject of section 7.

3. Tax #

The Union Budget 2026 and the Finance Act 2026 changed the after-tax position materially, and the conditions attached to the changes determine which structures qualify.

The Finance Act conditions access to notified data centre benefits on the facility being established under a scheme notified by the Central Government and on its being owned and operated by an Indian company. Both limbs are conditions rather than preferences, and the ownership limb constrains the structuring available to foreign capital, which supplies the majority of the sector's funding.

The Budget introduced a tax holiday extending to 2047 for eligible foreign cloud service providers operating through Indian data centre infrastructure. The beneficiary of that provision is the tenant rather than the operator, which makes it a demand-side measure: it improves the economics of operating from India for exactly the counterparties that dominate absorption, and therefore supports the rent line rather than the operator's own tax position.

Data centres are also referenced in the Budget as critical digital infrastructure eligible for infrastructure lending rates and accelerated depreciation. The first affects the cost of debt and the second the timing of tax deductions, and both improve returns without changing pre-tax earnings.

Diligence note. Notification is a condition precedent that operates at facility level. A financial model applying notified-data-centre benefits to a facility that has not been notified under a qualifying scheme is applying a benefit the facility does not have. The status should be evidenced rather than assumed, and where notification is pending, the model should carry both cases.

3.1 Accelerated depreciation and the timing of the deduction #

An accelerated depreciation allowance changes when a deduction is taken and not how much is deducted across the life of the asset. The benefit is the time value of the tax deferred, so it is worth more where the discount rate is higher and worth nothing where there is no taxable profit to absorb it.

Indian depreciation is computed on blocks of assets at prescribed rates under the Income-tax Act, so an allowance depends on the block an asset enters rather than on the line it occupies in a capital expenditure table. The allocation between building and plant is made in the fixed asset register, and it is a diligence item in its own right, because it sets the profile of the deduction and is capable of being challenged.

Model assumption — the uses table read by depreciation character

The land line carries no allowance at all. The building line and the plant line attract different rates, and the plant line is the larger of the two by a wide margin, which is why the allocation matters more in this asset class than in most real estate. Soft costs follow the assets they relate to rather than forming a class of their own, so a soft cost allocated to the shell is deducted on a building profile and the same amount allocated to the electrical package is deducted on a plant profile.

The condition that limits the benefit is the availability of profit to absorb it. A facility in construction and early ramp has none, so an accelerated allowance enlarges a loss carried forward rather than reducing a payment, and the value of that loss depends on when profits arise and on the period for which it survives. The ramp derived in Post 1 §5 therefore governs the tax benefit as well as the revenue, and the two move together in the same direction.

Where a minimum tax computed on book profit applies, part of the timing benefit is recovered in exactly the years the allowance is largest, because that is where book depreciation and tax depreciation diverge most. Whether such a computation applies to a given company is a question for the tax adviser, and it belongs in the model as a parallel computation rather than as an assumption folded into the effective rate.

The accounting consequence follows from the same divergence. The tax charge in the accounts and the tax actually paid differ, and the difference is carried as deferred tax. A coverage covenant is computed on cash, so a model that takes the accounting charge produces a ratio moving with an accrual rather than with cash, and it will diverge from the compliance certificate in both directions across the life of the facility.

Two diligence questions establish whether the modelled position is the real one. The first is whether the model computes cash tax from tax written-down values rather than from book depreciation. The second is whether any loss carried forward has been tested against the period for which it survives, because a loss that expires unused was never worth what the model credited to it.

The interaction with the notified benefit is the item most often modelled additively. An allowance deducted against income that is exempt or reduced under a notified benefit produces no cash saving, so the two provisions are modelled in sequence rather than added. Where notification is pending, the case without it carries the allowance at full value and the case with it does not, which is a second reason for the two-case discipline in the diligence note above.

3.2 Infrastructure lending status and the terms it reaches #

Classification of data centres as critical digital infrastructure operates on the terms available from a regulated lender rather than on the credit assessment itself. It changes the shape of the debt a bankable project can raise, and it does not make an unbankable project bankable.

Tenor and amortisation are the two variables the classification reaches that move a coverage test at an unchanged quantum. Lengthening the tenor reduces scheduled amortisation in each period, which raises the ratio without any change to the asset, and so relaxes the coverage constraint on the refinancing described in section 1.3. Where the loan-to-value test binds instead, the same lengthening changes nothing, which is why the question of which test binds should be settled before an infrastructure tenor is treated as an improvement.

The mismatch to test is between the tenor available and the weighted average lease expiry. Debt maturing beyond the lease profile is repaid in part from rent that has to be re-let inside the tenor, and a lender prices that as re-letting risk, usually by requiring amortisation to a balloon rather than by shortening the tenor. The balloon is then refinanced, which returns the exposure to the same market conditions that set the exit yield, and the sponsor has converted an operating risk into a capital markets risk without removing either.

3.3 The ownership condition and the structures it reaches #

The ownership limb constrains the vehicle rather than the source of capital, and the distinction is the one to establish first. A condition drafted at the level of the company that owns and operates the facility is satisfied by an Indian company whatever the residence of its shareholders, unless the condition is drafted to look through the shareholding. Which reading applies is a question for the instrument and for the notification issued under it, and it should be evidenced rather than assumed.

The last row is the row that connects the tax position to the exit. A notification attaches to a facility established under a scheme and to the company named in it, so a group reorganisation undertaken for an unrelated purpose, or a transaction structured as an asset sale, can move the facility out of the notified entity without anyone intending that result. Section 7.3 treats the same problem for the connection agreement, the open access approval and the group captive qualification, and notified status belongs on that list alongside them.

The model consequence is a switch with two positions rather than a rate. The diligence note above already requires both cases to be carried through the development. The addition here is that the switch has to be re-tested at completion, because the transaction itself is capable of moving it, and a benefit assumed in a buyer's model that the transaction extinguishes is a deduction from price discovered after signing.

4. The development spread #

Development cost and stabilised EBITDA are taken from the operating model in Post 1, which is where they are derived and where the assumptions behind them are stated. This post takes them as inputs and adds only the exit.

Model assumption — the same 20 MW block at stabilisation

The spread is a function of two variables and neither is the operating performance of the asset. The first is the gap between the yield at which the asset is built and the yield at which it is sold, which is set by the market's view of stabilised data centre risk relative to development risk. The second is stabilised EBITDA, which the preceding posts address.

The exit yield is the more sensitive of the two and the less controllable. A change of one percentage point in the yield a buyer applies moves the value of this asset by a larger amount than a change of the same proportion in EBITDA, because the yield operates as a divisor. This is why the sector's exit assumptions are the first thing a credit committee should test and the last thing a development plan should take for granted.

4.1 The income approach and the components of the exit yield #

The valuation applied at exit capitalises a sustainable earnings figure at a yield. The earnings figure is contested first, because it determines what the yield is applied to.

The yield is assembled from four components, and a seller should establish which of them a buyer is moving when a yield is quoted. The first is the return available on a long-dated claim carrying little risk in the currency of the cashflow. The second is the premium for the country and the currency in which the earnings are denominated, which is where a foreign buyer's yield diverges from a domestic buyer's, and it is the component that moves as the capital mix rebalances toward domestic sources. The third is a premium for the sector and for the illiquidity of a single asset in a market with few transactions. The fourth is an asset-specific adjustment covering tenant credit, remaining lease term, the technical position of the building against the density the market requires, and the security of the power position, and it is the only component within the seller's control.

4.2 Capitalisation yield and the EV/EBITDA multiple #

A capitalisation yield and an earnings multiple describe one relationship from opposite sides, and converting between them reconciles the exit assumption used above with the multiples quoted for Indian platforms.

Model assumption — conversion between an exit yield and an earnings multiple

IDCR 2026 places implied Indian EV/EBITDA between eighteen and twenty-five times, and that multiple values a different object. It values a platform: an operating business with a development pipeline, a land position, a connection portfolio and a demonstrated ability to repeat the development. Applying its reciprocal to a single stabilised asset would value that asset as though it carried the platform's growth. The gap between the two is the value the market attributes to the pipeline rather than to the building, and it is why a platform sale and an asset sale price differently on the same underlying capacity.

Comparable evidence for the single-asset yield is scarce in India. Most disclosed transactions are platform-level or partial-stake, terms are rarely published, and the assets differ from one another in tenant credit, lease length and power position by more than the yield difference being estimated. An exit yield in a model is therefore either imported from another market and adjusted, or built up from a required return. The diligence question is which of the two has been done, because a required return presented as a market observation is not evidence of what a buyer will pay.

4.3 Sensitivity of value to the exit yield #

Model assumption — the same stabilised earnings capitalised at different yields

Development cost is fixed while value moves with the reciprocal of the yield, so the spread is geared to the yield in a way the value is not. Across the range in the table, value falls by roughly a sixth. The spread over development cost falls by close to half, so a credit committee testing the operating case in detail and accepting the exit yield as given has examined the smaller of the two exposures.

4.4 Break-even exit yield and the margin against stabilised earnings #

The spread disappears at the exit yield equal to the yield on cost, because both quantities are stabilised earnings divided by development cost. That identity supplies a break-even requiring no assumption beyond the two inputs already taken from Post 1, and it is the first of the two break-evens a credit committee should hold.

Model assumption — break-even points on the block in section 4

The two break-evens sit at different distances from the base case, and that difference is the shape of the exposure. The exit yield would have to rise by more than half from the level assumed here before the spread vanished. Stabilised earnings would have to fall by rather more than a third. Both movements are large against the sensitivities derived in the preceding posts, which is the reason a base case shows a spread at all rather than an accident of the assumptions chosen.

The combined case is the one worth documenting, because a movement in the exit yield reduces the earnings headroom without touching earnings. The break-even earnings figure is development cost multiplied by the yield, so it rises as the yield widens.

Model assumption — break-even earnings at two exit yields

At the upper end of the range used in section 4.3 the earnings headroom falls to under a quarter, so a yield expansion arriving alongside a ramp shortfall consumes the spread considerably faster than either movement does alone. The two arrive together more often than independence would suggest, because a market that reprices data centre risk upward is a market in which absorption has slowed.

The same arithmetic produces the switching values described in section 2.4. Each variable can be expressed as the proportion by which it may move before the spread is exhausted, which puts a yield assumption and an earnings assumption on one scale and allows a committee to rank them. Ranked that way, the exit yield governs, which is the conclusion section 4.3 reaches from the other direction.

5. Erosion of the spread #

The spread is not realised automatically, and the mechanisms that erode it are the subjects of the preceding posts.

The last row is the one that most often surprises a seller. A buyer underwriting a twenty-year hold prices the risk that energy cost rises over that period, and a facility with contracted supply presents a lower risk than one exposed to tariff revision. The discount applied for that exposure is capitalised at the exit yield, so a recurring cost difference becomes a capital difference roughly ten times its annual size.

The same capitalisation works in the other direction, which is the argument for treating procurement as a capital activity rather than an operating one.

Model assumption — capitalisation of a recurring saving at a 10% exit yield

Against a development spread of ₹678 crore, a procurement decision taken during development is worth a substantial fraction of the entire spread. This is why Post 7 belongs in the capital plan rather than in the operating budget, and why procurement structuring undertaken after stabilisation captures a fraction of what the same structuring captures before a sale.

5.1 Conditions under which a recurring item capitalises #

The multiplier applied to a recurring saving is the reciprocal of the exit yield, so at the yield used here an annual amount is worth ten times its size when the asset is sold. The multiplier applies only where three conditions hold.

Durability. The saving persists over the period the buyer underwrites. A contract expiring inside the buyer's hold assumption capitalises over its remaining term only.

Transferability. The saving survives the transaction. A group captive structure depends on the ownership and consumption conditions set out in Post 7, and a change of control at the consuming or the generating entity can break them.

Evidence. The saving is visible in audited accounts, in metered records, or in a contract the buyer can read. A saving traceable to none of the three is treated as a forecast, and a forecast is not capitalised.

The fourth row reverses the sign of the investment that produced it. Where energy is recovered at landed cost, an improvement funded by the operator reduces the tenant's bill and leaves the operator's earnings unchanged, so the capitalised value accrues to the tenant while the capital was spent by the seller.

5.2 Irreversibility of each loss mechanism #

Every mechanism in the table above has a point before which it is a decision and after which it is a price. The interval between the two is short for some and runs the length of the programme for others, and knowing which is which determines where a developer's attention is worth spending.

The cost of a given defect rises through the programme in a defined way. A leak prevented at design costs a design change, priced in competition before any contract is let. The same leak prevented during construction costs a variation, priced by a contractor already on site and facing no competitive tension. The same leak identified at diligence costs a capitalised deduction, which is the annual amount multiplied by the reciprocal of the exit yield and is therefore the largest of the three by the multiple set out in section 5.1.

Two of the six cannot be reversed at all once the counterparty holds the benefit. A lease that returns the energy saving to the tenant, and a lease carrying an efficiency commitment the operator cannot evidence, are both positions the tenant would have to be persuaded to give up, and a tenant approached during a sale process understands why the approach is being made.

The sequencing consequence for a seller runs against instinct. The findings a seller can still fix in the months before a process are the documentary ones, being an incomplete metering record, a missing calibration certificate, or an unregistered charge. The findings that move the price are the contractual ones, and those were fixed or lost years earlier, at the design freeze and at the anchor lease. A seller who begins preparing at the point of appointing an adviser has access to the first set and none to the second.

6. What buyers discount #

An institutional buyer conducts diligence against a defined list, and each unresolved item is priced rather than negotiated away.

Power evidence. A signed connection agreement, sanctioned load, the energisation certificate, and the metering arrangement. A facility whose sanctioned load exceeds its measured peak, as examined in Post 2, carries a recurring demand charge the buyer will identify and price.

Contracted cashflow. Lease tenor, escalation, tenant credit, break rights, and the treatment of the energy line. A portfolio with a short weighted average lease expiry is discounted regardless of its current occupancy.

Efficiency and its contractual treatment. Measured annualised PUE under a stated measurement category, any PUE cap and its calibration, and whether the metering exists to demonstrate compliance. A cap the seller cannot evidence is a liability the buyer assumes.

Water and environmental position. Assured supply under drought conditions, any allocation constraint, and the consent position for the generator plant. The regulatory exposure on generation described in Post 6 is a known unpriced item, and a buyer aware of it will price it.

Land and approvals. Title, contiguity for future phases, and the status of every approval on which occupancy depends.

Field note. The documentary output of the connection process described in Post 3 is a financing and sale asset in its own right, not merely an engineering record. Developers who assemble that evidence continuously through construction present a diligence pack that supports the exit yield. Developers who reconstruct it at sale present gaps, and gaps are priced.

6.1 The data room #

A seller assembles the same evidence the lender required, in the order a buyer's advisers will read it. Each absent item is a finding rather than a gap.

Section B carries the most weight because a buyer cannot reconstruct it independently. Every document in it originates with a utility, a licensee or an inspectorate and is dated. Section E is the one most often assembled retrospectively, and a PUE record produced from a spreadsheet rather than from metered data under a stated ISO/IEC 30134-2 category evidences the calculation and not the performance.

6.2 Vendor due diligence and its limits #

Vendor due diligence is a set of reports commissioned by the seller before a process opens, covering the financial, tax, legal, technical and environmental position, and written so that a buyer and its lenders can be given reliance on them.

The value sits in timing rather than in content. A finding surfaced by the seller's own advisers arrives while the seller still controls the remedy and while no price is on the table. The same finding surfaced by a bidder's advisers arrives after a non-binding offer has been made, at a point where the only available response is an adjustment to that offer.

The reliance letter determines what the reports are worth. A report addressed to the seller alone is background reading for a buyer and is not underwritable by the buyer's lender, which will appoint its own technical adviser regardless, for the reason given in section 2.5. Reliance is normally extended on payment and against a liability cap, and the cap is itself a diligence item, because a report whose author's liability is capped well below the exposure it addresses supports the timetable rather than the price.

Three items in this asset class are re-run by every serious bidder whatever the vendor pack contains. The power position in Section B is re-run because the documents originate with third parties and a buyer will want them from the source. The metered performance in Section E is re-run because a measurement is verified at the meter rather than in a report, which is the subject of the site visit in section 7.5. The title chain in Section A is re-run because a buyer's counsel will not take another firm's search.

The scope is read before the findings. A workstream omitted, or a period excluded, is itself a finding, and a buyer's advisers begin by comparing the scope of each report against the checklist in section 6.1. The omission most common in this asset class is a technical scope that covers the building and stops at the point of supply, leaving the connection, the metering and the utility relationship outside the report, which is the section of the pack a buyer cannot reconstruct on its own.

7. Exit routes #

Several Indian operators are pursuing listings. A listing prices a platform carrying both colocation and accelerated compute revenue at a blend of the two multiples rather than at the higher, which is the mechanism behind the separate-vehicle argument in Post 11.

The capital position supporting these exits has shifted. Foreign institutional investors supplied the substantial majority of capital deployed through 2024, and the domestic conglomerate wave since then is rebalancing that mix. A more domestic buyer base changes both the currency in which returns are assessed and the exit yield applied, and the direction of that change is not yet established.

7.1 Locked box and completion accounts #

Both mechanisms produce a price. They differ in the date from which the buyer owns the economics, and in who carries the risk between signing and completion.

Three features of a data centre make the choice consequential. Recovered energy is consumed continuously and billed in arrears, so unbilled energy and the corresponding payable sit in working capital in amounts that are large against the rent line and that move with the tariff. Fit-out accruals close to completion are material, and their cut-off determines who funds them. Tenant deposits and prepaid rent are cash inside the borrower that is not the seller's to extract.

Under a locked box each is handled in the definition of leakage and permitted leakage rather than by adjustment. Under completion accounts each requires an accounting policy stated in a schedule and ranked above the seller's historic practice, because practice on the energy line varies between operators for the reasons in Post 1 §3.2.

7.2 Warranties, indemnities and the limitation architecture #

Warranties and indemnities allocate different classes of risk and are qualified differently. A warranty is a statement of fact, is qualified by what the seller discloses, and produces damages measured by the reduction in the value of what the buyer acquired. An indemnity is a promise to pay on a defined event, is not qualified by disclosure, and produces a payment measured by the loss actually suffered.

The limitation architecture is negotiated as a set: a de minimis below which an individual claim cannot be brought, an aggregate threshold below which claims cannot be recovered, a cap on total recovery, and time limits that differ between title, tax and general warranties. Disclosure operates against the warranties, so a matter fully disclosed is a matter the buyer has accepted, which is why the disclosure letter is negotiated with the attention given to the warranty schedule.

The warranties specific to this asset are the ones a general corporate schedule omits: that the sanctioned load and the connection agreement are as stated and in good standing; that energisation and the statutory clearances have been obtained; that measured annualised PUE for a stated period under a stated ISO/IEC 30134-2 category is as represented; that any PUE or availability commitment in a lease can be demonstrated from installed metering; that the consents for the generator plant are in force; and that the water allocation is as stated and unconditional. Each corresponds to a diligence item in section 6, and an item the seller cannot warrant will reappear as a deduction from price.

Escrow, retention and warranty and indemnity insurance secure recovery, and the choice follows the seller's position after completion. A fund vehicle distributing proceeds to its investors cannot stand behind a warranty for several years, so a buyer requires a retained fund or an insurer in place of the seller's covenant.

7.3 Change-of-control consents and conditions to completion #

A transfer moves a bundle of consents, and each consent depending on a third party's discretion belongs in the conditions to completion rather than in the warranties, because a warranty compensates after the event while a condition secures the position before it.

The group captive row carries the largest capitalised consequence, because the qualifying conditions attach to ownership as well as to consumption and are set out in Post 7. A share sale preserves the borrower and its consents and triggers every change-of-control provision. An asset sale leaves those provisions untouched and requires each consent to be obtained afresh. Neither route avoids the consent question.

7.4 The sale process from teaser to completion #

A sale runs through a defined sequence, and each stage exchanges a quantity of information for a quantity of commitment. The seller's leverage is highest early, when the information released is smallest, and falls as disclosure accumulates.

Two features distinguish this sequence for a data centre. The conditions to completion are dominated by third-party consents, so the interval between signing and completion is set by a licensee and a state rather than by the parties, and the long-stop date has to be negotiated against the consent with the longest recorded history rather than against an average of them. The second feature is that a teaser is a capacity statement, which makes the definitions in Post 1 §1.1 binding on it. A teaser quoting built capacity against a data room evidencing energised capacity produces a reconciliation at the first bidder question, and a reconciliation at that stage costs more than the capacity it appears to recover.

A non-binding offer is priced on what the seller disclosed and not on what is true, so a high indication won by omission is a price reduction deferred rather than avoided. The reduction lands during confirmatory diligence, at a point when the bidder population has narrowed and the competitive tension that produced the indication has fallen away.

Exclusivity is the point at which leverage transfers. Everything the seller intends to resist should be resisted before it is granted, because after that point the seller's alternative is to restart a process that every remaining bidder will know has failed once.

Completion in this asset class carries a mechanical step a corporate completion does not. The metering, the connection documentation and the utility relationship transfer as records and as a relationship rather than as assets, and a completion that moves the shares without moving the utility contact, the meter calibration file and the demand history leaves the buyer unable to evidence its own position at the first tariff revision it faces.

7.5 The management presentation and the site visit #

The management presentation tests the operator against the data room. The site visit tests the data room against the building, and it is the more consequential of the two, because a physical finding cannot be answered with a document.

The presentation covers the operating history and the incident log, the capacity position across the states defined in Post 1 §1.1, the customer relationships and the renewal outlook, the maintenance regime and its record, and the plan for the phases not yet built. A bidder is listening for two things: whether the operator's account of an incident matches the log filed in the data room, and whether the renewal outlook is supported by anything a tenant has actually said.

The site visit is attended by the buyer's technical adviser, and its scope follows the physical chain from the point of supply inward.

Section E is the section the visit exists to verify, because a measurement category can be confirmed only by identifying the meters and the boundary on which they sit. The adviser asks to be shown the instrument producing the reported figure and the point in the distribution system it occupies, and the measurement category defined in Post 9 §4 cannot be evidenced without both.

Two items on the list convert directly into the buyer's own capital plan rather than into a price adjustment. Battery age sets the date of a replacement the buyer will fund, and the spares position sets what the buyer must hold against a lead time it does not control. Both are read at the visit and neither appears in a data room index.

Sequencing matters more than the content of the visit. A site visit held after binding offers converts every physical finding into a renegotiation of the sale agreement rather than into a price input, and a seller who resists an early visit has told the bidder where to look.

8. What would change the calculation #

Three developments would alter the analysis at the level of the spread rather than at the level of individual line items.

The first is any compression or expansion in the exit yield applied to Indian data centre assets. The spread is a function of the gap between construction and exit yields, and the exit yield is set by capital market conditions rather than by anything a developer does.

The second is whether energisation risk is reduced systemically, through the Data Centre Economic Zones under consultation or through the wider adoption of the distribution licence route described in Post 3. A reduction in that risk raises what a buyer will pay for a de-risked asset and reduces the development spread available, because the spread is partly compensation for carrying the risk.

The third is the sector's capital requirement against the capital available. The Review places the cumulative capex required for its base case in the tens of billions of dollars, with roughly half already committed, and Deloitte's projection for cumulative Indian data centre investment by 2030 is larger again. A shortfall against that requirement supports the spread. A surplus compresses it.

8.1 Decision triggers and the evidence that would confirm each #

Each of the three developments is observable before it appears in a transaction, and each has a source that would register the change first.

The horizon over which these resolve is uneven. A single completed transaction with disclosed terms would change the exit yield assumption immediately, because the market currently has none, and one credible observation will carry more weight than its sample size deserves for exactly that reason. The policy and licence triggers resolve over a longer period and are visible in stages, so they can be monitored rather than awaited. The capital mix moves continuously and is observable only in arrears.

The discipline this imposes on a model is the discipline of the Review's own falsification table in section 2. An assumption is recorded together with the observation that would falsify it, and it is re-tested on a stated cycle rather than at the point where a transaction forces the question and the answer arrives too late to act on.

FAQ #

What is the development spread on an Indian data centre? The difference between development cost and the value of the completed asset at a stabilised exit yield. On the model in section 4, an asset built at a mid-teens yield on cost and sold at a stabilised yield of ten percent produces a spread of ₹678 crore on the block modelled there.

What do lenders test before financing an Indian data centre? Contracted capacity at close, weighted average lease expiry, tenant credit, energisation certainty, power cost certainty, land title, and sponsor completion support. Energisation certainty is the test most specific to this asset class in this market.

What changed in Indian data centre tax in 2026? The Finance Act conditions notified-data-centre benefits on the facility being established under a notified scheme and owned and operated by an Indian company. The Budget introduced a tax holiday to 2047 for eligible foreign cloud service providers and referenced data centres as critical digital infrastructure.

Why is procurement a capital decision rather than an operating one? Because a recurring saving is capitalised at the exit yield when the asset is sold. A saving secured during development is worth roughly ten times its annual value at exit, and the same structuring undertaken after stabilisation captures a fraction of that.

What do institutional buyers discount at diligence? Unevidenced power position, short weighted average lease expiry, PUE commitments that cannot be demonstrated from metering, unresolved water and environmental exposure, and incomplete land and approval records. Each is priced rather than negotiated away.

What does a construction lender require before the first drawdown? A defined list of conditions precedent covering corporate authority, land title, the construction contract, the connection agreement and approvals, insurance, the security package and its registration, legal opinions, and evidence that the agreed proportion of equity has been contributed. The list governs how much of the programme the sponsor funds before the lender is committed to fund anything.

Sources #

• Finance Act 2026 and Union Budget 2026 provisions on notified data centres

• Houlihan Lokey, Data Center India Edition, December 2025, via IDCR 2026

• Global Data Center Hub, Q1 2026 capital tracker, via IDCR 2026

• Deloitte, India data centre investment projection to 2030, via IDCR 2026

• India Data Centre Review 2026 (v2.3, edition cutoff 28 July 2026), Chapters 4, 12 and 14 — India Energy Atlas

Sources and uses, the development spread, the loss mechanisms and the capitalisation table are modelled by India Energy Atlas and are labelled as model assumptions. IDCR 2026 figures are quoted at the locked edition snapshot of 13 July 2026; live Atlas products may carry newer records.

This concludes The Indian Data Centre Playbook. The series index and the underlying models are at energymap.in. For the measured GenAI load profile analysis underpinning Posts 2, 3, 4, 6, 7 and 10, see the working note.

India Energy Atlas builds India's grid intelligence layer — substation headroom, interconnection queues, market prices and carbon intensity in one place. See energymap.in/pricing.