Iron Mountain

IThe business

Iron Mountain began in 1951 in an underground facility near Hudson, New York, where it stored business records.³ Today it is a company of four trades under one name. It keeps paper and tapes in warehouses. It scans and processes documents. It leases data center capacity to the largest cloud companies. And it takes old computer hardware out of corporate offices and data centers, wipes it and resells or recycles it. It says it serves more than 240,000 customers in 61 countries, including about 95% of the Fortune 1000, and no customer is more than about 3% of revenue.³ It employed about 29,400 people at the end of 2025.³ It has been a real estate investment trust, a REIT, since 2014. Its headquarters is in Portsmouth, New Hampshire; in 2022 its releases carried a Boston dateline.⁴³

Most of what it earns still comes from the oldest trade. A customer puts boxes of records into one of its buildings and pays a monthly fee per box, plus charges for pick-up, retrieval and destruction. The company holds more than 740 million cubic feet of records. The average box stays 14.5 years, and 93.4% of records customers stay each year.⁷⁸ That is the business that the data center and recycling arms are being built on top of.

The pieces are not equal. Storage rental, the monthly fee for space, was $4.05B in 2025, 59% of revenue, and services were $2.85B.³ Records Management alone brought in $4.38B, data center $0.80B, and information destruction, which includes ALM product sales, $1.21B.³ The records business has a high margin because the same building and the same truck serve the same box for years.

Records: price, not volume

Here is what I find most important about the oldest trade. The number of boxes is nearly flat. Global RIM storage volume was 726.3 million cubic feet in the first quarter of 2024 and 740.5 million in the second quarter of 2026, up about 2% in two and a quarter years.⁷ Total volume, which adds the recycling and fine arts arms, was 747.9 million, up 1.6% in a year.⁷ Yet storage revenue per square foot rose from $10.07 in the fourth quarter of 2024 to $12.19 in the second quarter of 2026, up 21%, and storage profit per square foot from $8.14 to $9.57 C (calculated).⁷ The company calls this revenue management. It is price increases on customers who rarely move their boxes.

That is a good business so long as customers accept the increases. The 10-K lists the shift from paper and tape to other technologies as a risk.³ I cannot measure how much of each price rise customers absorb because they find it cheaper than moving, and how much they absorb because they have not noticed yet. What I can say is that the retention rate has crept up, from 92.6% to 93.4% in six quarters, which is not what a customer revolt looks like.⁷

The company owns only about a quarter of the square footage of its more than 1,300 buildings and leases the rest.⁸ It runs over 98 million square feet and owns 24 million.³ Storage rent, the lease cost of those buildings, was $501M in 2025, an operating cost the company pays before it reaches Adjusted EBITDA.⁵ That matters when I compare it with REITs that own most of what they use.

Data centers: the new engine

Iron Mountain operates 31 data centers in 21 markets, with 528.5 megawatts that can be leased, 97.1% of it leased, and says that five of the largest hyperscalers are customers.⁷³ A hyperscaler is a cloud giant that rents whole buildings, or large parts of them, at once. The leases are long. The weighted average term still to run is 10.2 years, and the new leases in the second quarter were priced at $205 per kilowatt per month.⁷ Renewal pricing was up 11.8% on a cash basis.⁷ Revenue was $263M in the second quarter, up 39%, at a 52.2% margin.⁷

The build-out is large. On 30 June 2026, 160.2 megawatts were under construction, 114.0 of them pre-leased, and 684.2 megawatts of land and shells were held for development, a total potential of 1,373 megawatts.⁷ The company leased 13 megawatts in the second quarter and 75 in July, so 110 megawatts in the first seven months, including 51 in Mumbai and 25 in London for hyperscalers.⁴⁸ Leasing comes in lumps. A quarter of 13 megawatts and a month of 75 are the same business seen at different times.

The project table shows what it costs. The seven projects with a stated budget add up to 103 megawatts and $1,449M, or $14.1M a megawatt C (calculated).⁷ A megawatt leased at the 2026 year-to-date average of $199 per kilowatt per month brings in about $2.4M of revenue a year, and at the segment’s 52% margin about $1.25M of EBITDA, a yield of about 8.9% on cost E (estimate).⁷ That yield is my estimate, built from the company’s pricing, margin and budget table. The company does not publish a development yield. It also shows that 45 of those 103 megawatts, London LON-3 and Madrid MAD-2/3, were not pre-leased.⁷

Digital and recycling

Digital solutions is a roughly $600M business inside Global RIM that scans documents and runs an AI-based platform called Insight DXP.⁸ Asset lifecycle management, or ALM, took in $333M in Corporate and Other revenue in the second quarter, up 67% for the segment and 88% for ALM, helped by an enterprise contract and by the decommissioning of hyperscale data center equipment.⁴⁷ The company bought 80% of ITRenew in January 2022 to build this arm.¹⁵ The market is large and fragmented, which the company puts at $35B.⁸ It is also a business that sells used hardware, so some of its revenue follows what servers and memory fetch on resale, and I cannot judge that from the filings.

How it got this big

Much of the growth was bought. In May 2016 it completed the Recall acquisition for $2.17B, paying $331.8M in cash and issuing 50.2 million shares.¹⁶ In January 2018 it paid about $1.34B for IO Data Centers, its entry into the large-scale data center business.²⁶ In January 2022 it acquired ITRenew.¹⁵ Since 2022 it has also spent $574M on Project Matterhorn, which it calls a program to accelerate growth through a global operating model, and which it treats as a one-off in AFFO.³ Revenue went from $3.51B in 2016 to $6.90B in 2025, and debt from $6.25B to $16.43B C (calculated).¹⁷

The year so far

The first half of 2026 was strong. Revenue was $3.97B, up 20.0% and 17.0% on an organic basis. Adjusted EBITDA rose 18.8% to $1.43B and AFFO 19.6% to $859M, or $2.87 a share, up 19.1%.⁴ The company raised its 2026 guide to revenue of $7.94B to $8.01B, Adjusted EBITDA of $2.945B to $2.975B and AFFO per share of $5.87 to $5.93, about 14% above 2025 at the midpoint.⁴ Its chief executive called it a record quarter and credited the growth plan.⁴ I give the sources of that growth in the next sections, because $134M of the $317M rise in second-quarter revenue, 42%, came from Corporate and Other, which earns the lowest margin C (calculated).⁴⁷

IIThe moat

A durable advantage is a cost or a habit that a rival with money and talent cannot copy in a reasonable time. Iron Mountain has four businesses, so the answer is four answers. I ask the same question of each: does it earn more on a box, a megawatt or a server than the next company would, and does the extra last?

In records the answer is yes. The first reason is trust. Customers hand over boxes of contracts, medical files and legal evidence. Moving them means a new vendor, a new audit and a new chain of custody, and the product is mostly invisible when it works. The second is the building and the truck network. A competitor needs warehouses, racking and a delivery fleet near the same customers before it can offer the same retrieval time. The third is inertia. The average box stays 14.5 years.⁸ A customer who finds a price increase irritating still has to pay to take the boxes out, and then pay another vendor to put them in.

The evidence is in the price. Storage revenue per square foot rose 21% in six quarters while volume rose 1.6% in a year C (calculated).⁷ That is what pricing power looks like. It is also what a slow squeeze looks like if customers decide to shrink their archives. I cannot tell which from outside. Records Management retention of 93.4% argues for the first reading.⁷

In data centers, the answer is weaker. A wholesale data center is a building with a lot of power and cooling, and the 10-K says customers choose on power availability, security, location, connectivity and price.³ Digital Realty and Equinix both reported record bookings in the third quarter, and Digital Realty alone has about 3.2 gigawatts of capacity against Iron Mountain’s 0.53.²⁷¹⁰ What Iron Mountain has is land and power it already owns, 684 megawatts of it, a 20-year history of running data centers, and relationships with five hyperscalers that it also serves in records and recycling.⁷³ That helps it win. It does not stop a customer from taking the next building from someone else. I do not count the data center business as protected. I count it as a good business that needs a lot of capital to stay in the game.

In ALM, the advantage is scale and a clean record of chain of custody. When a hyperscaler retires thousands of servers, it wants every drive wiped and logged. The company says it is the only player offering a single vendor for the whole job and describes the market as fragmented.⁸ The 10-K lists hyperscalers and corporate customers that manage their own recycling as competitors.³ I treat this as a moderate advantage in a young business.

The last row interests me most. The company says the opportunity is to offer its other services to its 240,000 customers. The 10-K says only about 5% of customers use more than one business unit.³ That tells me two things. The room to grow is large. And after years of effort, 95% of customers still use one thing. A cross-selling push that works would be visible in revenue growth outside the four trades. I have not seen it yet.

So I call this a strong advantage in the old business, a thin one in the new. The right comparison is the cash. The records business pays for the data center build, and the data center build pays for the share price. The advantage that matters most for the price is the one that is hardest to see.

IIIThe competitive advantage

The cleanest way to see how the market values Iron Mountain is to put it beside the companies it is compared with. Equinix and Digital Realty are the largest listed data center landlords. Extra Space Storage is a REIT that rents out space by the month, as Iron Mountain does in its oldest trade. Each reports a different adjusted measure of cash earnings, so the table compares the price against each company’s own 2026 guide, not against a common figure.

Two things stand out. Iron Mountain trades at 20.1 times its own AFFO guide, below the two data center landlords and just above the storage REIT. Its dividend yield of 2.9% sits between Digital Realty’s and Extra Space’s, because it keeps a large part of its cash to build.

The second thing is what that multiple implies. In 2025 the data center segment produced 15% of segment EBITDA, though 35% of second-quarter revenue came from the three growth businesses.³⁴ Suppose I value the rest of the company, which is mostly records, at 14 times EBITDA. That is about where the company traded in 2019 and 2021, when its own supplemental showed 12.3 times and 14.8 times total enterprise value to Adjusted EBITDA.²³²¹ The rest then comes to $33.6B. My enterprise value of $53.0B, which adds net debt and partner interests to the market value, leaves $19.4B for data centers that earn about $560M of EBITDA this year. That is 35 times E (estimate). At 13 times for records the data centers have to be worth 39 times, and at 15 times for records, 30 times E (estimate). The market is paying a data center multiple, and a high one, for a business that is mostly not data centers.

That is not an error by itself. The data centers are growing 39% and are far from finished. The 103 megawatts under construction with a stated budget would add about $130M of EBITDA at my 8.9% yield, and the 684 megawatts held for development would add a great deal more if it were built E (estimate).⁷ A buyer at $118.66 is paying for most of that in advance.

What the numbers say about quality

The operating results are good. Revenue rose 12% in 2025 and 20% in the first half of 2026. Adjusted EBITDA margin is 37%. The records margin is 44.7%. AFFO per share rose from $3.07 in 2020 to $5.17 in 2025, 11.0% a year C (calculated).²²⁵ The proxy shows a share price return of 247% over five years, from $100 to $347, against $138 for its peer group.¹² I do not argue with any of that. The question is the price paid for it, and what must come true to repay it.

One more point on the advantage. Because the company borrows to build, its cost of money is part of how well it competes. It is rated BB- by S&P and Ba3 by Moody’s, both with stable outlooks, and its newest bonds carry 6.25%.⁷⁶ A rival with a better rating borrows more cheaply and can accept a lower return on the same megawatt. In records, where cost of money matters less, that is no problem. In data centers it is the main way a better-funded rival beats it.

IVManagement

William Meaney has run the company since January 2013. Before that he was chief executive of the Zuellig Group, a private trading and distribution company, from 2004 to 2012.¹² He sits on the board of State Street.¹² Barry Hytinen has been chief financial officer since January 2020; he came from Hanesbrands and, before that, Tempur Sealy.¹² Mark Kidd runs data centers and ALM, and led data centers from 2013, having joined the company in 2003.¹² The independent chair of the board is Pamela Arway, a retired American Express executive.¹²

I could not find a succession plan in the filings I read. The proxy says only that a voluntary departure by Mr. Meaney would count as retirement, because he met the age and service tests at the end of 2025.¹² That makes a change possible without being scheduled. I cannot judge a person I have not met, but I can say that a 13-year chief executive who has led a tripling of the share price is a key-person risk.

How they are really paid

Mr. Meaney was paid $17.05M for 2025, of which $11.40M was stock awards, $2.10M options, $2.27M a cash bonus and $1.20M salary.¹² In 2024 it was $18.36M and in 2023 $14.90M.¹² The ratio of his pay to the median employee’s $37,949 was 449 to 1.¹² Because the stock more than tripled, the proxy’s measure of pay actually received, which counts the change in value of awards, was $118.5M for 2024.¹² The pay itself is not what concerns me. The measures are.

The annual bonus is 40% on Adjusted EBITDA and revenue, 30% on AFFO per share in constant currency, and 30% on strategic goals the board sets. The 2025 formula paid 90.2% of target before a 20% personal adjustment for the chief executive.¹² The long-term awards are performance units over three years. They vest on return on invested capital in the third year, which must reach a minimum, on total revenue, and on total return against the MSCI US REIT Index.¹² The chief executive also receives options.¹²

What does this pay them to do? To grow revenue and EBITDA, and to beat REITs on the share price. Only the bonus, and only 30% of it, is tied to a per-share cash measure. A revenue target is easiest to reach by building and buying, and building here is paid for with borrowed money. I do not say the team behaves that way. I say the pay allows it, and it is the reason I look hard at the record of what they did with the money. The minimum return on invested capital in the long-term awards is a useful brake. I could not find the level it is set at.

Their own money

Mr. Meaney reported 394,307 shares at 4 March 2026, worth about $47M at $118.66 C (calculated), and 2,620,582 vested options.¹²¹ Executives must own a multiple of salary, six times for the chief executive, which is $7.2M C (calculated), and hedging and pledging are banned.¹² All directors and executive officers together held 2,445,554 shares, 1.7% of the company with options, about $290M C (calculated).¹² The longest-serving director, Kent Dauten, holds the most.¹² These are real stakes, and the guideline is met many times over.

The trading is the other side. In the filings from 2 January to 3 November 2026, Mr. Meaney sold 779,382 shares for about $85.8M C (calculated).¹³ Much of it follows a monthly pattern: he exercises 38,474 options struck at $37 and sells the shares, on the first days of the month. That looks like a pre-arranged plan, though I did not verify one. In March he also received 650,674 shares from vested awards, and had 256,040 withheld for tax.¹³ Three other executives sold in smaller amounts: Mark Kidd 54,000 shares for $6.1M, Mithu Bhargava 59,725 for $6.2M and Greg McIntosh 26,839 for $3.2M C (calculated).¹³ I found no open-market purchase by any officer or director in 2026.¹³ Selling is normal for executives with large awards. It also means that the people with the most information are not adding to their stake at a price near its high.

What they did with the money, over a full cycle

I take 2016 to 2025 as the cycle. It includes the Recall deal, a pandemic, a rate shock, and the AI build-out. Over those ten years operating cash flow added up to $9.5B. Dividends took $6.9B, capital spending $9.2B and acquisitions $3.8B C (calculated).¹⁷ The difference came from borrowing. Long-term debt went from $6.3B to $16.4B C (calculated).¹⁷ The company has not bought back shares in this period; the XBRL record shows none since 2012.¹⁷

I draw three conclusions. First, they paid shareholders very well: 18.9% a year on price and dividends, with the dividend up 61% from $2.01 to $3.22 C (calculated).¹²⁸¹⁷ The multiple did a lot of that work, as the chart in Valuation shows. Second, growth was paid for with debt, not retained cash, and the amounts are rising. Capital spending was $0.33B in 2016 and $2.27B in 2025.¹⁷ Third, they did not issue common stock to pay for it, which protects the per-share numbers today. The 2023 to 2025 cash flow statements show no sale of common stock, and partner equity in data center joint ventures came to $231M, $93M and $95M in 2024, 2025 and the first half of 2026.³⁶ A common share sale remains the lever they have not pulled. If the market closed to the company's bonds, it would be the next one.

There are things I like less. Project Matterhorn cost $574M over four years and was added back in full to AFFO. That is $532M of cost in 2023 to 2025 alone C (calculated). A one-off that lasts four years is a running cost. The company says cash restructuring ends in 2026 and I count that in its favour in my numbers.³⁸ Stock pay doubled from $74M in 2023 to $140M in 2025, and it is also added back.¹⁷ And the dividend was flat at $2.47 for three years, 2020 to 2022, while debt grew.¹⁷

VFinancials

A real estate company’s accounts hide as much as they show. Net income is struck after depreciation and amortization of $1.02B in 2025, which is up 57% since 2020.¹⁷ So nobody values it on earnings. Analysts use funds from operations, or FFO, which adds back depreciation on buildings. The company goes further, to an adjusted FFO called AFFO, which adds back more items and takes out the money for recurring upkeep. I want a third step, what an owner could actually take out, and that is an estimate.

Why not plain earnings

Diluted earnings per share were $0.49 in 2025. AFFO per share was $5.17. Price over 2025 earnings would be about 240 times, and that says almost nothing C (calculated). Price over AFFO is 20.1 times the 2026 guide, which tells me more. But the company’s AFFO leaves out three real costs, and the table walks through them.

I take each deduction in turn, because each can be argued.

Stock pay. Stock-based compensation was $140M in 2025, up from $74M in 2023.¹⁷ The company adds it back to Adjusted EBITDA, FFO and AFFO. But it is pay. It dilutes holders, and the company paid a net $94M in the first half of 2026 in connection with employees’ stock awards, so cash goes out as well.⁶ I deduct about $150M for 2026 E (estimate).

The cost of winning customers. The company capitalizes the cost of customer relationships, commissions and intake work, and adds back their amortization, $286M in 2025, to reach AFFO.⁵ The cash for these things is in investing cash flow, not capital spending: $75M for customer intangibles and $96M for contract costs in 2025, and $70M in the first half of 2026.³⁶ In a business that grows by selling, this is a running cost. I deduct $140M E (estimate).

Upkeep. Recurring capital spending was $147M in 2025 and is guided to $150M in 2026.⁵⁶ Non-real-estate depreciation, which covers racking, trucks, shredding equipment and computers, was $296M in 2025.⁵ Some of that gap is for the growth businesses, but not all. If half of the $210M difference is real upkeep, that is $105M E (estimate). The range I show is $50M to $200M, and I cannot narrow it from outside.

The result is about $4.58 a share, 22% below the AFFO guide and nine times the earnings per share. It is not that AFFO is wrong. It is that the company defines a number that suits a growth story, and an owner has to look through it. At $118.66 the price is 25.9 times my owner earnings. I use the company’s AFFO in the valuation because that is what peers report and what the market watches, and I let the haircut show up in the discount rate, the payout and the multiples I choose.

Ten years

The chart shows the pattern. AFFO per share was $2.62 in 2016 and $5.17 in 2025, a rise of 7.8% a year C (calculated). From 2020 it grew 11.0% a year, faster than before, and the 2026 guide is another 14%.²⁸²²⁵⁴ Earnings per share tell a different tale: $1.90 in 2022, $0.63 in 2023, $0.61 in 2024 and $0.49 in 2025.¹⁷ The gap is depreciation on new data centers and interest on the debt that paid for them, which the company adds back or capitalizes. The dividend was $2.47 for three years and has risen 40% since, to $3.456 a year.¹⁷⁴

Revenue, shown in the first chart, grew from $3.51B in 2016 to $6.90B in 2025. Adjusted EBITDA grew from $1.09B to $2.57B, a rise of 2.4 times against 2.0 times for revenue, so margins widened from 31% to 37% C (calculated).¹⁷²⁸⁵ Part of that is data centers, which earn a higher margin. Part is the records price increases. Part is the end of the Recall integration. In 2025 revenue grew 12.2% and organic revenue 10.2%, the best in the series.⁵

Who paid for the growth

Look at the chart for what happened after 2021. Operating cash flow rose from $0.76B to $1.34B, a good record. But capital spending went from $0.61B to $2.27B, and dividends from $0.72B to $0.92B. From 2021 to 2025 the company spent $5.4B more than it earned in cash from operations C (calculated). Debt rose $7.7B in those years.¹⁷ The money did not come from selling shares or assets. It came from lenders and, in the data center joint ventures, partners.

I want to be fair about this. A growth company can sensibly borrow when the return on a project exceeds the cost of the money. The question is how much of the cash flow depends on that being true, and the next table separates the two.

The dividend is not the problem. It is covered 1.6 times by the cash the base business generates, and AFFO payout is 59%, which the company wants to keep in the low 60s.⁸ The problem is the other $2.05B. Every dollar of growth is financed from outside. The company says the growth in Adjusted EBITDA supports more than $2.0B a year of borrowing at the same leverage.⁸ At 4.8 times that is right if EBITDA grows by about $400M a year, which it did in 2025 ($338M) and is guided to do in 2026 ($386M at the midpoint) C (calculated).⁵⁴ If EBITDA growth slowed while the spending continued, leverage would rise.

The balance sheet

Long-term debt was $17.35B on 30 June 2026, and net debt $17.28B, which the company puts at 4.8 times lease-adjusted EBITDA, inside its 4.5 to 5.5 times target and well inside the 7.0 times its credit agreement allows.⁷⁶ The rating is BB- from S&P and Ba3 from Moody’s, both with stable outlooks, which is below investment grade.⁷ Adjusted EBITDA covers interest 3.3 times and the fixed charge coverage ratio is 2.5 times against a 1.5 times floor.⁷⁶ Weighted average interest is 5.6%, 83% of the debt is fixed, and the average maturity is 4.6 years.⁷ A 1% rise on floating-rate debt cost $41.9M of net income at the end of 2025, about $0.14 a share C (calculated).³

The maturity wall is the part I watch. Debt due from 2027 to 2031 is $11.3B, with $3.5B in 2031 alone.⁷ The newest bond, 6.25% notes due 2035 sold in June 2026, costs more than the 4.875%, 5%, and 5.25% notes due in 2027 to 2030.⁶ If the whole 2027 to 2031 stack is refinanced at about 6.25%, interest rises by roughly $120M a year, $0.40 a share at full effect. I model a drag of $0.35 in the base case, $0.50 in the bear, which assumes 7% money, and $0.20 in the bull, which assumes lower rates E (estimate).

The balance sheet is stretched but not at the edge. Book equity is negative, at −$0.96B in June 2026, which mostly reflects years of dividends and buildings carried at cost less depreciation; I do not read it as distress.⁴ The real limits are the ratings and the lenders’ patience. Leverage has fallen from 5.7 times in 2016 to 4.8 times, while Adjusted EBITDA rose 2.4 times.²⁸⁷ That is an argument in management’s favour.

What the history of the multiple shows

The price has risen 3.7 times from the first close in my data. AFFO per share doubled. The other factor of 1.9 is the multiple. Over the first ten October closes it ranged from 8.5 to 27.3 times AFFO, with a median of 12.9 C (calculated).^{12825242322212019185} The company changed in those years from a records business to a records business with a growing data center arm, and the market changed its mind about what that earns. I cannot say where the right multiple lies. I can say it was 12.4 times in 2016, and 20.1 times now.

VIValuation

Start with what would go wrong

Before any numbers, I ask what would most likely make this a bad investment from here, and work back from that. For Iron Mountain the old business is not the likeliest culprit. The likeliest problems are the price, the debt and the new business. Here are three ways it goes badly wrong, in the order I think most likely.

1. The business does fine and the multiple falls. The stock now trades at 20.1 times the AFFO guide. The median over the first ten Octobers was 12.9 times, and the low was 8.5 C (calculated).¹²⁸⁵ If my base AFFO comes through and in 2036 the market pays 13 times it, close to that median, the return from $118.66 is about 6.4% a year with dividends E (estimate). That is only 1.5 points above a ten-year Treasury at 4.90%.²⁹ Nothing has to break for this to happen.
2. Leasing slows while the building does not. The company has committed $1.45B to 103 megawatts of development, and 45 of those megawatts have no tenant.⁷ It plans $2.2B of capital spending this year, with about $1.5B of it funded from outside E (estimate).⁶ If hyperscale demand paused for a few quarters, as it can, the spending would continue on the debt and leverage would climb from 4.8 times towards the 5.5 times top of its range. The company would then have to slow down, issue shares or accept a lower rating.
3. Money gets dearer and records stop absorbing price rises. About $11.3B of debt falls due from 2027 to 2031, and the newest bond costs 6.25%.⁷⁶ Refinancing it all at that rate takes about $0.35 a share out of AFFO E (estimate). At the same time the records business has had 21% more revenue per square foot from the same boxes in six quarters.⁷ If customers start to resist, or move boxes to cheaper storage, the growth falls first in the part that funds everything else.

I can’t put honest odds on the second and third. I can show what the first does, and I do below.

What an owner can take out, and how far it is from reported earnings

The financials walk from earnings of $0.49 a share to AFFO of $5.17. I take the further step to what an owner could pull out without letting the business run down. For 2026 that is $4.58 a share, with a range of $4.10 to $4.97 E (estimate). The company’s AFFO guide of $5.90 is 1.29 times that. The difference is stock pay, the cash cost of winning customers and upkeep the company treats as growth. I explained each above, and I cannot narrow the upkeep from outside. At $118.66 the price is 25.9 times owner earnings, against 20.1 times AFFO.

What the price implies

The data center row needs a word. I take total enterprise value, subtract 14 times the EBITDA of everything except data centers, and divide what is left by the data centers’ expected EBITDA. The records multiple is a guess anchored on what the company itself reported in 2019 and 2021.²³²¹ If the guess is wrong in either direction the answer moves, which the Competitive advantage section showed. But the direction is clear. The market is paying an AI infrastructure price for a segment that earns 15% of segment profit.

Reverse DCF: what the price needs

At my 9.5% hurdle, the price needs AFFO per share to grow 11.2% a year for ten years, from $5.90 to about $17 before the drag, and the dividend to grow with it C (calculated). That is almost exactly what the company did from 2020 to 2025, 11.0% a year.²²⁵ So the price asks for the last five years to be repeated for ten, on a base that is three times as large as in 2020. My base has 6.5% a year, which already fades from 11.5% in 2027. At an 8.5% hurdle the price needs 8.9% a year, and at 10.5% it needs 13.4%. The verdict depends on the rate I require, and on whether the first half of that decade, which was a build-out, can last twice as long.

The discount rate

I discount at 9.5%. The capital asset pricing model with the 4.90% ten-year Treasury and Damodaran’s 4.20% implied equity risk premium gives 9.1% on Iron Mountain’s beta against the market over 119 months, 9.9% over 60 and 10.7% over 36 C (calculated).²⁹³⁰³¹ The three betas are 1.01, 1.19 and 1.39, from price returns with dividends not added. I chose 9.5% because it is the cost of the company’s newest debt, 6.25%, plus a premium of 3.25 points for owning the equity behind $17B of it, and because it sits within the range of the three betas. My Costco page uses 8.5% and my Enbridge page 9.0%. A BB- borrower with a build-out underway deserves more than either, in my view. So I show the whole range below. At 9.0%, my dividend discount base is $94, against $86 at 9.5%.

The best alternative use of the same dollar is a bond. The Treasury pays 4.90%. Iron Mountain’s own notes due 2035 pay 6.25% and rank ahead of the shares.²⁹⁶ A holder of the shares needs a reason to expect more than that, and the AFFO yield is 5.0% with growth on top. Inside this site, Enbridge is the nearest comparison for a heavily indebted income stock: its page has a price 6% below its base value, a 5.6% dividend yield, and a base return of 9.9% a year against a 9.0% hurdle. Iron Mountain’s dividend yields 2.9%, and my base returns 7.8% a year against 9.5%, with the price above my base value. On the demand side of the same build-out, Vertiv, which sells power and cooling to data centers, shows a price 89% above its base value when I wrote it, and Constellation Energy, which sells electricity to them, shows 20% above. I do not think those three are independent bets. They share one view, that AI infrastructure spending continues, and that is also the main support for Iron Mountain’s growth case.

Three ways to value it, and one rule

I value the company three ways and take the same weight from each. The rule is set before the numbers: bear, base and bull are the plain average of a dividend discount, a net asset value and a multiple of AFFO. None of the three is tuned to agree with the others.

The dividend discount is the cash an owner is paid. I take AFFO per share, grow it in each case, subtract the refinancing drag, pay out a share of it that rises from 59% towards 72% as growth capital spending slows, discount at 9.5%, and add a value after 2036 equal to the dividend divided by the discount rate less growth. The net asset value values the businesses by what they earn at an assumed cap rate, adds the data centers under construction at their expected value less the cost to finish, and subtracts debt and partner interests. The multiple method applies a price to AFFO taken from the stock’s own history and the peer table. Each is a different way to be wrong.

Net asset value

This method asks what the parts are worth. It is the usual way to value a REIT, and its weak point is the cap rate. A cap rate is annual profit divided by value, so 7% is the same as 14.3 times profit. Iron Mountain publishes no cap rates and no net asset value, and the records business is not a block of rentable offices, so every rate here is my estimate. Here is how I built it.

For the records, digital and ALM businesses I take 2026 Adjusted EBITDA of about $2.40B, less $133M of recurring capital spending and less $150M of stock pay, and capitalize it at 7.0% in the base, 8.0% in the bear and 6.25% in the bull E (estimate). The base rate is 14.3 times. It sits at the top of the 12.3 to 14.8 times the company showed in 2019 and 2021, and below the 17.9 times at which the whole company now trades.²³²¹⁴ For the operating data centers I take $560M of 2026 EBITDA less $17M of recurring spending, capitalized at 6.0% in the base, 7.0% in the bear and 5.25% in the bull E (estimate). For the 103 megawatts under construction with a budget, I assume a stabilized yield of 8.9% on the $1,449M, from the pricing and margin shown earlier, capitalized at the same rate, less the $472M still to spend E (estimate).⁷ I add the 684 megawatts held for development at $0.30M a megawatt, a land-and-power price I cannot verify E (estimate).⁷

The grid shows how much the cap rates matter. A rise of three-quarters of a point in the records cap rate, from 7.0% to 7.75%, takes about $9 a share off the base.

The multiple method, and the exit hidden in a discount

The multiple method asks what the market has been willing to pay. Over the first ten Octobers it paid 8.5 times AFFO at the lowest and 12.9 at the median, and in October 2024 it paid 27.3 C (calculated).¹²⁸¹⁸ I use 13.5 times in the bear, close to the median, 17 times in the base, between the median and where the stock trades now, and 23 times in the bull, between Digital Realty’s 22.2 and Equinix’s 24.0 E (estimate).²⁷¹⁰⁹ The reason the base is above the median is that the company is no longer only a records business. The reason it is below the present multiple is that most of its profit still is.

A dividend discount has a multiple hidden inside it. When I stop in 2036 and capitalize the dividend, I am assuming a buyer then pays 11.7, 11.4 or 11.4 times AFFO in the three cases C (calculated). Today’s price is 20.1 times. That hidden gap is much of why the dividend discount sits below the price. Here is what the base case returns if the buyer in 2036 pays more.

Only the last row, a buyer in 2036 who pays a multiple near today’s data center peers, returns more than my 9.5% hurdle. That is the real bet in this price: not a faster business, but a multiple that stays at the level of a company the market has decided is digital infrastructure, ten years from now.

Three cases

My base case is Iron Mountain carrying on roughly as it has: records revenue growing 6% to 7% on price, data centers energizing the 325 megawatts the company expects over the next 24 months and leasing a similar amount again, ALM growing but with lumpy margins, and AFFO per share growing at the pace it faded to over ten years.⁸ That gives AFFO of $8.76 in 2031 and $11.12 in 2036 E (estimate). The bull case has AFFO per share growing 14% and then 12.5% for two years, as the company has done on average since 2020, with a lower refinancing drag and a market that sees a digital infrastructure company. The bear case has leasing pause, records price rises fade to the level of inflation, and the new money cost 7% E (estimate).

What I can’t judge

Some inputs I can’t know, and I would rather say so than pretend. I don’t know how much of the $2.2B of annual capital spending is really upkeep; I show a range from $50M to $200M of the gap. I don’t know the development yield, because the company does not publish one, and my 8.9% is built from its lease pricing and budget table. I don’t know the cap rate at which a records business sells, because there are no recent comparable sales. I don’t know how much of the AI-driven demand for capacity is lasting. I don’t know how much of the 17% organic growth in the first half is price and how much is one-off recycling contracts. And I don’t know the lease terms on the 45 megawatts without a tenant. Where I can’t know, I give a range and a reason.

Multiply it out

My base needs several things at once. Hyperscale demand keeps filling new capacity at today’s rents: I would give that 70%. Records pricing keeps growing storage revenue 7% or more with flat volume: 80%. Debt stays available near 6.25% and leverage stays near five times: 75%. Construction and power arrive on time and on budget: 75%. And no large equity raise at a low price: 80%. Multiplied, that is about 25% E (estimate). They overlap, since a good market helps every line, so the true figure is higher, perhaps 35%. And missing one does not ruin the case. But the price needs more than my base. It needs all of that, and a buyer in 2036 who pays what the market pays for data center landlords today.

Quality and price are two questions

How good is the business? Good. The records business has a 44.7% margin, a 93.4% retention rate and pricing power that has shown up in every quarter I checked. The data center business has long leases, hyperscaler customers and a 52% margin. Revenue and AFFO have grown for years, and leverage has fallen. How good is the price? Poor, on my numbers. At $118.66 it is 35% above my base and 7% above my bull, and even the multiple method, the most generous of the three, finds $136 only in the bull. A good business can be a poor purchase at the wrong price. On this evidence that is what this looks like. I say that knowing the stock has returned 18.9% a year over ten years. The question is the next ten.

The margin of safety

My margin-of-safety price is $62, 30% below my base value. At $62 the dividend discount returns 10.2% a year in the bear, 11.8% in the base and 13.0% in the bull C (calculated). The bear case at that price returns a little above my 9.5% hurdle. I use a 30% cushion for businesses with predictable rents. Here the debt, the build-out and the gap between AFFO and owner earnings argue for at least that much. The price is 91% above it. The stock last closed below $62 on 14 November 2023, at $61.88. It last closed below my base value on 8 January 2026, at $87.32, and below my bull value at $109.96 eight trading days before my price date.¹ Third-quarter results are due on 5 November and are not in my values, which use the second quarter and the guide.¹⁴

VIIThe case for the price

What most people miss

Three facts cut in the company’s favour. First, the dividend is not what the debt pays for. The base business generates about $1.6B after upkeep and customer costs, and the dividend takes $1.0B, so it is covered 1.6 times before any growth spending E (estimate).⁸⁶ Second, leverage went down while the business grew. Net lease-adjusted leverage was 5.7 times in 2016 and 4.8 times in June 2026, over which Adjusted EBITDA rose 2.4 times C (calculated).²⁸⁷ Third, the build-out is mostly pre-sold: 114 of the 160 megawatts under construction are pre-leased, a backlog the company says supports another $370M of data center revenue growth beyond 2026, before any lease signed since June.⁷⁸

The bull case

In plain words: records are a durable cash business growing on price, and leasing data centers to hyperscalers adds a second engine growing 39% on long leases. In August William Meaney called the second quarter a record, pointed to 110 megawatts leased through July and raised the 2026 guide.⁴ Its presentation says the dividend will grow in line with AFFO per share and that data centers are underwritten on 10 to 15 year pre-leases.⁸ Tobey Sommer of Truist raised his price target to $140 from $130 on 1 May 2026, citing data center lease trends.³² A target is an opinion about a price a year out, not a measure of value. I found no checkable current statement from any well-known investor, so I name no one else.

Against reality

Much of the bull case is true today. First-half revenue rose 20%, data center revenue 39%, AFFO per share 19%, and the guide went up.⁴ If the company keeps leasing 100 megawatts or more a year near $200 per kilowatt a month, holds leverage near 4.8 times and borrows at 6.25% or less, a price near my bull value of $111 is possible, and the multiple method alone gives $136 at 23 times AFFO E (estimate). The price is $118.66. It needs AFFO per share to grow 11.2% a year for ten years, to about $17 before the drag, against my base of $11.12 after it C (calculated). That is what the company did from 2020 to 2025. My bull case fades from it, as growth on a larger base needs more capital. The price also rose 8% in eight trading days with nothing new in the filings, and third-quarter results are due on 5 November.¹¹⁴

What a 10x would take. Ten times $118.66 is $1,187 a share, about $356B of market value C (calculated). At today’s 20.1 times AFFO that needs about $59 of AFFO per share, ten times the $5.90 guide, or about $17.7B a year in total against $1.77B guided for 2026 C (calculated).⁴ In ten years that is 26% a year, against the 11% of the last five. At today’s margins it means revenue near $80B, ten times the $8B guided, for a company that doubled revenue in nine years.⁴¹⁷ Its shares rose 3.7 times in ten years, with AFFO per share up 2.0 times and the multiple the rest C (calculated).¹²⁸⁵ A tenfold move is not plausible, and I would not build a case on it.

Where I come down

The second quarter and July leasing moved me: they put my 2026 base on the guide, and 75 megawatts in a month show the slow second quarter was timing. They did not move me on the price, which sits 35% above my base and 7% above my bull, or on the gap between AFFO and owner earnings. Leasing above 150 megawatts for the year at stable prices, a 2027 guide above $6.70 a share, or a lower ten-year Treasury would raise my values. They stay at $69, $88 and $111.

VIIIRisks, and what would change my view

The three ways this goes badly wrong are at the start of Valuation. These are the rest.

• Key person. William Meaney has been chief executive since 2013, and the proxy says he met the retirement tests at the end of 2025.¹² He sold 779,382 shares in 2026 on a regular schedule.¹³ A departure without a named successor would be a risk for a company in the middle of a build-out.
• Customer and demand cycle in data centers. A small number of hyperscalers take most of the new capacity, and their own build plans, power supply and chip purchases decide how much they rent. Leasing was 13 megawatts in one quarter and 75 in the next.⁷⁴ In the latest session Vertiv’s shares fell 12.9% in one session while this stock rose 1.7%, which shows how far the market’s view of this theme can swing in a day.⁹¹
• Credit and refinancing. The debt is rated BB- and Ba3. Interest cover is 3.3 times and about $11.3B falls due by 2031. A downgrade or a closed bond market would raise the cost of every renewal.⁷ Seventeen percent is floating rate.⁷
• Decline of paper. Storage volume is flat only because customers keep their boxes. The 10-K lists the shift from paper and tape to other technologies as a risk.³ If customers start destroying more than they store, the pricing that carries the earnings would be tested.
• ALM and resale prices. Corporate and Other revenue rose 67% in the second quarter, to $333M, and the segment lost $74M of EBITDA in the first half after overhead.⁷ The business sells hardware and components, so it depends on resale prices that I cannot forecast.
• Construction, power and equipment. The company is building 160 megawatts and plans many more. Delays in power connections or equipment would defer rent while the interest on the borrowing continues.⁷ It capitalized $78M of interest in 2025, which would hit earnings once the buildings open.³
• Tax status and structure. It must pay out most taxable income as a REIT, and holdings in taxable subsidiaries such as the recycling and digital businesses are limited. That limit rose from 20% to 25% of assets on 1 January 2026.⁶ Growth in those subsidiaries makes the limit more relevant.
• Currency. About 34% of 2025 revenue came from outside the United States, and 15% of debt is not in dollars.³⁷ Revenue growth in the second quarter was 18.5% as reported and 17.6% in constant currency.⁴
• Information. The company does not publish a development yield, a cap rate for its buildings, owner earnings or a net asset value. The company changed its AFFO definition at least three times since 2016, most recently for 2023.²⁸²²¹⁹ My values rest on estimates that I explain, and the answer moves with them.

What would raise my base value: the third-quarter report on 5 November showing leasing above 150 megawatts for the year and a stable AFFO guide, and then a 2027 guide above $6.70 in February. What would lower it: a guide for 2027 near the 2026 pace of growth or below, new bonds priced at 7% or more, or a pause in data center leasing while building continues.⁴¹⁴