Summary

  • Equinix publishes a per-cabinet revenue figure whose divisor is an average number of billed cabinets that the filing never prints, so nobody outside the company can reproduce it.
  • The same word covers a billing status, a cage-space unit called a Cabinet Equivalent, and a billed-space-against-capacity ratio, and Equinix reports 77% utilisation in its annual filing against 76% in its own earnings deck for the same year end, with no reconciliation located.

One word, three jobs

Start with the definition the annual report actually prints. In Item 2 Properties, footnote 4: "MRR per cabinet represents average monthly recurring revenue recognized divided by the average number of cabinets billed during the fourth quarter of the year" (2025 Form 10-K, printed p. 43).

That sentence fixes a numerator — monthly recurring revenue, recognised, averaged across a quarter — and a denominator: cabinets being billed, averaged across the same quarter. Neither is installed capacity. Neither is floor area. A cabinet that exists and stands empty does not enter. A cabinet billed for part of the quarter enters an average whose underlying counts the filing does not show.

Footnote 3 on the same page defines a second quantity, utilisation, as "cabinet space billed versus total cabinet capacity." Different denominator entirely: space rather than revenue, capacity rather than a billed count. That footnote also concedes that power limitations can make physically available space unusable, which means the denominator is itself a judgement rather than a measurement.

The third use sits in the product documentation. In private-cage colocation, Equinix defines a Cabinet Equivalent — CabE — as "equal to the total cage space needed to support one standard Equinix cabinet (and its associated hot and cold aisle space)" (Equinix private cage documentation). That is real estate including aisle, not revenue and not a billing event, and it does not map one-to-one onto cabinet count: the same page puts eight extra-wide cabinets in a 10-CabE cage, and rates a cage by dividing its power draw cap in kVA by its CabE size — the page's own example divides 45 kVA by 10 CabE to give 4.5 kVA.

So a billed cabinet is a revenue event. The cabinet in the utilisation ratio is a space measure against capacity. A Cabinet Equivalent is a cage-space-and-power unit in which eight cabinets can occupy ten units of something called a cabinet. I found no reconciliation between the three anywhere in the four documents I read for this piece.

What the filing prints, and the cell it leaves blank

The annual report gives the per-cabinet figure by region for the fourth quarter of 2025: Americas $2,694, EMEA $2,418, Asia-Pacific $2,355. The total row's cell for that metric is blank. No worldwide value appears in the table or its footnotes.

Read the gap between the highest and lowest of those three. Americas minus Asia-Pacific is $339, which is 14.4% of the Asia-Pacific figure — my arithmetic, not a rate of change the company published. What produces it is mix, and the filing does not decompose the mix. It does tell us the Americas figure excludes Infomart non-IBX tenant income, while the revenue-recognition note lists tenant and subtenant rent among the recurring components.

The numerator is a pool, not a price. Item 8 Note 1 records that recurring revenue exceeds 90% of revenue and comprises colocation — cabinet space and power — interconnection, managed infrastructure, and tenant and subtenant rent, billed monthly and recognised ratably over the contract term (2025 Form 10-K, p. F-16). Interconnection is the part of that pool with the fattest margin and the least to do with floorspace. A customer buying one cabinet and two cross-connects and a customer buying one cabinet and forty both land in the same divisor.

The deck publishes what the filing does not

The quarterly presentation prints a worldwide figure the annual report leaves blank: MRR per cabinet of $2,574 for Q3 2026, $2,538 for Q2 2026 and $2,436 for Q3 2025 (Q3 2026 earnings presentation, slide 23). The first-to-last difference is $138, or 5.7% of the Q3 2025 figure; also my arithmetic.

Then look at the method on slide 47: "(current quarter recurring revenues / 3) divided by ((quarter end cabinets billing prior quarter + quarter end cabinets billing current quarter) / 2)." That is not the filing's calculation. The filing averages one quarter's revenue against one quarter's billed count. The deck divides a quarter's revenue by three, which annualises nothing and instead converts quarterly revenue into a monthly rate, then averages two quarter-end counts. Both methods can be defended on their own terms.

They are not interchangeable, and the deck's definition does not print the two quarter-end dates it averages.

The perimeter is curated and the deck says so: xScale joint-venture recurring revenue excluded globally; Infomart non-IBX rent and Equinix Metal excluded in the Americas; the TIM acquisition in Asia-Pacific; the BT Group acquisition in EMEA. Curating a perimeter for period comparability is a legitimate choice. It also produces a denominator that no customer can locate their own contract inside.

Two ratios for one year end

The utilisation figures should agree and do not. The annual filing, Item 7 Capacity Trends: "Our cabinet utilization rates were approximately 77% and 78%, as of December 31, 2025 and 2024, respectively" (2025 Form 10-K, p. 50). The earnings deck prints 76% for FY2025/Q4. I found no rounding note, scope note or definitional difference to explain the point. Both are reported here as each document printed them, and I attribute the divergence to neither document.

The quarterly filing updates the same ratio on what it describes as the same basis: approximately 78% at 30 September 2026 and at 30 September 2025, again with the warning that power limitations can prevent use of physically available cabinet space (Q3 2026 Form 10-Q, p. 43). It also restates the revenue model in the same form of words as the annual report — "Our business is primarily based on a recurring revenue model comprised of colocation, interconnection and managed infrastructure offerings" — and repeats that contracts generally run one to five years and "thereafter automatically renews in one-year increments."

What a tenant can check

Here is the whole of what a customer or an investor can verify from these documents. A tenant knows their own invoice: their cabinet count, their power rating, their cross-connects, their term. An investor knows three regional averages for one quarter and one worldwide average for three quarters, computed on two different denominators. Between those facts sits the number everybody quotes — what a cabinet costs — and it cannot be rebuilt from either end. You cannot divide, because the billed-cabinet count is not published. You cannot multiply back, because the count that exists is a period average over dates the documents leave implicit.

Nothing here is a finding that a published figure is misstated, manipulated or unlawful. Pooled recurring revenue, a period-average divisor, a curated perimeter and a regional cut with a blank total cell are ordinary issuer practice. The observation is narrower and more useful: a per-unit figure with an unreproducible denominator is a presentation instrument, not an accountability instrument. The benchmark a buyer actually needs — one cabinet, one month, one market, one power rating — sits one level below the granularity at which Equinix discloses.

Where this piece stops

Nine boundaries hold, dated to 13 November 2026. Neither 'cabinet equivalent' nor 'CabE' appears in the searchable body text I read of the 2025 Form 10-K or the Q3 2026 Form 10-Q — a finding scoped to those main documents on that date, not to image-only content, separate exhibits or documents I did not open. No absolute cabinet count, installed or billed or total, was located in the reviewed filings' searched text, so this article states none and does not reconstruct the denominator. The filing does not say why its total-row cell is blank and I supply no remedy for the silence.

The 77% and 76% utilisation figures are reported exactly as each document printed them; the gap is attributed to no rounding, scope, definition or error. The deck's worldwide average and its exclusion list are not reconciled to the filing's regional figures anywhere I read, so the two sit side by side and are never averaged or substituted for one another. The reviewed documents print no churn rate, no renewal rate, no re-pricing statistic and no per-tenant invoice or tariff, and nothing here is derived from a price list I did not read.

No published number is alleged to be misstated, manipulated or unlawful — the subject is a disclosure perimeter, not a fraud. The two-quarter-end averaging is the deck's own method, and the specific quarter-end dates are absent from that definition, so I do not assert them. The one-to-five-year term with automatic annual renewal is the company's general description of its contracts, not the term of any tenancy I identified.

BTW has written about Equinix's interconnection fabric and its expansion record elsewhere. This piece works only the units, because the units are where a public number stops being checkable and starts being merely quotable.