Revised practical article; source details and figures are dated to their cited filing or event, with current evidence reviewed through 24 September 2026.
A data centre building is not one asset with one replacement date. It combines land, buildings, power systems, cooling, core systems, servers, networking, leasehold improvements, software, and contractual access to electricity and fibre. Some equipment may be refreshed every few years. A building may have a much longer useful life. The cash an owner must reinvest therefore depends on the vintage, utilization, technical requirements, and service commitments of the operating assets.
Equinix’s 2025 Form 10-K reports $23.6 billion in property, plant and equipment at 31 December 2025 and $1.9 billion of depreciation expense for the year [1]. It identifies a range of estimated useful lives: core systems, 3–40 years; buildings, 12–60 years; leasehold improvements, 12–40 years; personal property, 3–10 years; and capitalized internal-use software, 3–5 years [1]. Those ranges make a point that a single “depreciation equals maintenance” assumption cannot capture the renewal cycle.
Separate project spending from keeping the current service running
The same filing says Equinix’s property, plant and equipment includes the costs incurred to build out or acquire its data centres. It records construction in progress separately and describes major ongoing investment in new and expanded facilities. Management also reports a non-GAAP measure, adjusted funds from operations (AFFO), which includes “recurring capital expenditures” defined as spending to extend the useful life of data centres or other assets required to support current revenue [1]. Its 2025 recurring capital expenditure adjustment was $284 million.
That disclosure is especially useful because the company tells us what it counts in its own recurring-capex measure. It still does not prove that $284 million is the complete economic maintenance need for every investor’s owner-earnings model. It is one issuer-defined measure within a particular non-GAAP framework. Check the reconciliation, the assets included, whether spending is cash or accrued, and what recurring work may be classified elsewhere.
The economics of replacement investment are also different from growth investment. Livdan and Nezlobin model capital with non-geometric efficiency profiles and show how replacement cost and the productive capital stock can diverge from each other [2]. In plain language, replacing an old asset and adding capacity are different economic jobs, even when the cash-flow statement groups them together. That helps explain why a building’s book depreciation is a weak stand-alone estimate of cash that must be reinvested to keep service quality competitive.
Capacity has to earn its keep
A new data centre can be an attractive long-term investment and still lower near-term distributable cash. The valuation asks whether contracted demand, utilization, power availability, customer credit, construction cost and funding support the expected return. A site with available megawatts but delayed grid connection may not generate the expected revenue. A site that reaches capacity may need power, cooling and network upgrades before it can serve more customers. A newly built site adds potential capacity; it is not automatically “growth” if replacement of old equipment is necessary to deliver existing contracts.
For each project, an analyst can track:
- cash already spent and remaining committed spend;
- whether land, construction, equipment and power are owned, leased or contracted;
- expected usable capacity and the date it can serve customers;
- customer commitments, contract duration, and the conditions under which customers can delay or cancel;
- expected utilization, operating costs, debt and lease service;
- recurring refurbishment and equipment refresh once the project opens.
Equinix’s filing discloses substantial construction activity and unaccrued capital-expenditure commitments for land, infrastructure equipment and labor associated with data-centre developments [1]. These commitments matter for funding and delivery risk. They are not all immediately due and not all maintenance spend. The model needs dates and categories before deciding how they flow into owner earnings or debt capacity.
Non-GAAP cash measures still need a bridge
REITs and infrastructure operators often report funds from operations, adjusted funds from operations, EBITDA or recurring-capex measures. These can provide useful information about property cash generation and distribution capacity. They are not automatically equivalent to Buffett’s owner-earnings concept. Each measure has an issuer definition, and different issuers can exclude different items. Read the reconciliation, then ask whether the excluded cost is actually avoidable or whether it has simply moved to another period or category.
A sensible valuation may combine several views: property-level cash flows, project-level returns, debt and lease obligations, refinancing exposure, an equity cash-flow model, and a distribution test. It should also account for equity issuance. Equinix discloses that it may issue shares through an at-the-market program, and that further issuance can dilute shareholders [1]. A growth-funded distribution is still funded by capital; the per-share ownership claim can shrink while total cash grows.
The owner-earnings conclusion should preserve uncertainty
For a data-centre business, the major analyst judgment may be the portion of total capital spending required to keep the existing service competitive. The public filing provides depreciation lives, recurrent capex, construction commitments and future plans. Those are strong evidence inputs. A precise maintenance-capital estimate still requires asset-level history and project detail. If a company does not disclose that split, show cases or state that the number is unavailable from public disclosures.
Draw the service boundary before estimating maintenance
A data-centre operator sells a service, not simply a building. Before estimating cash requirements, draw what the customer contract promises and what the operator owns. One site may include land, shell, electrical connection, substation, generators, fuel arrangements, uninterruptible power systems, cooling, fire suppression, security, network pathways and monitoring. The customer may own servers and racks; another contract may include customer equipment, managed services or shared infrastructure. The point at which responsibility changes affects both capital needs and service liability.
Create an asset hierarchy for each major location. At minimum, separate civil structure and building systems from electrical and mechanical equipment, network infrastructure, security controls, software, tenant fit-out and customer-specific assets. Capture in-service date, remaining useful life, capacity, utilization, service history, failure consequence, replacement lead time, vendor, current replacement quote and whether an asset is leased or owned. Useful-life ranges in the 2025 filing are wide because asset classes have very different jobs [1]. An aggregate depreciation charge cannot tell us which part of a site will require cash next year.
For each asset, ask what failure would mean contractually and commercially. Could the company move load to another site? Is there redundant capacity? Would a failure trigger service credits, a customer departure or regulatory breach? How long would replacement take, and could a vendor supply a compatible part? A low-cost component with a long lead time may be more important to the cash reserve than a large asset with redundancy. Rank risks by the expected operational consequence and recovery time, not only by book value.
Split project capital by purpose and by date
Every material construction project should have a project sheet that distinguishes:
- Replacement: spending that retires or renews an existing asset needed to deliver the present service.
- Compliance and resilience: spending required by safety, environmental, security, power-quality or customer standards.
- Expansion: new capacity or geography intended to add serviceable demand.
- Customer-specific work: fit-out or equipment tied to a contract, with a clear recovery mechanism.
- Efficiency upgrade: spending expected to lower energy, labor or outage costs, with a measured payback.
- Unclassified amount: spending the public disclosure does not permit the analyst to allocate.
The classifications can overlap. A new cooling plant may replace an obsolete system and allow more capacity. Use a like-for-like replacement estimate as the renewal component and show the incremental cost as expansion or upgrade when records allow. If the split cannot be supported, report a range and keep an unresolved item; do not allocate all project spend to growth merely because management calls it a “development” or all spend to maintenance because the site was already open.
Timing matters as much as category. Record cash spent to date, unpaid invoices, land and equipment deposits, construction commitments, expected commissioning date, power availability, customer move-in, and the date contracted revenue begins. Compare cash spend with a monthly funding schedule. A project can appear attractive on a stabilized annual return and still require temporary funding beyond the company's liquidity or borrowing capacity. Equinix reports construction-in-progress and unaccrued commitments alongside its planned development activity [1], which is exactly why a model needs dated obligations rather than one annual capex assumption.
Worked investment screen: an illustrative expansion
Assume an operator proposes a new facility requiring KES 10 billion of total construction and equipment cash over two years. At stabilized occupancy, management expects KES 1.8 billion of annual revenue and KES 800 million of cash operating costs before maintenance, financing, tax and central overhead. Assume annual renewal spending for the operating facility is KES 150 million once stabilized. These figures are purely illustrative; they are not Equinix data or an industry benchmark.
At stabilization, the project's initial annual cash contribution before financing and tax would be KES 850 million: revenue less operating cost and renewal. Dividing by KES 10 billion gives an 8.5% simple stabilized cash yield. That shortcut is not yet a project return. It ignores the construction period, the time to reach occupancy, cost overruns, working capital, lease expense, corporate costs, taxes, financing and the risk that the expected revenue never arrives.
A more useful schedule places the capital outflow in the months or years when it is due, then phases capacity and customer cash receipts. Model a base case in which power is available on schedule and signed customers occupy the site as contracted. In a downside case, delay grid connection by six months, raise equipment cost, slow customer installation and test whether deposits or capacity payments compensate for the delay. Add a stress case in which a major customer cancels or defers a portion of capacity. The owner should see the low point in cash and the financing required before the stabilized yield is reached.
The project gate should require evidence at each stage: land and permits secured; utility capacity confirmed; construction price and contingencies agreed; funding available; customer contracts signed and credit reviewed; equipment ordered with realistic lead times; and commissioning passed. If a stage fails, define whether management stops, resizes or defers the project. This avoids investing the remaining cash simply because a large amount has already been spent.
Read customer contracts as cash-flow documents
Data-centre contracts can vary in term, power reservation, space, cross-connects, escalators, service levels, deposits, renewal options, termination rights and customer fit-out. Do not model contracted revenue as guaranteed cash. Identify the minimum commitment, customer cancellation or downgrade rights, payment dates, credit support, power pass-through and service credits. Separate signed backlog from pipeline and from management's estimate of demand.
For each large customer, compare reserved capacity with installed load, actual power draw, invoiced recurring revenue and collection. A customer may reserve capacity before full deployment; the operator may need to build the site well in advance. That creates a period when committed spending precedes cash revenue. Conversely, a contract with minimum payments can reduce demand risk only if the customer is creditworthy and the contract is enforceable in the relevant circumstances.
Electricity and cooling costs are central to delivered service. The model should distinguish energy passed through to customers from costs the operator bears, and show how tariffs, fuel, efficiency and utilization alter gross contribution. A low utilization rate can raise cost per occupied unit even when the building is operating normally. Where redundant power or cooling is maintained, some capacity may be intentionally idle as resilience and should not be judged as wasted solely because it is not billable every hour.
Reconcile AFFO and other non-GAAP measures
Equinix's AFFO reconciliation and its $284 million recurring-capital adjustment provide a company-defined lens on recurring spending [1]. Compare that measure with cash capital expenditures, depreciation, construction activity and the operating obligations described in the filing. Ask whether recurring capex includes replacement servers, network upgrades, customer-specific equipment, tenant improvements or only the issuer's stated categories. Check whether the measure includes capital that is paid in cash this period or amounts accrued for later payment.
A non-GAAP adjustment can be informative without being a complete owner-earnings measure. If the company labels an expenditure “non-recurring” but a similar project occurs each year across different sites, it may be recurring at the portfolio level. If a project is a genuine expansion, omitting it from a cash flow measure may make sense for one question about existing sites; it still matters when estimating the capital required to maintain market position and future earning power. State which question the number answers.
A useful bridge presents reported operating cash, total cash capex, company-defined AFFO, an analyst estimate of renewal cash and the remaining growth/project capital in adjacent columns. It should avoid subtracting all future committed construction spending from one year's earnings while also showing how those obligations will be financed and what capacity they create. A reported distribution can be covered under an issuer-defined metric and still depend on debt, share issuance, asset sales or temporary working-capital timing.
Financing, leases and dilution belong in the same decision
Map secured and unsecured debt, variable and fixed rates, maturities, covenants, guarantees, lease liabilities and refinancing needs. Match each obligation with the site's operating cash and contract term. A long-lived building financed with debt maturing well before customer contracts renew creates refinancing exposure even if today's interest coverage looks comfortable. Include interest expense and principal repayment according to the cash-flow measure being used; do not subtract debt twice if a valuation model already moves from enterprise value to equity value through net debt.
Check whether growth is funded by retained cash, debt, asset sales, joint ventures or equity issuance. A growing total AFFO can coexist with flat or falling AFFO per share if the share count expands. Compare gross new issuance, at-the-market sales, share-based compensation, acquisition shares and repurchases. If repurchases merely offset compensation or new shares fund developments, the investor needs the cost and per-share outcome, not only the announcement amount.
The decision should show distribution headroom under both base and downside occupancy. Include maintenance, interest, maturities, committed construction and share dilution before describing cash as available to owners. If the company reports a dividend, compare it with recurring cash generation after the relevant obligations and explain whether the policy depends on external funding.
A field test for asset-heavy owner earnings
When public data are incomplete, ask management or operations for a five-year asset renewal log, project-level capex, commissioning delays, outage history, energy intensity, utilization, customer concentration, recurring-capex reconciliation and vendor lead times. Cross-check planned work with current condition and actual invoices. If the answer is unavailable, make the estimate provisional.
Livdan and Nezlobin's work on non-geometric depreciation supports the broader point that productive capacity and accounting depreciation need not move in lockstep [2]. For an analyst, the practical output is not a more elaborate depreciation formula by itself. It is an asset-by-asset renewal plan, a dated construction and funding schedule, and an explicit range for cash that can be distributed while preserving the service.
Power, utilization and customer economics
Power is an operating input and a capacity constraint. A facility may have a megawatt allocation on paper but lack completed substations, grid connection, backup equipment, permits or customer load needed to earn a return. Record contracted power, energised power, usable IT load, committed customer load, installed load and billed load as different figures. Where the issuer reports only aggregate capacity, mark facility-level utilization as unavailable rather than dividing two incompatible totals.
Track revenue and cash contribution by customer or contract where disclosed. A large customer may have a lower rate but predictable multi-year utilization; a smaller customer may pay more but require expensive fit-out and support. Compare contract term with the remaining life of the customer-specific equipment. Record the cost and recovery mechanism for power, cross-connects, installation, service credits and security. A long-duration contract can support project financing, but only if the customer can pay and the contract remains enforceable under plausible business conditions.
Energy-efficiency projects can be renewal, compliance or growth, depending on the counterfactual. Replacing an inefficient chiller may reduce costs and improve capacity, while a new utility interconnection may enable an expansion. Estimate both the capital outlay and the expected cash savings using actual energy consumption and tariff assumptions. Run a case with lower utilization: the energy-saving equipment may still consume maintenance and financing cash while revenue falls.
Build a paired, not one-variable, downside
A spreadsheet should avoid changing only occupancy while leaving every other cost and funding assumption fixed. In a real delay, construction costs may rise, commissioning shifts, customer move-in slips, interest accrues longer and vendor invoices arrive before revenue. Pair those effects in a delayed-delivery scenario. In a demand slowdown, lower occupancy may coincide with customer price renegotiation, slower collections and reduced ability to fund the next project from operating cash.
For each case, calculate the lowest cash balance, debt and lease service, distribution coverage, capital still required to complete committed projects, and the amount of equity issuance that may be needed. Then explain which project or distribution management could defer. If a project is contractually committed, do not model it as an optional outflow without documenting the relevant cancellation right.
This is also a useful way to distinguish a strong existing portfolio from risky expansion. Existing facilities may generate durable recurring cash, while development projects carry construction and absorption risk. Value them separately or report separate ranges. A portfolio-level growth multiple can conceal projects that have not yet reached power-on or customer occupancy.
Public data limits should lead to a practical evidence request
An analyst using only a Form 10-K can estimate reported cash flow and see total capital, depreciation lives, project commitments and issuer-defined recurring capex. That evidence may not disclose each site's physical condition, equipment replacement dates, contract margin, repair history or customer-specific investment. A precise maintenance number from the public filing alone may therefore be impossible.
The next diligence request should be specific: renewal capex by asset class and site for five years; project-level committed spend and expected opening date; power and construction milestones; capex paid versus accrued; customer fit-out recovery; utilization and contracted capacity; service credits and outage costs; and share issuance needed to fund the pipeline. Each missing item should have a named effect on the owner-earnings range.
The resulting investment note can still be decisive. It can say that current portfolio cash is durable within a stated range, but the distribution case is sensitive to project spending and dilution; or that a valuation depends on new capacity reaching contracted utilization by a certain date. That is more useful than presenting one payout ratio without construction and funding context.
Trace a project from announcement to cash returned
For each announced development, keep four amounts separate: total expected project cost; cash spent to date; cash committed but not yet paid; and uncommitted future capacity spend. The annual report may combine certain assets or project costs, so document the definition. Then connect the project to capacity, customer contracts, expected revenue start, stabilization period, replacement schedule and financing. A capacity announcement does not establish that all capacity will be energized, occupied or collected.
At commissioning, reconcile project cost against the approved budget. Record overruns, delays, capitalized interest, tenant incentives and costs that management moved between asset classes. Then test the project's actual return after it operates: occupied capacity, price, power pass-through, energy use, maintenance, service credits, collections, local overhead and financing. Compare actual results with the original underwriting at the same months after opening. This reveals whether weak performance reflects slow ramp, higher operating cost or an inaccurate demand thesis.
Use separate return measures for unlevered project economics and equity cash after financing. The first helps compare sites independently of capital structure; the second shows what ordinary shareholders may receive after debt service, tax, lease obligations and required reserves. State which one the spreadsheet reports. A project can be value-creating before debt and still create a cash shortfall for the parent if debt amortizes too quickly or if the parent must finance the next construction phase.
Review the model per share, not only per megawatt
The final review should reconcile site-level capacity and project returns to group cash, debt, leases, minority ownership, corporate costs, tax and diluted shares. A project may be attractive while the group-level equity claim is thin because of financing or issuance. Keep the operating asset case separate from the shareholder case, then show how a change in utilization or renewal spending changes cash per diluted share. That is the number a continuing ordinary owner can compare with the market price.
Workbook map
The Project 2 workbook shows Equinix FY2025 property, depreciation and issuer-defined recurring capex separately from a fictional KES project screen in Asset Heavy. Evidence Vectors records the 31 December 2025 period end and 11 February 2026 SEC filing date separately, and labels the KES project screen as fictional. That screen calculates stabilized contribution and simple cash yield after renewal spending; it does not infer Equinix maintenance capex from the company’s recurring-capex label. Profile Stress applies a distinct utilization/power and renewal-spending scenario to an explicitly normalized index. Use the article’s asset, lease, power and construction checks before turning either example into an issuer estimate.
Technical appendix
Technical appendices
The worked calculations and source boundaries below extend this article's examples. They keep assumptions and unresolved inputs visible so readers can reproduce the steps without mistaking an illustration for an issuer or sector estimate.
References
[1] Equinix, Inc., Form 10-K for the year ended December 31, 2025, filed February 11, 2026, https://www.sec.gov/Archives/edgar/data/1101239/000110123926000032/eqix-20251231.htm.
[2] D. Livdan and A. Nezlobin, “Investment, Capital Stock, and Replacement Cost of Assets When Economic Depreciation Is Non-Geometric,” Journal of Financial Economics, vol. 142, no. 3, pp. 1444–1469, 2021, https://doi.org/10.1016/j.jfineco.2021.05.021.
Companion resources
The workbook and technical appendices are available with research membership. The video production guide maps this article to a planned episode; a published video is not yet available.
No public comments yet. The first thoughtful question can start the discussion.