What this covers

Jim Chanos, the short seller, is interviewed by Compound248 for Business Breakdowns. The conversation centers on a short case against legacy data center landlords, with Digital Realty as the main target. Chanos works through the accounting in detail. Companies report maintenance CapEx at about 10% of total capex, which against a 15-year depreciation life implies equipment lasting 150 years. Replacing failed air conditioning can be booked as growth spending if management says it will bring in a tenant or lift rents. He also says management's high-teens to low-20s marginal returns on stabilized centers work only by ignoring most capex. Leverage of roughly 9x at Digital Realty, and whether its investment-grade rating can hold, is a separate and contested point.

The conversation then widens to commercial real estate generally. Chanos explains why cap rate figures flatter reality: NOI excludes corporate overhead, and leasing costs are capitalized. In office, he argues, a 4% or 5% cap may be closer to 1% or 2%. He dates the office downturn to 2018-2019, before the pandemic. He also recalls that New York real estate took until 1995 or 1996 to recover from its late-1980s peak.

The last stretch is more general. Chanos compares this month's bank failures to the savings-and-loan crisis, framing them through Rate Duration vs Credit Duration. He argues that earning alpha requires a backed Variant Perception. He also argues that short sellers should be held to the same standard as bulls. He names what would prove him wrong: returns above the cost of capital, together with a pickup in data usage from AI.

Sharpest takeaway

Chanos argues that legacy colocation data center REITs are poor businesses whose incremental returns on capital have been below the cost of capital since about 2016, and that their reported economics are inflated by capex accounting, so their valuations are vulnerable as private equity turns from buyer to seller and rates rise.

  • Incremental returns on capital for legacy data center REITs have been negative or far below cost of capital since 2016, so growth destroys value
  • Maintenance capex is being reclassified as growth capex, and the 150-year implied maintenance life is an accounting artifact
  • Rising rates and private equity sellers push cap rates into the 8-10% range, versus the 5% caps at which the big REITs trade

The argument · threads12 threads · 76 claims
0.85

Reported legacy data center returns and earnings are overstated because capex and reinvestment are understated.

12 pointscentrality 5/5
  • Capital expenditure by AWS, Azure, and Google in data centers began to pick up around 2016 and 2017, and this shows up in the numbers and marks the start of the hyperscaler share gains.

    “the CapEx really began to pick up at AWS and Azure and Google in this space in 2016, 2017, and you see it in the numbers.”

  • Companies report maintenance capex at only about 10% of total capex, which against a 15-year average depreciation life implies that maintaining the air conditioning, HVAC, forklifts, and racks lasts 150 years, which is absurd.

    “So that means that they're telling you with a straight face that the maintenance CapEx for the air conditioning, the HVAC, the forklifts, the rack is 150 years.”

  • Since 2016 Digital Realty required $11 of new capital to generate $1 of new revenue at a 50% EBITDA margin, which is $0.50 of gross cash flow per $11 invested, implying a payback of more than 20 years on a gross basis before maintenance capex.

    “Digital Realty required $11 in new capital since 2016 to generate $1 in new revenues at a 50% EBITDA margin.”

  • Chanos suspects many companies are moving building operating expenses into SG&A to make their NOI look better.

    “we suspect a lot of companies are now putting more and more building operating expenses in SG&A to make their NOIs look better if you will.”

  • Replacing failed air conditioning in a data center is treated as growth capex, not maintenance, if management says the replacement will bring in a new tenant or allow rent increases, so the company can call the whole cost growth spending despite the replacement being unavoidable.

    “If you replace the air conditioning and you can say that you will bring in one new tenant or you will be able to raise rents on any kind of meaningful number of existing tenants, you can call the entire ticket growth CapEx.”

  • Legacy data center companies have spent 150% to 175% of their depreciation and amortization on capex for years, so reported EBIT is overstated and depreciation itself may be understated.

    “the data center guys have had CapEx at 150% to 175% of their depreciation and amortization.”

  • On an EBIT basis Digital Realty earns about a 2% return and Equinix 5-6%, and even adding back depreciation the returns are single digits for DLR and low double digits for EQIX.

    “They're 2% at DLR, and they're 5%, 6% at EQIX. And even if you add back the depreciation, the numbers are still single digits for DLR and low double digits for EQIX.”

  • EBITDA is not the right metric for legacy data center companies because it excludes the recurring capital spending needed to keep facilities operating, so it overstates the cash economics.

    “we'll get to that why EBITDA is not the right metric for these companies in a little bit”

  • REIT net operating income excludes corporate overhead, which for public REITs runs 50 to 150 basis points of capital per year, and companies may be shifting building operating expenses into SG&A to make NOI look better.

    “The first problem is that it doesn't include corporate overhead. And for most publicly traded REITs, that's anywhere from 50 to 150 basis points of capital a year”

  • The thesis would be wrong if legacy data center businesses showed returns on invested capital above the cost of capital for a heavily leveraged business, combined with a pickup in data usage from AI, but returns are far below that level so the path will be bumpy.

    “Show me ROICs above the cost of capital for a heavily leveraged business, and I'll change my mind.”

  • The fulcrum of the bear case is how management defines the money reinvested in the business, since it never seems to generate free cash flow and Digital Realty has no real revenue growth after inflation.

    “This is the fulcrum part of the bear case is, these guys will tell you their cash-on-cash marginal return is high teens, low 20s”

  • Digital Realty is shrinking on a real (inflation-adjusted) basis, and because the company is not growing in real terms, depreciation may be understated, meaning its reported EBIT is overstated.

    “in fact, DLR is shrinking on a real basis, it gets back to one of the real cruxes of our story, which is that depreciation is not only a real extent, it may be understated for these companies.”

0.64

Peak-price private equity purchases and peer multiples imply steep downside for legacy valuations and asset marks.

9 pointscentrality 4/5
  • Paying 40x existing EBITDA only makes sense if the strategic growth plan, such as new customers and geographies, works out, so buyers at that price are betting heavily on future execution.

    “So I would say that if you're paying 40x existing EBITDA, you would better hope that, that happens.”

  • Data centers are being offered for sale at cap rates in the high single digits, which is disastrous for valuations of the large publicly traded data center REITs that trade at much lower caps.

    “we thought that private equity would become a seller of data centers. And that's exactly what is turning out to be the case right now”

  • Private equity bought data centers at peak prices, including DigitalBridge's purchase of Switch at about 40x EBITDA, with a number of deals done at 25 to 30x EBITDA in 2020 and 2021, and these purchases are likely to prove regretful.

    “DigitalBridge's purchase of Switch, which just closed a month or 2 ago at 40x EBITDA. And a number of deals were done around 25 to 30x EBITDA in 2020 and 2021.”

  • DigitalBridge paid the highest prices for data center assets, so if the thesis is right its equity AUM risks being mismarked, which is bad news for its shareholders.

    “So the highest prices paid have been paid by DigitalBridge for these assets. If we're right, that's not good news for DigitalBridge shareholders”

  • Moving Digital Realty to a 9-10% cap rate, which is what its subsidiary, Cyxtera and GDS trade at, would imply a decline of roughly 70% or more from current levels.

    “DLR to be down 70 plus has to be a 9% or 10% cap”

  • Several publicly traded data center companies trade at roughly 9% caps, including Digital Realty's Singapore REIT DCRU, Cyxtera and GDS, while the two largest players trade at about 5%.

    “Digital Realty's own subsidiary, it's DCRU REIT in Singapore, trades at a 9% cap. Cyxtera trades at a 9% cap. GDS ... trades at a 9% cap.”

  • Given the risks and asymmetry of returns on the short side, a short seller who is wrong will recognize it quickly.

    “Given the risks on the short side and the asymmetry of returns on the short side, if the short seller is wrong, they're going to know it.”

  • Chanos would not be a financial buyer of data centers because he sees no economic return in them and does not want to compete with Amazon, Google and Microsoft.

    “I wouldn't be in this business. I don't think there's an economic return in it. And I don't want to be competing with Amazon, Google and Microsoft.”

  • Private equity managers with money to deploy often make late-cycle investments they later regret, and the data center buyers entered late and are finding the business is not what it appeared to be.

    “because they have money to put to work, they end up doing things they regret later.”

0.56

Legacy data center returns on incremental capital have fallen below cost of capital and keep worsening.

6 pointscentrality 4/5
  • Returns on incremental capital in the data center business peaked around 2016, and since then incremental returns have been poor and in some cases negative, meaning the business has gotten worse even as the stocks kept rising for a few years.

    “the business peaked pretty much around 2016. Ever since then, incremental returns have been terrible and in some cases, negative on capital.”

  • Legacy colocation data centers are a bad business whose true economic returns on capital fall below their cost of capital, and this gap is getting worse; Chanos looks for flawed business models first, with questionable accounting and poor management as added signals.

    “The problem with the colocation legacy data centers is it's just really a bad business and that underlines a lot of what we do on the short side.”

  • On a same-store basis, data center REIT returns have been flat for several years, and Digital Realty's same-store returns have been negative for the last handful of years.

    “on a same-store basis, the returns are, in many cases, flat or in the case of Digital Realty, DLR, negative on a same-store basis for the last handful of years.”

  • The risks to the short are a renewed rise in total data center usage, possibly driven by AI, combined with firm pricing and returns on capital that climb above the cost of capital; Chanos would change his mind only after seeing ROICs above cost of capital for a heavily leveraged business.

    “Show me ROICs above the cost of capital for a heavily leveraged business, and I'll change my mind.”

  • If the incremental return on invested capital is negative, growth does not help, because a company that grows is liquidating faster, so the faster the growth, the worse the value destruction.

    “if your incremental return on invested capital is negative then it doesn't matter how fast you grow. In fact, if you grow, you're liquidating faster.”

  • Management teams claim marginal cash-on-cash returns in the high teens to low 20s on stabilized centers, but Chanos cannot get near that, and the claim works only by ignoring about 90% of capex when calculating returns, which fools both the market and management itself.

    “If you're ignoring 90% of your CapEx in calculating your returns, I think you're not only fooling the market, you're fooling yourself.”

0.45

Digital Realty's cash burn and roughly 9x leverage contradict its investment-grade rating.

7 pointscentrality 3/5
  • Including distributions and before asset sales or acquisitions, Digital Realty burned about $2.7 billion last year, roughly $220 to $230 million a month, which must be financed or covered by asset sales just to keep operating.

    “Inclusive of distributions and before asset sales or acquisitions, we have them burning $2.7 billion last year. So that's about $220 million, $230 million a month.”

  • Digital Realty's interest cost on net debt was about 1.6% to 1.7% last year because it locked in fixed rates through green bonds and long-dated debt, but with negative cash flow and asset sales, its cost of debt can only go up from here.

    “Their interest costs on net debt last year was, I believe, 1.6% or 1.7%. Now that's the good news. The bad news is it's only going up from those levels.”

  • Digital Realty's net debt plus preferred is just over $19 billion on about $2.2 billion of EBITDA, roughly 9x leverage, which Chanos believes is not investment grade and risks a downgrade to junk, which would sharply raise funding costs.

    “net debt and preferred is a little over $19 billion. On $2.2 billion of EBITDA, you're almost at 9x leverage.”

  • Digital Realty says its $2.5-2.7 billion of capex is all growth while it is trying to sell assets for cash, which Chanos argues is contradictory because a cash-strapped company could simply cut growth capex.

    “why don't you just cut back on your growth CapEx? And we haven't gotten a good answer to that.”

  • Digital Realty must raise or sell roughly $230 million every month to keep the lights on, and with capital more expensive this need becomes an increasing problem, compounded by a flywheel that reversed in 2022.

    “the need to raise that amount of money every month is going to become an increasing issue here.”

  • If Digital Realty is short of cash and selling assets, it should cut growth capex first, and the company has not given a satisfactory answer for why it does not.

    “why don't you just cut back on your growth CapEx? And we haven't gotten a good answer to that.”

  • The rating agencies still rate Digital Realty just barely investment grade, a standing the speaker believes is not justified given negative free cash flow and rising leverage.

    “The rating agencies still have these guys just barely at investment grade. I don't think it is, not with the negative free cash flow and where the leverage is going.”

0.43

Data center REIT valuations rest on narrative and optimistic cap rates rather than fundamentals.

6 pointscentrality 3/5
  • Data center REITs are among the most expensively valued stocks in the market even though their underlying business is weak, creating a dichotomy between poor economics and very high multiples.

    “the data centers, represented by the big REITs, are some of the priciest stocks we see in the entire marketplace.”

  • The data center REITs kept performing for several years after returns peaked because they were riding a narrative that data growth benefits landlords, not because of fundamental analysis; Chanos prefers shorts where the business differs from the narrative.

    “really was on the back of a narrative than it was analysis in our view. And we love those kinds of shorts where the business is quite a bit different than the narrative.”

  • As rates rise, private equity buyers realize data centers are capital-intensive, so servicing debt and capex exceeds cash flow, and they become sellers rather than buyers, which is why data centers are being put up for sale at cap rates in the high single digits.

    “we thought that private equity would become a seller of data centers. And that's exactly what is turning out to be the case right now”

  • Wells Fargo reported private data center deals pricing at 8% to 10% cap rates, while Digital Realty and Equinix trade at implied caps of about 5.4% and 5.6%, and at an 8% cap the downside to the stocks is considerable.

    “we thought that private equity would become a seller of data centers. And that's exactly what is turning out to be the case right now”

  • Legacy data center companies are technology operating businesses that provide services, so they should be valued like tech companies, and at 100x earnings they do not deserve tech multiples, let alone a REIT valuation based on NOI or EBITDA.

    “these are technology operating businesses. They're service companies, and they should be valued really like a tech company, and they don't cold water at tech company valuations, not at 100x.”

  • Commercial real estate cap rates of 3% to 5% are gross cash flow measures that exclude corporate overhead, typically 50 to 150 basis points of capital a year, and also omit leasing costs that companies capitalize, so the true cap rate can be much lower than advertised.

    “when you are paying 3% caps and 4% caps and 5% caps for commercial real estate, everything has to go right, because that's a gross cash flow number.”

0.40

Hyperscalers are better businesses that both compete with and lease from legacy data center landlords.

5 pointscentrality 3/5
  • Enterprises keep data in one of three ways: running their own on-site IT, using colocation data centers where a third party maintains servers and facilities, or using hyperscale cloud providers such as AWS, Azure, and Google Cloud that host data and sell add-on services.

    “There's really three ways for an enterprise to maintain its data.”

  • Hyperscalers such as Microsoft and Amazon have higher returns, higher growth, and higher returns on capital than the legacy data center REITs, yet trade at lower cash-flow multiples, so an investor who wants data center exposure should buy them instead.

    “If you like this business, buy Microsoft or Amazon, they're cheaper. They're better businesses and they're cheaper.”

  • The hyperscale cloud providers are the largest tenants of the legacy data center operators even though they are also their biggest competitors, a bad position for a landlord, although the hyperscalers still lease space because building a new center makes no sense where they only need a fraction of a facility.

    “So you do have this bad dynamic where your largest competitors are also your largest tenants. That's never a position you want to be in as a landlord”

  • Cyxtera, which trades at a 9% cap and has economic troubles, is a substantial customer of Digital Realty, representing over 20% of the Singapore-listed DCRU REIT and a major but smaller share of consolidated Digital Realty.

    “In their DCRU REIT, they're, I think, over 20% of that. And I think quite a bit less, however, in the consolidated, but still a major client.”

  • Hyperscaler growth has dropped from roughly 30% and 40% to 10% and 20%, and this slowdown creates a question of whether legacy operators can claw back share or whether weaker data demand will push down occupancy and rents.

    “growth at the hyperscalers is downshifting dramatically... it's gone from basically 30% and 40% to 10% and 20%.”

0.36

Commercial real estate's fixed location and gross cap rates, like office decline, make it a poor asset.

9 pointscentrality 2/5
  • Chanos shorted office REITs in 2020 and watched them outperform in 2021 as investors saw them as safe, but they then fell sharply in 2022 and 2023 as vacancies rose and rent rolls dropped.

    “we got short the office REITs in 2020 and watch to our horror as the stocks outperformed in 2021 as people saw them as safe”

  • Real estate cannot be moved, so in good markets its location advantage is attractive, but in bad markets and poor credit conditions commercial real estate becomes a very bad asset, and New York City real estate did not recover from its late 1980s peak until 1995 or 1996.

    “It becomes in bad markets, poor credit markets, a really bad asset, and everybody forgets that. From '88, '89, the peak in New York City real estate at that cycle, New York City real estate did not begin to recover until '95 or '96”

  • Broker commissions, tenant improvements, and free rent can be capitalized and written off over the lease life, so capex for many office companies is actually increasing in a contracting market, and a 4% or 5% office cap could really be a 1% or 2% cap on true economics.

    “So you can capitalize your broker commissions, you can capitalize tenant improvements, you can capitalize free rent and write it off over the estimated life of the lease.”

  • Office rents and occupancy peaked in 2018, and office problems were visible by 2019, before the pandemic, and after the pandemic it became clear that many service businesses would not return to five days a week in the office, so incremental office demand would downshift.

    “Office rents and office occupancy peaked in 2018. ... it became apparent to us ... that a lot of service businesses were not going back to 5 days a week in the office post-pandemic.”

  • When commercial real estate trades at 3 to 5 percent cap rates, everything has to go right because the cap rate is a gross cash flow number that ignores costs, so the risk-reward looks poor.

    “when you are paying 3% caps and 4% caps and 5% caps for commercial real estate, everything has to go right, because that's a gross cash flow number.”

  • Commercial real estate's location fixity is both an advantage and a disadvantage, because a business can threaten to relocate across a river or to Florida but a building cannot move, so in bad credit markets it becomes a poor asset.

    “It's the advantage and disadvantage of real estate location. But it's not portable. You can't threaten to move the business across the river or to Florida.”

  • If total data demand remains strong, legacy colocation landlords could claw back market share from hyperscalers, but if demand is downshifting broadly, occupancy and rent pressure will continue.

    “growth at the hyperscalers is downshifting dramatically... it's gone from basically 30% and 40% to 10% and 20%.”

  • The firm concluded that service businesses would not return to five days a week in the office, so demand for office space would downshift, and this was the basis for shorting office REITs in 2020.

    “it became apparent to us... that a lot of service businesses were not going back to 5 days a week in the office post-pandemic. And that incremental demand for office space was going to downshift.”

  • New York City real estate peaked in 1988 or 1989 and did not begin to recover until 1995 or 1996, a lesson that investors tend to forget.

    “It becomes in bad markets, poor credit markets, a really bad asset, and everybody forgets that. From '88, '89, the peak in New York City real estate at that cycle, New York City real estate did not begin to recover until '95 or '96”

0.35

Higher rates and private equity retreat strain data center financing, capex, and supply dynamics.

3 pointscentrality 3/5
  • Many private equity data center deals were financed in mortgage REITs or pools that are not buyout funds, so managers earn management fees on committed capital and are unlikely to earn meaningful performance fees.

    “They're basically earning a management fee of the committed capital. And I would bet at this point, they're not expecting to see much in the way of performance fees.”

  • Equinix's interest cost on net debt was below 3% last year, showing that the major data center REITs had locked in very favorable financing before rates rose.

    “Equinix, their interest cost on net debt was below 3% last year.”

  • Chanos argues it is inconsistent to predict that high financing costs will reduce new data center building and tighten supply while also claiming the REITs are spending heavily on growth capex, because you cannot have it both ways.

    “Well, wait a minute, I thought there was all this growth CapEx from these guys. You can't have it both ways.”

0.34

Fact-based short selling is legitimate, and management must directly answer specific allegations.

8 pointscentrality 2/5
  • Short sellers are held to a much higher standard than bulls, even though bulls send out many buy recommendations without scrutiny, and Chanos believes short sellers should be held to the same standard, with the bright line being not trading on or inducing trades on information known to be false.

    “You cannot trade on or induce others to trade on information you know to be false. And that's the bright line.”

  • When management calls a short thesis outrageous lies and then refuses to address the specific allegations, offering only nondenial denials, it invites further scrutiny rather than ending the controversy.

    “But that is always a far better approach than saying these are outrageous lies and then you don't address them because at the end of the day if you have this sort of nondenial denial”

  • For a fundamental short, the other side is easy to find because bulls are highly paid and motivated sell-side analysts, while the thoughtful bear case is hard to hear, and this asymmetry lets short sellers test their hypothesis against plenty of pushback.

    “It's quite easy to see highly paid, highly motivated people and what they're thinking about stocks bullishly. But it's not so easy sometimes to hear the bear case on something”

  • Chanos does not produce large 40-page reports on short candidates, but he defends any short seller's right to publish them as long as the opinions are based on facts and the facts are not knowingly misstated.

    “having opinions about facts is what makes markets. We don't put out big reports, that's not our business model.”

  • Reed Hastings's response to a bear thesis, rebutting it point by point without recrimination and saying why he thought the short seller was wrong, is a gold-standard model for how management should respond to short sellers.

    “One of the gold standards was what Reed Hastings did a number of years ago to a bear thesis where he just rebutted it point by point thoughtfully without recrimination”

  • As a young sell-side analyst, Chanos was invited by a company's CEO to visit operations and debate a short recommendation, and the CEO rebutted it thoughtfully, with Chanos judging he was right on about half of the points and the company right on the other half.

    “the CEO invited me out to where they were and spent the day with me and with the CFO and thoughtfully rebutted what I believe. I think I was right at about half of it, and I think they ended up being right on about half of it.”

  • Chanos's hedge fund is effectively long the NASDAQ via QQQ and those hyperscale names, and short the legacy data center REITs.

    “In our hedge fund, we're long the NASDAQ, the QQQ. We're long those guys in effect and short the data centers.”

  • You cannot outperform the market by believing what the market believes, since that yields only beta returns, and an investor needs a meaningful, backed variant perception at major points to earn alpha, which applies to value buyers and fundamental shorts alike.

    “You have to have a variant perception. You're not going to outperform the market if you believe the same things that market believes.”

0.21

Bank exposure to commercial real estate creates vulnerability, as Silicon Valley Bank's failure shows.

3 pointscentrality 2/5
  • Local Law 97 fines are seen by city government insiders as a backdoor way to raise property taxes, since almost no building is expected to comply.

    “Exactly. It's a backdoor way to raise property taxes.”

  • Silicon Valley Bank failed because it was drowned in deposits it could not profitably deploy, and First Republic had a lesser version of the same problem through fixed-rate mortgages.

    “It did. It did. That's a great way to put it. First Republic, to a lesser extent, with fixed-rate mortgages.”

  • Regional banks hold most of the commercial real estate exposure in the banking system, making them the most vulnerable to commercial real estate losses.

    “The regional banks have most of the exposure.”

0.19

Power costs and New York regulation affect legacy data center operators differently.

2 pointscentrality 2/5
  • Power costs are a pass-through to customers for Digital Realty but not for Equinix, which hedges its power exposure, so rising power costs affect the two companies differently.

    “It's a pass-through for DLR. It is not a pass-through for Equinix, they go out and hedge.”

  • New York City's Local Law 97 requires buildings above about 25,000 square feet to meet strict green energy standards starting next year, applies retroactively to existing buildings, and imposes fines that a city official reportedly described as a backdoor way to raise property taxes, with even One Vanderbilt noncompliant.

    “It's retroactive, including residential. And the reason for that is they're going to charge fines if you're not compliant. And when I said to somebody, 'Well, nobody is going to be compliant.' My friend at city government says, 'Exactly. It's a backdoor way to raise property taxes.'”

0.07

Recent bank failures reflect duration and bond losses rather than credit, resembling the savings and loan crisis.

3 pointscentrality 1/5
  • The bank failures in the last couple of weeks were a credit duration and bond-portfolio problem, not an asset quality problem, because banks took in excess deposits and gambled in the bond market, so losses came from long-duration fixed-income holdings rather than bad loans.

    “what we just saw in the last 2 weeks in the U.S. was a credit duration problem, not an asset problem in terms of quality. It really was banks that took on excess deposits and gambled in the bond market that sunk.”

  • Regional banks hold most of the commercial real estate exposure, and the recent banking stress was mainly about bond and fixed-rate mortgage portfolios rather than credit, with Chanos not yet seeing credit problems.

    “The regional banks have most of the exposure. Again, I think what happened in the last couple of weeks since we've been talking has really been much more about bond and fixed mortgage portfolios than credit. I don't think we've seen the credit problems yet.”

  • The savings and loan crisis started as a rate duration problem in the late 1970s and early 1980s and turned into a credit problem and a fraud problem as the 1980s went on, which is the pattern Chanos sees in current banking stress.

    “the S&L started out as a rate duration problem in the late '70s, early '80s, and then morphed into a credit problem and a fraud problem as the '80s went on.”