Why the Bears are Wrong About Equinix
There have been a couple of short reports on data centers, in which they lump Equinix in with the rest of the data centers and paint the industry with broad brushes. The crux of the various reports (though not all are fully public) seems to be:
The cloud companies are competitors and not partners to data centers providers
Data center companies are destroying value as ROIC is lower than WACC, no pricing power, etc.
Growth is slowing
I’ll address Equinix specifically.
Cloud Companies Competitors or Partners?
The notion that the cloud companies are competitors to Equinix assumes that Equinix is the home of the cloud. Said another way, that Equinix builds and operates (or leases) the data centers where the cloud companies process and store data. The notion is that Equinix is building data centers and leasing them to Amazon, Microsoft, Google, etc., in principle is not true. Incidentally, it is true with respect to their joint ventures, but the JVs also help to build the moat around Equinix. If Equinix is not the home of the cloud, what is it? It is the on-ramp to the cloud.
Amazon places its cloud in big data centers (often where land and power are cheap) to execute its processing and store data. Customers don’t rent space and connect to AWS in AWS data centers. They do it via interconnection focused data centers, such as the ones Equinix operates. This is illustrated well below. Notice the Northern Virginia area. It has a white circle and two orange ones. If you could connect directly to AWS, you would not need the two orange circles that are shown on top of the white circle. Just a brief aside, you see all the orange lines leaving the United States? Those are subsea cables. Where do you think those originate and land? Often in Equinix data centers, who serve more than 30 subsea cable providers, including Google’s link to Chile and Los Angeles.
AWS Private Network in North America in 2016 (White is AWS Region, Orange is Edge Location)
The importance of interconnected data centers is illustrated in this analogy. Foxconn is a major manufacturer of Apples’ iPhones. You don’t roll up to a Foxconn factory and buy an iPhone. If you want an iPhone, you go to your local Apple store. The hyperscale data centers (the white dots above) are the Foxconn factories for AWS, while the Apple Stores for them are the orange dots (the interconnected data center). And who owns the orange dots? They are certainly not owned by Amazon, nor Google, nor Microsoft. Those are carrier-neutral data centers that specialize in interconnection. These are owned by the Equinix’s of the world. The below chart (which I created from various peering data basis) shows the importance of interconnected data centers to AWS, Azure, Google Cloud, and other of the worlds leading websites and software providers. Furthermore, if you look at peering databases, you will find that google does not own one facility where you can connect to their cloud directly. Amazon only owns two, and these are incidental examples, as they are both in Bahrain. Check for yourself at https://aws.amazon.com/peering/locations/.
Furthermore, if Equinix was building data centers to house the cloud, remember, these are large single use facilities, you would think that these monster cloud providers would be a massive percentage of Equinix’s revenue. Well, the chart below shows that they are not. The top 10 customers represent 18.2% of revenue, down from 18.6% in Q1 of 2021. Yes, the top 10 list is populated by the large cloud providers, but whom do you think has the leverage here. The largest cloud provider that is only 2.6% of revenue, or Equinix, which provides the cloud provider access to thousands of customers in their data centers? I think Professor Porter would say Equinix, although the proof is in the pricing power.
The first two charts come from the billings for the co-location space, cabinet rental. The number of cabinets is growing steadily, with an occasional big jump from acquisitions. That is where you see the monthly recurring revenue per cabinet drop, so the MRR per cabinet billed is probably not a good same-store-sales metric. Furthermore, MRR per cabinet is also hampered by unfavorable growth accounting. New cabinets are deployed throughout the quarter, so you get revenue from them, but not a full quarter’s worth, while the denominator (the number of cabinets) is burdened by the full number of cabinets.
Price per cross connect is probably a cleaner picture as it is not burdened by the above growth accounting issue, but you do see some declines around large acquisitions. Despite that, this does show you the value and pricing power of the interconnected data center.
How about revenue growth? The most telling way to view this is to look at the normalized growth, meaning ex acquisitions and with constant currency. Equinix consolidated got to a low of 7%. The low was in a period when North America’s growth anemic. What went wrong in North America? Well, nothing went wrong. Sale cycles are particularly long in this industry. Equinix decided to buy Verizon’s data center assets, which weren’t well run. Customers had planned on leaving the Verizon data centers. This is well known and was factored into the purchase price. However, it did take time for that acquisition to be digested and for churn to normalize, which it has. Growth in the Americas in the last two quarters were 10%, with 8% and 7% for the prior two, respectively. It looks like all three regions are starting to fire on all cylinders.
How about the returns for the business? Let’s start with that I think that it is misleading to take the GAAP financials and try to calculate a ROIC from them for two reasons. First, you have a time mismatch, as you deploy capital into a data center, which immediately goes to the denominator of your ROIC calculation, but the numerator takes time to arrive at steady state and have a full year’s worth of earnings. Second, the GAAP deprecation likely overstates the economic depreciation of the asset, more on this later. The preferred way is to look at the cash return on investment for a stabilized data center. This gives you a full year’s worth of earnings of data centers at a stabilized level of utilization and divides it by the total capital put into the facility. Below is their ROIC history. For illustrative purposes, I include the ROIC for expansion data centers, but the variability of the returns for these are likely driven by the utilization rates changing at the various measurement times.
Now let’s tackle the depreciation, as it affects both the ROIC and the valuation. I’ve heard the multiple 100x EPS being thrown around by the bears. Let’s start by saying that data centers have very long lives and require little maintenance in the early stages of their lives. If you can get your hands on MoffettNathanson’s data center initiation of coverage report, they have a nice section on this. In it, they discuss a data center’s maintenance capital as a percent of initial investment totaling around 5% for the first decade of a data center’s life. It ramps up a bit in the teens, and it is followed by a brief respite in the second decade and ends that decade with some heavy spending. They make a point of saying that if you are constantly adding new data centers to your portfolio, then the portfolio average of maintenance capex as a percent of revenue moves up very slowly. Now, this is great, but it ignores the maintenance expenses that aren’t capitalized and just flow through the income statement as repairs and maintenance expenses. Setting that aside, the REIT industry standard earnings metric is AFFO. This metric corrects uses maintenance capex (actual cash spent) in place of GAAP depreciation. When you do this, you get vastly different ROIC calculations and earnings multiples than what is often thrown around by the bears. As an examples Equinix’s 2022E AFFO multiple is 22.3x, not the 100x that was recently mentioned.
I think the bears fundamentally do not understand how the cloud is delivered to customers via interconnected data centers. They believe that it is the cloud players that hold all the cards in the relationship, when in fact it is the interconnected data centers who hold the cards, as they are the choke point between the cloud providers and the customers. Furthermore, the use of GAAP numbers for earnings multiples and returns calculations on the surface tries to appear to be morally pure, while in reality it is a disingenuous attempt to prove their narrative that data centers have low returns and are expensive.
Disclaimer: Views are author’s only. For entertainment purposes only. Not a solicitation nor investment advice. Due your own due diligence.