Analysts defend Equinix as investigation into short report launched

The report accused Equinix of manipulating its AFFO and artificially boosting its profitability, overselling power at its facilities and not being able to capitalise on the growth opportunities presented by AI.

For a summary and details of the accusations made in the report, read here first.

Shortly after the release of the report, Equinix received a subpoena from the US Attorney's Office for the Northern District of California, the company revealed yesterday.

“Receipt of these types of inquiries is not unusual in these circumstances, and Equinix intends to fully cooperate in this matter. The company does not expect to comment further on such matters until appropriate to do so,” it said.

BMO Capital Markets was quick to reaffirm its outperform rating on Equinix. On Thursday last week, analysts Ari Klein and John Kin wrote in a note to clients they “disagree with the rehashed cloud risk thesis, and while Equinix was perhaps aggressive on its recurring capex definition, it does not seem unreasonable, with the potential risk that valuation gets slightly more expensive."

Klein and Kin said the issue of overselling power was "the most concerning" issue "and where we lack visibility."

Speaking with Capacity, Daniel Golding, CTO and an advisor to several data centres at Appleby Strategy Group thinks there’s an obvious explanation for Klein and Kin’s concerns.

“It’s to do with the nature of Equinix’s business model,” he says.

“They’re a retail colocation provider. They sell power up to a flat rate and allow tenants to draw up to the limit. But on average, the power they’re using isn’t at the limit.

"Not overselling power would mean that Equinix facilities were running at something like 40-50% of their available power most of the time. What investor would be happy to hear that?”

This is in stark contrast to Digital Realty, which Hindenburg often compares Equinix to in its short report.

“Digital Realty has a different model, they sell capacity on triple net leases, if they have 30MW of power, they might sell it in six 5MW chunks. Then you can’t oversell. But Equinix might have 100s of customers who are not going to reach their cap. They are entirely different scenarios.”

“A competent data centre operator can use its experience and appropriate modelling to determine the degree to which it can oversell without jeopardising its ability to deliver for each customer. To our knowledge, Equinix has never faced any sort of notable issues stemming from doing so,” Nick Del Deo, a senior analyst and managing director at MoffettNathanson wrote in a report to investors.

“I cannot think of a situation in the last 20 years, where an Equinix data centre went offline because of oversubscribing power, it simply hasn’t happened,” agreed Golding.

This seemingly nullifies Hindenburg’s claims that overselling power poses an operational risk to Equinix. But does it mean that its other claim that lack of available power means Equinix is ill-positioned to capitalise on AI infrastructure investment in its data centres?

“No one is considering putting Nvidia GPUs in a legacy Equinix data centre” Golding firmly states.

“Equinix wouldn’t even want them to. The way they will capitalise on AI is the same way they’ve cashed in on the cloud, interconnections. This is a more lucrative business on revenue per mw anyway.”

“Equinix has multiple facilities in virtually all its markets, forming connected data centre campuses,” Del Deo writes.

“The facilities vary in age and power density. Equinix can direct customers into facilities that suit their requirements while optimising its power utilisation across its portfolio.”

“For example, networking deployments, which are not power-intensive (and are the core of Equinix’s business) can be slotted into facilities with lower power densities while high-performance workloads can be slotted into newer facilities, and they can all be connected to one another with campus or metro cross-connects”.

“We’d note that Equinix need not directly house power-hungry AI deployments to benefit from AI, just as Equinix doesn’t directly host large footprint cloud compute nodes in any meaningful way to benefit from cloud growth,” Del Deo continues.

A large portion of Hindenburg’s short report was dedicated to dressing down Equinix for misrepresenting maintenance capex as growth capex.

“When Equinix has the option to do a maintenance operation they often go for full replacements. At older facilities, when things break down and need to be replaced, there’s normally a more efficient, or more effective piece of technology that can fill the gap, so they can reduce their power usage effectiveness,” Golding says.

Even if Equinix has pushed the boundaries of what should be classified as maintenance vs growth capex, neither Golding nor Del Deo believe that the grey area is large enough to skew the numbers too significantly.

The majority of Equinix’s growth capex will be spent on building new facilities, which correlates with opportunities to grow revenue.

Even if an adjustment is published Del Deo and his team suspect that the numbers would be modest.

That some maintenance capex gets funnelled into growth capex through liberal interpretations is well understood by Equinix’s institutional investors, Golding and Del Deo argue.

“Maintenance capex figures reported by Equinix and others are clearly not representative of the capital that would actually be required to sustain their businesses. We have yet to meet an investor or observer who takes these metrics at face value,” Del Deo writes.

Ultimately the argument made by Hindenburg is that the figures released by Equinix are misleading investors, whereas the analysts say that the accounting definitions and the definitions that would be used to assess the value of the company are different.

“Industry-wide definitions of recurring vs. non-recurring capex are not necessarily consistent with how an investor would think of those splits for evaluating business performance or valuation,” Del Deo says.