Iron Mountain Data Centers: 452 MW and the conversion chain

Summary

Iron Mountain's data-centre segment reported $204 million of Q3 2025 revenue, up 33% year over year, on a 452 MW operating portfolio that is 97% leased, with segment adjusted EBITDA of $107 million at a 52.6% margin — capacity, not announcements, is what the segment is currently monetising.
The forward story rests on conversion: 203 MW under construction is 61% pre-leased, 685 MW is held for development, and management guides current backlog to drive at least 25% data-centre growth in 2026 and roughly $250 million of growth beyond 2027, before any new leasing.
The binding constraint is delivery, not demand. Only 3 MW commenced in the quarter against 13 MW of new leases signed and 11 MW signed after quarter end, which tells you where the schedule risk sits: power, construction and customer acceptance.
Two disclosed weaknesses matter more than the headline: a third-party transcript contradicts itself on the 18-month energisation figure, and no disclosure captured here gives capital cost per megawatt, power contract terms or a defined backlog→revenue conversion rate.
The falsifiable signal to watch is commencement volume: if energised megawatts and rent commencements do not track the guided schedule, pre-leasing is being funded ahead of the cash it promises.

A data-centre company can sign a lease that produces no revenue for years. That single sentence explains most of what is interesting about Iron Mountain's data-centre business, and most of what is dangerous about reading its press releases quickly.

Iron Mountain is best known for records management and shredding, businesses built on physical custody and recurring fees. Its data-centre arm inherits that logic — long leases, contractual escalation, high utilisation — but operates a different machine. A data centre is a power-conversion asset wrapped in real estate. Land, shell, substation capacity, cooling and fibre are assembled over years; only then does a customer's equipment arrive, pass acceptance tests, and begin to be billed. Between the signing of a lease and the first invoice sits everything that matters.

The company's third-quarter 2025 reporting gives an unusually clear view of that sequence. The numbers are large enough to be material to the parent and granular enough to test the conversion chain step by step.

What the segment actually earned

The parent reported total third-quarter 2025 revenue of $1,754 million against $1,557 million a year earlier, roughly 13% reported growth and 12% at constant currency, with record adjusted EBITDA of $660 million, net income of $86 million, and AFFO of $393 million, or $1.32 per share, according to the Q3 2025 earnings release as filed with the SEC. The company also raised its quarterly dividend by 10%, a decision that reads as confidence in cash generation rather than in pipeline.

The data-centre segment is the part that matters here. It produced $204 million of revenue in the quarter against $153 million a year earlier — 33% reported growth, 32% at constant currency — with organic storage rental growth of 32%. Segment adjusted EBITDA was $107 million, up $41 million, at a 52.6% margin, an improvement of roughly 900 basis points year over year. Renewal pricing came in at +14% on a cash basis and +19% on a GAAP basis, with churn of 0.3%, as presented in the company's Q3 2025 earnings presentation and recounted in a third-party transcript of the earnings call.

Those are not early-stage figures. A 52.6% segment EBITDA margin and a 0.3% churn rate describe a mature, nearly full asset base. The 97% leased figure for the 452 MW operating portfolio says the same thing: there is very little empty capacity left to sell in the buildings that are already live. Growth from here cannot come mainly from filling existing halls. It has to come from energising new ones.

That reframes the entire investment question. The reported growth rate is a lagging indicator of decisions made years ago. The leading indicator is the schedule on which new megawatts cross from construction into service.

Three buckets, three different economics

Iron Mountain's disclosure sorts its capacity into three categories, and conflating them is the most common error in reading the segment.

Operating — 452 MW, 97% leased. This is the revenue machine. Nearly every megawatt here is under contract, which means the segment's current revenue is largely locked and its growth is dominated by contractual escalation and renewals rather than by new sales. The +14% cash renewal spread is the clearest evidence of pricing power in this bucket: customers are renewing at materially higher rates rather than leaving, which is what a 0.3% churn rate confirms.

Under construction — 203 MW, 61% pre-leased. This is the visible pipeline, and the pre-leasing percentage is the single most informative number in the disclosure. Sixty-one per cent of the capacity being built already has a customer attached. The corollary is less comfortable: the remaining 39% — roughly 79 MW — is being built on speculation against future demand. Speculative construction is rational when demand is deep and power is scarce, and it is also where a capacity cycle can hurt you if demand thins between groundbreaking and delivery.

Held for development — 685 MW. This is land and power rights, largely not yet capitalised into a building. Total developable capacity is put at 1.3 GW, meaning the company expects to roughly triple the operating portfolio over time. But the gap between 'held for development' and 'operating' is precisely the conversion cost this article is about: entitlement, utility interconnection, shell construction, fit-out, and only then customer acceptance. None of those steps shows up as revenue while it is happening.

Read together, the three buckets describe a company that has already converted most of its live capacity into cash and is now making a large, multi-year bet that it can do so again. The bet is not that demand exists. It is that delivery will not slip.

Where the chain actually stalls

The quarter's leasing and commencement data exposes the bottleneck more sharply than any commentary.

In the third quarter of 2025, Iron Mountain signed 13 MW of new leases. After the quarter closed, it signed an additional 11 MW — and that second transaction is worth reading closely, because it involved transferring a customer's previous 25 MW London lease into an expanded 36 MW lease covering the entire Chicago data-centre site, as detailed in the company-hosted results release. Roughly 300 leases totalling 11 MW were renewed. And new capacity commenced in the quarter amounted to 3 MW.

Sit with that last figure. Thirteen megawatts signed, eleven more signed later, eleven renewed, and three commenced. Leasing is running several times faster than delivery. The signed-but-not-yet-billing capacity is accumulating.

That is not necessarily a problem — it is how the business is supposed to work, since leases are signed before fit-out completes. But it means the segment's near-term revenue trajectory is already determined by construction and energisation schedules, not by the sales team, and it means any slippage in those schedules lands directly on revenue with a delay measured in quarters.

The Chicago transaction illustrates the mechanism from the other direction. Moving a customer out of London and into a 36 MW Chicago footprint does not create new demand; it reallocates it, while locking the customer into a larger commitment. For the company that is a win — a bigger lease, longer duration, and a released London position to re-let at prevailing rates. For a reader trying to forecast revenue, it is a reminder that headline megawatt signings mix genuine expansion with consolidation and migration.

Why the backlog needs a conversion rate

Management's forward guidance is unusually specific, and unusually dependent on that conversion rate. The company expects data-centre revenue growth of nearly 30% for 2025. For 2026, it guides that current backlog alone drives at least 25% growth before any new leasing. For 2027 and beyond, current backlog is expected to contribute approximately $250 million of growth, again before new leasing. Roughly 450 MW of capacity is described as energising over the next 24 months.

One passage in the third-party transcript puts roughly 250 MW in the next 18 months, while another puts roughly 150 MW in the same window plus another 200 MW in the following six months. Those two accounts cannot both be right. The discrepancy is unresolved in the transcript, and it matters precisely because the 2026 guidance rests on it. Anyone modelling the segment should treat the 2026 growth figure as anchored to the company's own materials and treat transcript-derived capacity timing as soft until it is reconciled.

More structurally, 'backlog' is not a standardised financial measure. Across the industry it can mean signed leases, signed leases plus renewals, or a broader committed-capacity figure. The disclosure captured for this article does not define it, does not state the historical conversion rate from backlog to commenced revenue, and does not separate contracted-but-not-commenced from contracted-and-commencing. Each of those gaps is a place where optimism can persist for a while without being contradicted by reported numbers — and then be contradicted all at once.

For the 2026 guide, a useful sanity check is arithmetic rather than narrative. If the segment earns roughly $204 million a quarter, then at least 25% growth implies something above $1 billion of annualised revenue next year, with roughly $250 million more of growth expected beyond 2027 from the current backlog alone. Those figures imply a sizeable scheduled ramp. Whether the ramp is achievable is a construction and power question, not a demand question.

Capital sits in the ground before it earns

There is a second chain running alongside the revenue chain, and it is the one that determines whether the growth is worth having.

Building a data centre requires capital years before the first invoice: land, substation and utility works, shell, mechanical and electrical plant, generators and cooling. The 203 MW under construction and the 685 MW held for development represent a very large future capital programme, and the 61% pre-leasing on the construction bucket is the main mitigant. Pre-leasing converts construction risk into delivery risk: the customer is committed, but the company still has to spend the money, and it does not collect rent until the megawatts are live.

Held-for-development capacity is more exposed still. A 685 MW land and power bank is an option on future demand, and options cost money to hold while paying nothing. The relevant question is not whether the option is valuable — with data-centre demand strong, it plausibly is — but whether the carrying cost and the timing of utility interconnection are being managed so that the option can be exercised when demand arrives rather than after it has moved elsewhere.

What the available disclosure does not give a reader is the arithmetic of that trade. There is no capital cost per megawatt, no cost of capital or return-on-invested-capital figure for the segment, no power purchase or utility contract terms, no lease commencement schedule and no customer concentration detail. Those omissions are normal in a quarterly REIT release, and they are also exactly the inputs needed to judge whether the expansion compounds value or merely grows the balance sheet. The claims that can be checked — revenue, margin, leasing, renewals, churn — come from the company's filings, the company's investor page for quarterly results and its press-release archive. The claims that cannot be checked should be labelled as such rather than inferred from the confident ones.

What would falsify the story

This is the part professional readers should hold the company to, because the thesis is clean enough to test.

Commencements must accelerate. Three megawatts in a quarter cannot produce a 25% revenue step-up next year on its own. Either commencement volumes rise sharply in the coming quarters, or the 2026 guidance is being carried by pricing and renewals rather than by new capacity.

Pre-leasing on the construction bucket should hold or improve. If 61% drifts down while construction spending rises, speculative capacity is being added into a softening market — the classic prelude to a supply overhang.

Renewal spreads should stay positive. A +14% cash spread and 0.3% churn are strong. If spreads compress toward zero while churn stays low, pricing power is fading even though customers are staying.

The energisation schedule needs to reconcile. The 18-month capacity figure is currently contradictory across sources. A company-confirmed schedule would settle it.

Capital discipline needs a denominator. Until capital cost per megawatt and segment returns are disclosed, growth in megawatts is not evidence of growth in value.

None of these tests requires a view on artificial-intelligence demand, hyperscaler spending plans or power-market politics. They require only the segment's own operating disclosures, read in sequence. That is the advantage of a conversion-chain framework: the questions are answerable from the record, quarter by quarter.

The bounded conclusion

Iron Mountain's data-centre segment is, on the latest disclosure, a nearly full and highly profitable asset base attached to a large, partly pre-committed expansion. The 452 operating megawatts earn real money at a 52.6% margin. The 203 megawatts under construction are more than half spoken for. The 685 megawatts held for development are a claim on a future the company expects to be large.

Everything in between — the 13 MW signed against 3 MW commenced — is the part that is not yet decided. The gap is where construction schedules, utility interconnection and customer acceptance live, and it is where the segment's next two years of revenue will either materialise or slip.

A power-first expansion converts backlog into durable operating income only if energisation arrives roughly on schedule and capital costs stay inside the returns the leases were priced to deliver. Neither condition can be verified from the disclosure captured here. Both will be visible in the next several quarters of commencement data, pre-leasing percentages and segment margins — and if they fail, they will fail first in small numbers long before they fail in a headline.