The warehouse that thinks.

Seen from outside, a data center is a warehouse. High ceilings, few windows, security at the entrance. The resemblance ends at the facade. Inside, value lies not in square meters nor in concrete structure, but in contracted power, in electrical and mechanical redundancy, in cooling capacity, and in connectivity, and it lies in a long-term contract with a tenant of excellent credit who rents not a space but a critical infrastructure that cannot stop. Valuing this asset as one would value a logistics warehouse is a category error, and it is an error the Brazilian market will commit at scale over the coming years as demand for computing capacity transforms the data center into the most contested real estate asset class of the cycle.

The difficulty is not conceptual in the abstract. The three classical approaches to valuation: market, income, and cost, remain the correct vocabulary. The difficulty is that each behaves atypically when applied to an asset whose value is dominated by specialized infrastructure and by a single structuring contract.

Why the market approach barely works

The comparative market approach depends on transactions of similar assets under similar conditions. For data centers in Brazil, those comparables are rare, heterogeneous, and often opaque. Two assets with the same built area can have radically different values depending on available power, redundancy level, supported load density per rack, and lease quality. Built area, the central variable in nearly every property valuation, becomes nearly irrelevant compared to megawatt capacity and the facility's availability classification.

This does not invalidate the approach but requires the valuer to abandon the square meter as the comparative unit and to think in terms of the unit that effectively prices the asset: critical IT capacity. Where transactional evidence exists, it must be normalized by this metric, not by area, or one risks comparing things that are incomparable.

Income dominates, and the cap rate deceives

The value of a stabilized data center derives primarily from the income it generates, and the income approach underpins most valuations. One caveat comes before the rate: because a large share of the asset is infrastructure that wears out and becomes obsolete in short cycles, direct capitalization by a single rate tends to be coarse, and discounted cash flow, which explicitly models the decline in utility and reinvestments over time, generally serves better. With capitalization adopted, the problem concentrates on the rate. A single cap rate applied to total income hides that this income remunerates two superimposed assets with distinct risk profiles: the building itself, with risk similar to a good-quality logistics asset, and the specialized critical infrastructure, whose risk is different, because it is exposed to technological obsolescence, high maintenance costs, and a degree of specificity that drastically reduces the universe of alternative users.

One way to address this is to decompose income and recognize that the portion attributable to specialized infrastructure does not deserve the same rate as the real estate component. Another is to abandon direct capitalization and treat the asset by discounted cash flow, with obsolescence embedded in the projections. Both routes fight the same error: treating all income as if it were as safe as that of a warehouse leased to a good tenant, or, at the opposite extreme, penalizing the entire asset with a risk premium that is justified only for part of it. The tenant's credit quality, generally a global cloud operator, pulls contractual risk down, but does not eliminate the physical asset's obsolescence risk, and the two effects must be separated.

Where replacement cost comes in

The cost approach is usually treated as the poor cousin of real estate valuation, useful for insurance and little more. In data centers, it regains relevance, because a significant portion of value resides in infrastructure assets that are, in essence, specialized industrial installations: substations, generators, uninterruptible power systems, cooling units, building automation. For these components, depreciated replacement cost is a legitimate value reference, provided that depreciation captures not only physical wear but also functional obsolescence, which in this sector is the dominant factor.

And here lies the risk that distinguishes data centers from nearly any other property. An asset designed for the load density of an earlier generation may be functionally outdated for the loads required by artificial intelligence processing, which demand far greater density and cooling. Obsolescence in this sector is not measured in decades but in short technological cycles, and the depreciation applied must reflect that reality, lest one carry on the balance sheet a value the market does not recognize.

Energy as a ceiling on value

There is a factor that in Brazil conditions all others: energy availability. A well-located plot with an approved project and available capital is worth little as a data center if there is no contractable electrical power at the connection point. Contracted energy and the position in the connection queue cease to be operational details and become a determinant of value, in some cases the determinant. A valuation that does not investigate real availability and long-term cost of energy is valuing an intention, not an asset.

The bridge to fund balance sheets

Real estate investment funds are starting to carry data centers, and this brings the discussion to accounting ground. Under the investment property standard, these assets can be measured at fair value, and the fragility of comparable information pushes them to the lower levels of the fair value hierarchy, where the subjectivity of inputs requires redoubled discipline in documentation and auditability. The valuation that supports a fund's NAV must be as defensible as one that supports a transaction, because the quotaholder's confidence depends on it.

Specialized property and the limit of comparables

There is a category of assets that valuation standards treat apart, specialized property, defined by the absence of an active market because the asset is rarely sold in isolation and its value is tied to a specific use. Refineries, hospitals, and certain industrial facilities fall into this category, and the dedicated-purpose data center moves in the same direction. When an asset is specialized, the hierarchy of approaches inverts relative to common property: in the absence of market evidence and, in some cases, of observable market rent, depreciated replacement cost ceases to be a secondary reference and becomes the primary basis of value, precisely because it is what a market participant would have to spend to reproduce the asset's utility.

Recognizing a data center as specialized property has immediate practical consequence. It means the valuer must investigate the most probable and economically viable use of the asset, and ask whether that infrastructure has value beyond the current contract and operator. An asset designed bespoke for a single tenant, with specifications few other users would benefit from, carries specificity risk that must enter the calculation. The same asset, designed with flexibility to accommodate different tenants and different load densities, preserves alternative value and deserves distinct treatment. The question of who else this asset would serve is as relevant as the question of how much it earns today.

The lease structure defines the risk

Data centers are not a single product, and the way space is commercialized changes the risk profile and therefore the valuation approach. In the basic infrastructure model, the owner delivers the structure prepared with energy and cooling and the tenant installs its own equipment, in a long contract approaching a pure real estate lease. In the colocation model, the operator commercializes space, energy, and services to multiple clients, and revenue depends on occupancy, client mix, and renewal capacity, with a profile closer to an operational business than a lease. In the model dedicated to a single hyperscaler, the entire asset serves a very long-term contract with a tenant of exceptional credit, and value depends almost entirely on the quality and duration of that contract.

Each arrangement pulls weight toward a different approach. The long contract with a single strong-credit tenant favors capitalization of contracted income, with attention to weighted-average contract duration, agreed adjustments, and tenant strength. The multi-client model requires projection of occupancy and renewal, bringing the valuation closer to project cash flow. Ignoring this distinction and applying a generic cap rate to any data center, as if all shared the same contractual risk, is the shortcut that delivers a solid-looking number with fragile foundation. The contract is part of the asset, and valuing the asset without reading the contract leaves half of the problem out.

The principle

The data center exposes a truth that holds for all specialized real estate valuation: the correct approach derives from the nature of the asset, not from the valuer's preference. Treating critical infrastructure as concrete, or megawatt capacity as square meters, is comfortable and wrong. The valuation that serves investor and auditor recognizes what the asset is, decomposes what it is worth, and records the reasons.