DuPont Fabros Technology is a triple-net REIT that develops and owns wholesale data centers in the US.  As of FY2012, it owns 11 data centers in 4 submarkets with 205.5 MW of critical load.  Additionally, it is also developing 3 sites with capacity of 49.4 MW.

On the surface DFT is a great business with stable cash flow in what should be a growth industry.  The emergence of mobile devices and cloud computing is driving up demand for data centers.   It trades at lower multiples than its data center peers.  It looks even more attractive compare to retail triple-net REITs like NNN and O that are trading at 19 – 20x 2013 FFO

Relative Valuation

*Source: P/ FFO from Bloomberg; Cap rate from Green Street Advisors

I had written about DFT as a long opportunity before and have taken a second look after its recent disastrous 3Q12 earnings call had caught my attention.  I believe, after further analysis, the market has mis-priced DFT.  Shorting DFT at $24 is a compelling opportunity because there are several problems not being reflected in the stock price.

Problem 1: Wholesale data centers are commoditized

Ever wonder how Google can seem to answer any questions instantly?  The answer lies within its “data factories”.  First, a group of computers index the internet and literally crunch data nonstop 24/7.  The results are sent to an interconnected network of computers which then forwards the results to the end users.

The first group of computers, along with other IT infrastructure, are housed in “wholesale data centers” where interconnectivity and latency is less of a concern.  These data factories are akin to the back office of an investment bank.  Its function is vital, and it does not matter where the work is done.  Wholesale data centers compete on cost and efficiency.

The second group computers are housed in data centers called “internet exchange points”.  These IXPs are where AT&T talks to Verizon.  IXP is a classic example of the network effect and has wide “moat”.

DFT’s portfolio is 100% commoditized, wholesale data centers with limited moat.

Problem 2: Headwind against at least 51% of revenue

The world of technology of today is vastly different than 5 years ago.  Before the arrival of smartphones and cloud computing, most companies would outsource their data centers to companies like DFT.  Now, at least to internet companies, data centers have become a core part of strategy.  DFT’s top 3 tenants are Microsoft, Facebook and Yahoo.  Each has the technical expertise and financial resources to develop and operate its own centers.   In fact, they are all owners and renters at the same time. 

All else being equal, the top 3 and other big internet companies such as Amazon, Google and Oracle, prefer owning over renting.  This dynamic has placed a cap on DFT’s rental growth because rent has to be significantly lower than owning for the tenants to renew. 

To management’s credit, DFT has successfully lowered its tenant concentration.  The top 3 represented 51% of 2012 revenue, down from 88% in 2007. 

Problem 3: Obsolescence Risk

Each data center is designed at a single point in time with assumptions about future IT infrastructure and applications.  This approach is problematic obviously as technology is constantly evolving.  On the demand side, the need for hardware adaptability, new power density and higher energy efficiency have posed hard questions for legacy data center owners. 

Some of the older centers can be retrofitted by upgrading cooling systems and power infrastructure.  This is expensive and is akin to replacing the battery of an iPhone 4.  Other legacy centers are too old.   They either have to be redeveloped completely or risk rendering useless, just like the iPhone 3GS, which has to be replaced because it is obsolete, even if the hardware is still in function.   As such, I believe any “1^st generation data centers”, without major redevelopment, can only last 20 years or less.

Problem 4: Modular data centers

Modular is an approach to deploy scalable data center capacity using standardized components with multiple power and cooling options.  Modules can be shipped to be added, integrated or retrofitted into the customer’s space according to his/her needs.  They are cheaper than traditional centers.

Source: http://en.wikipedia.org/wiki/Modular_data_center

DFT’s approach is the exact opposite to the modular approach.  It builds centers with huge capacity and equips them with the latest bells and whistles. 

DFT’s approach has several disadvantages relative to modular:

• Bells and whistles are expensive.
• Lacks flexibility to meet clients’ unique needs which makes the center more difficult to lease up.  NJ1 is a good example.
• More vulnerable to the emergence of new center designs’ with better power and capabilities than less “flashy” centers.

At best, modular will compete with traditional data centers and limit rent growth.  At worst, modular will reduce lifespan of the latter to less than 10 years.

Problem 5: True earnings is much lower than AFFO because “depreciation is real”

Investors value REITs based on FFO/AFFO multiples or NAV derived from cap rate.  DFT is no exception.  Management provides the usual REIT metric (FFO, AFFO, NOI etc) in its reports and guidance.

*Source: Company Reports, Bloomberg

FFO is a useful metric for REITs because it adds back GAAP depreciation.  Real estate depreciates at a different rate than GAAP so adding back depreciation makes sense.  Unfortunately, the investor community has incorrectly classified data centers as “normal real estate” and inadvertently make the mistake of accepting FFO/AFFO/Cap Rate as earnings metrics.  At the very least, an adjustment to account for technological obsolescence based on estimated useful life has to be deducted from FFO/AFFO.  The cash capex numbers that were used to calculate AFFO are maintenance capex that keeps the centers running and do not account for true depreciation.

Data centers construction cost per SF often triples other commercial real estate type because of expensive infrastructure spend on power supply, switchgear, cooling system, etc.  

Source: company reports

About 66% of DFT’s capitalized real estate costs are electrical and mechanical components with well-known usable lifespan that rarely exceed 30 years.  For the purpose of this analysis, land, buildings, fire and security system are assumed to last forever.

2012 true depreciation expense should be A * 66% / B = $63mm.

Valuation

A business with limited moat should trade at 15x earnings or less.  Assuming 25 years of useful life for margin of safety, DFT should be worth $0.8 *15 = $12/share.

Catalyst 1: Goldman Sachs has gone modular

http://www.datacenterknowledge.com/archives/2012/09/19/goldman-sachs-goes-modular-with-io/

• Modular is now competing for the wholesale, enterprise clients of traditional data centers

Catalyst 2: DFT went from “modular isn’t a solution” to “we now construct in a modular fashion” in 3 months

<Q - David Shamis>: Okay. Great. And how should we be thinking about the recent announcement by Goldman to move to modular data centers?

…..

<A - Hossein Fateh>: ...if it starts with 250 kilowatts, incrementally goes to 800 megawatts. So, I think there's more noise and fluff than reality there. So, not knowing the detail of that deal, I – long term, I do not feel that modular is a solution. There is absolutely an application for modular data centers. But I don't think there is a solution for large companies that wanted to build – that need their own data centers.

Source: 3Q12 earnings call, Bloomberg

< Hossein Fateh>: As previously stated, we're now able to build out in smaller incremental megawatts instead of the full phase. Scott Davis will discuss this later in the call.

…..

< Scott A. Davis>: Lastly, we have devised greater flexibility for the facility to be constructed in a more modular fashion, allowing for multiple phases of construction in logical building blocks, sized to better match our expected demand.

Source: 4Q12 earnings call, Bloomberg

• This confirms the modular approach to data centers is the future...
• Which means DFT’s existing portfolio lifespan is more likely to be less than 20 years than not, and...
• Rents more likely to roll down than up due to competition from newly designed inventory.

Catalyst 3: The super wholesale lease extension in 3Q12

• Extended existing leases with top 3 tenants by 4 more years, but at rates that are 18% lower than existing rents.  Management used the term “super wholesale rate” to describe the extension in the earnings call.
• I call the extension a “reality check at market rate”.  The fact that DFT has to swallow this tough pill is a testament to the problems ( # 1 - 4) that were mentioned above.

Conclusion

The short thesis is based on my assertion that

1. Depreciation is real and is ignored.
2. Modular is a bigger threat than people realize.
3. DFT’s existing portfolio is out of date and will surprise to the downside.

I will leave you with this excerpt from the 4Q12 earnings call.

The analyst was asking about data center VA3.

<Q>: Thanks guys, I appreciate the time. In terms of VA3, I want to focus on that for just a minute, so we've now seen two tenants leave that space, Yahoo! a few quarters ago and now the tenant this quarter. If I kind of I guess between those tenants there's really been no backfilling of that space. Just looking at it on paper, it was built in 2003……. I don't want to say obsolete, but maybe inferior and what's the road to leasing that up?

So after a few exchanges, management’s final response to the question was

<A - Hossein Fateh>: Yeah, we would need to upgrade all the electrical wires. We would need to upgrade the cooling system. I mean, essentially we would need to build a new building.