Cyxtera Technologies: A Data Center Business Built on the Wrong Real Estate

Welcome to the 193rd Pari Passu newsletter.

Cyxtera Technologies was the product of a 2017 carve-out acquired by BC Partners and Medina Capital and assembled into the third-largest public retail colocation operator in the US. What began as a portfolio of data centers was repositioned as an independent enterprise infrastructure platform serving over 2,300 customers across roughly 60 facilities. By 2021, the business had gone public via a SPAC merger and had grown steadily amid the data center supercycle driven by AI demand. Unlike most competitors, Cyxtera was built on leased rather than owned real estate. Then, in the middle of the largest data center buildout in history, it filed for Chapter 11. The business was not the problem. The capital structure behind it was. With its entire funded debt floating-rate and almost all of its real estate leased, rising rates and a near-term maturity wall left the company unable to refinance.  

In June 2023, Cyxtera filed for Chapter 11, which culminated in the sale of its assets to Brookfield Infrastructure Partners. What makes this transaction interesting is not the distress itself but the company’s post-filing performance. Alongside the asset purchase, Brookfield directly bought the underlying real estate at seven of Cyxtera’s most important data centers from their landlords, converting the business from a tenant to an owner-operator in a way the standalone recapitalization path could not have delivered. The combined platform, now rebranded as Csquare, has since doubled EBITDA and filed confidentially for an IPO in the coming weeks. 

The data center industry sits at the center of the AI buildout, and Cyxtera is a useful lens on how it works. In today’s writeup, we will start by walking through Cyxtera’s business model and why retail colocation occupies a particular niche in the data center market. From there, we will cover the corporate history, including the BC Partners carve-out, the de-SPAC, and the buildout of the leased facility footprint. After discussing the path to distress and the finance lease structure that masked the business’s true economic leverage, we will break down the company’s dual-track bankruptcy and its post-emergence transformation into Csquare.

Cyxtera Technologies primarily operates in the retail colocation segment of the data center industry. Colocation itself is the business of renting physical space, power, cooling, and security to house other companies’ servers. To understand retail colocation, it helps to separate the two ends of the data center market. At one end sits wholesale, where operators such as Digital Realty lease enormous blocks of space and power to a handful of tenants, including the major cloud providers (e.g., AWS and Microsoft Azure). At the other end sits retail colocation, where a provider such as Cyxtera rents out smaller increments of space, anywhere from a single cabinet to a few rows, to a large number of enterprise customers, such as a regional bank or a healthcare provider, who want to house their own servers in a secure, well-connected facility without building one themselves. If wholesale is renting out an entire warehouse to one or two tenants, retail colocation is renting individual storage units within it to many. 

Operators like Cyxtera provide everything around them: the building, the power, the cooling, and the security and uptime guarantees. For a mid-sized enterprise, it is far cheaper and faster than building and staffing its own data center. That decision reflects a broader choice about where a company runs its computing. It can rent capacity on demand from a public cloud provider like AWS, Azure, or Google, which owns the servers, or run its own servers in a facility it controls, such as a colocation site like Cyxtera’s. Many do both, an arrangement the industry calls "hybrid" setups. 

The retail colocation model is attractive due to its revenue stability. Cyxtera’s colocation contracts (~83% of 2022 revenue) typically run on fixed three-year terms billed monthly, and roughly 90% of revenue is recurring, reinforced by high switching costs [1]. Once a customer has physically installed servers, networking gear, and storage inside a Cyxtera facility and integrated those systems into broader IT operations, moving to a competing provider means physically relocating hardware, retooling connectivity, and risking downtime on workloads that often cannot tolerate it. The result is an average monthly churn rate of 0.8%, an industry-leading low [2]. Customer concentration is modest, with ~32% of monthly recurring revenue coming from the top twenty customers. 

This model carries high operating leverage. The cost of running a data center, the lease payments, the staffing, and the baseline power and cooling are largely fixed. That means once a facility is built and the lights are on, each incremental customer that fills empty cabinet space flows through to EBITDA at very high margins. However, operating leverage cuts both ways. When revenue softens, the fixed cost base does not shrink with it, and the same leverage that expands margins on the up compresses cash on the way down. As we will see, this dynamic sits at the center of Cyxtera’s distress. 

Colocation is also capital-intensive in two places: the building and the equipment inside it. Cyxtera leased its buildings, so it paid rent rather than capex. On the other hand, it still had to buy and install the power, cooling, and the network fabric that brought the new space online. This is where its capex went, about 13% of revenue from 2018 to 2022. Because the buildings ran through rent, that figure understated the true cost of Cyxtera’s real estate. An owner would have capitalized the buildings and reported higher capex while growing. 

The Interconnection System

If colocation is the foundation of Cyxtera’s business, interconnection is what makes Cyxtera’s facilities valuable rather than interchangeable. Customers in a data center don’t just sit next to each other, storing servers. Many of them need to talk to each other and the outside world as quickly and securely as possible. Interconnection is the physical and virtual cabling that links one customer directly to other customers in the same building, to outside networks, and to cloud providers without sending traffic over the public internet. Here is why that matters. Let’s say a bank housed in a Cyxtera facility wants a direct, private line to a payments processor three rows away, and a second one running straight into Amazon’s cloud. Over the open internet, traffic is slower and less secure. A direct connection inside the building, called a cross-connect, is faster, more private, and more reliable. Across its facilities, Cyxtera houses more than 40,000 of these direct links, connecting customers to over 300 internet carriers, such as Verizon, and to major cloud platforms such as AWS, Google, and Microsoft [3]. 

Interconnection is sold as its own product. Customers pay a monthly recurring fee for each cross-connect, and the segment is extremely profitable. Although interconnection accounts for only ~11% of revenue, its margins exceed 95% since once cabling is in place, each additional connection costs almost nothing to provide. Additionally, the more connections a customer builds within a facility, the more its day-to-day operations depend on being there, and the more disruptive leaving becomes [4]. Therefore, interconnection is Cyxtera’s highest-margin product and strongest retention tool, and the closest thing a colocation operator has to a real moat. The question was how to widen it, and Cyxtera’s answer was automation. 

Digital Exchange

What set Cyxtera apart from legacy colocation owners was its focus on making interconnection programmable.  The flagship product was the Cyxtera Digital Exchange, an in-house network that customers controlled through software rather than through physical work. In simple terms, this lets customers set up new network connections on demand through a web portal or programming interface, rather than waiting for a technician to run a cable by hand. By treating the physical data center like software, Cyxtera offered cloud-like speed with the control and security of dedicated infrastructure. 

The company extended this idea with two adjacent products. Enterprise Bare Metal lets customers rent dedicated, high-performance servers inside a Cyxtera facility on a monthly basis. A company running AI training jobs, for instance, could use the Bare Metal platform to get the latest NVIDIA chips without buying them outright or signing a multi-year lease. It is the same idea CoreWeave has since built into an entire business, renting out dedicated GPU capacity for AI at scale, though for Cyxtera, Bare Metal was a minor product rather than the company itself. The second product, SmartCabs, applied the same on-demand idea to standard colocation. A SmartCab was a colocation cabinet that came with built-in power and the Digital Exchange’s configurable network fabric already in place, so a customer running its own servers, as opposed to leasing capacity as in the Bare Metal on CoreWeave model, could begin running in days rather than waiting weeks for an installation crew. The two products solved opposite problems. Bare Metal was for customers who did not want to own hardware, while SmartCabs was for customers who did want to own hardware but wanted it deployed quickly. In revenue terms, both were small. Bare Metal ran at roughly 1% of revenue, while SmartCabs were reported within core colocation rather than as a separate line [5]. 

To summarize, these three layers reinforce one another: colocation gets the customer in the door, interconnection makes leaving expensive, and the Digital Exchange lets a customer run dedicated infrastructure with the flexibility of the public cloud. Together, these three layers kept churn low and customers in place, right up until the balance sheet, not the business, forced the issue.

Market Backdrop

The backdrop to all of this is a data center supercycle driven by AI. Training and running large AI models takes enormous amounts of computing, far more than traditional software, and that computing has to live in a physical data center somewhere. That demand is pulling data center construction to record levels. In addition, traditional business computing ran at roughly 5–15 kilowatts per rack. Modern AI and high-performance workloads are increasingly demanding 50 kilowatts per rack or more, with some deployments exceeding 100 kilowatts, which, for context, is enough to power 60–80+ homes [6]. The shift matters because data centers were not designed around this kind of density. Cooling, power distribution, and floor layout all assumed the older 5-15 kilowatts range, which means an enormous amount of existing capacity cannot host modern AI workloads without major upgrades. The result is a structural imbalance between what AI workloads require and what the installed base can deliver. 

At the same time, enterprises are rethinking where their computing should live. Throughout the 2010s, the default answer was the public cloud, with everything moving to AWS, Azure, or Google Cloud. By the 2020s, that view had softened. Cloud bills grew faster than expected, some workloads ran more cheaply on dedicated hardware, and governments increasingly required certain data to reside on customer-controlled infrastructure. The result was a shift back toward hybrid, with more workloads moving off the public cloud and onto a company’s own servers, often housed in a colocation facility [7]. Retail colocation sits at the center of that shift because a colocation facility is where the customer’s private servers can sit just one cable away from the major cloud platforms, plugging straight into AWS or Azure without routing over the public internet.  For Cyxtera, the move back to hybrid kept demand steady through 2021 and 2022, as enterprises kept servers in colocation running alongside the public cloud. 

On the supply side, the binding constraint became electricity. By 2022 and 2023, building a new data center required the local utility to extend the grid to the site and approve the load. As demand exploded, utilities in major markets like Chicago, Silicon Valley, and Phoenix could not keep up. Wait times for new grid connections in those markets stretched well past three years. The result was that pricing power shifted to operators that already controlled facilities with secured power, because new competitors could not come online quickly to undercut them. 

Against that backdrop, Cyxtera sat as the third-largest public retail colocation provider, ranking behind two much larger rivals. Equinix had built the largest global interconnection ecosystem in the industry, which was difficult for customers to abandon or replicate elsewhere. Digital Realty worked on the other end of the market, leasing vast capacity to the biggest cloud companies. Cyxtera could not match Equinix’s connections or Digital Realty’s scale, so it competed on speed. By running its data centers like cloud services, it let customers turn up capacity through a web portal in minutes rather than weeks, which suited enterprises that wanted to retain control of their own hardware without the slow deployment that normally came with it [3]. None of which was the issue. What undid Cyxtera was its business model around the company’s controlled data center capacity.

The Sandwich Lease

Most large data center operators are structured like real estate investment trusts (REITs) and own the buildings they operate. Cyxtera had a different model. In all but two of its global locations, it leased the space that housed its data centers, renting large blocks of wholesale capacity from landlords (in some cases, direct competitors such as Digital Realty), subdividing and reselling that space to its own retail customers. 

The industry calls this a “sandwich lease” because the operator sits in the middle, paying rent on one side and charging customers on the other, profiting on the spread [8]. 

The case for the leasing structure is capital efficiency. Owning roughly 60 data centers worldwide would have cost billions upfront. Leasing lets Cyxtera reach a global footprint on a fraction of that upfront capital, while also giving the company a chance to step back from a market by letting a lease lapse at expiry. In a stable rate environment, the spread between wholesale rent and retail pricing works well. The other side of that capital efficiency was a large, fixed lease expense, owed under contract, whether or not the business performed. 

However, because Cyxtera did not own its real estate, it effectively paid its landlords’ margin on top of its own costs, which left it at a structural price disadvantage relative to rivals that built and owned their facilities. In many markets, customers chose Cyxtera for its interconnection density, deployment speed, and service rather than price. Still, the model meant it competed from a higher fixed-cost base than a rival that owned its own real estate [9]. As mentioned above, its reported capex essentially does not capture the true cost of its real estate, which sat off the balance sheet as leases rather than owned assets. 

In fact, Cyxtera’s lease commitments were enormous. By the end of Q1 2023, Cyxtera carried roughly $1.4bn in total future lease payments across its finance and operating leases, most of which were long-term and non-cancellable [10]. The trouble was that these leases were fixed. When business was good, the rent was cheap relative to the revenue it supported, and earnings rose quickly. When revenue softened, the rent did not, because the company had to pay under contract regardless of what happened to revenue. 

Cyxtera is a relatively new company, and the lease-heavy structure that defined it was inherited rather than chosen. To see how, we start with the business’s origins. Its operations began inside CenturyLink, a large US telecommunications carrier, now known as Lumen. CenturyLink had accumulated a portfolio of high-quality, well-located facilities, but ran them as a small side business. Its focus was on its core telecom business, and the colocation operation was set aside and under-managed. CenturyLink did not own much of this real estate, so the data centers carried sizable lease obligations [11]. 

In 2017, private equity firms BC Partners and Medina Capital bought the portfolio, spanning 57 data centers and roughly 195 megawatts of power capacity at the time, out of CenturyLink for roughly $2.2bn, with Cyxtera assuming about $568mm of capital lease obligations and other liabilities at close [12]. The thesis was that a dedicated management team could turn those underutilized assets into a retail colocation platform that any network could plug into, not just one. Under CenturyLink, the facilities ran mainly on CenturyLink’s own network, so customers were largely stuck with it for connectivity. Opening the sites to every major carrier lets customers pick whichever providers they want. The acquisition closed on May 1, 2017, and, combined with Medina Capital’s security and data analytics operations, Cyxtera was born. Lumen retained a roughly 10% equity stake. 

To finance the carve-out, the company put in place a $1.3bn first lien credit package: a $150mm RCF, a $815mm first lien term loan due 2024, and a $310mm second lien term loan due 2025 [1]. Against the roughly $2.2bn purchase price, that implies an equity contribution of a little over $1bn and a loan-to-value in the low 50s, a reasonable structure for a carve-out of stable, contracted infrastructure assets [13]. On a standalone basis, the divested colocation business generated an estimated $137mm in EBITDA, implying a ~16x entry multiple.

Over the next several years, the company repeatedly amended and expanded the facilities, borrowing a $100mm incremental first-lien term loan in May 2019 [14]. Notably, Cyxtera’s funded debt was entirely floating-rate. Cyxtera’s original portfolio also included a cybersecurity software division assembled through acquisitions. That fit proved awkward in practice, because selling cybersecurity software and selling data center space and power involve fundamentally different sales cycles and buyers. On December 31, 2019, Cyxtera spun off its cybersecurity business as a standalone company, Appgate, which allowed Cyxtera to focus exclusively on retail colocation while maintaining a commercial relationship with Appgate for its own security needs [3]. 

The company’s path to public markets ran through a SPAC. Starboard Value Acquisition Corp., a special purpose acquisition company sponsored by Starboard Value, the activist hedge fund, was incorporated in November 2019 and completed its IPO on Nasdaq in September 2020. On July 29, 2021, Starboard completed its business combination with Cyxtera in a deal that valued the combined enterprise at roughly $3.4bn, and the surviving public company changed its name to Cyxtera Technologies, listing under the ticker CYXT [15]. The de-SPAC delivered approximately $654mm in proceeds, including a $250mm investment from Fidelity and Starboard clients. Cyxtera used the proceeds to repay the entire $310mm second-lien term loan, among other corporate purposes, leaving the first-lien term loans and revolver as the company’s remaining funded debt [1]. As expected, Cyxtera’s ~16x EBITDA multiple was priced at a discount to its peers at the time (Equinix ~31x, Digital Realty ~26x). 

From the 2017 carve-out to the de-SPAC,  the business grew steadily on cloud adoption rather than the AI demand that would come later. The growth was modest but unspectacular: capacity rose from the ~195 megawatts across 57 facilities it inherited from CenturyLink to roughly 250 megawatts across more than 60 data centers by 2023, with revenue climbing from $703mm in 2018 to $746mm in 2022 and adjusted EBITDA from roughly $156mm to $239mm [16].

Cyxtera entered 2022 with momentum, trading at a multiple predicated on continued growth. On paper, the business was still growing. The problem was never Cyxtera’s operations. It was the capital structure underneath them. That structure worked only as long as capital stayed cheap. When the cost of capital turned in 2022, the structure, not the business, drove Cyxtera’s distress.