Gravel2Gavel Construction & Real Estate Law Blog
By one count, worldwide there were some 11,800 data centers in early 2024. Within that census are facilities so small that they fit in office building closets, while others are among the largest manmade structures on the planet. How are we to make sense of this diverse population?
Data centers house computers, networks and switches to store, process and disseminate large quantities of data of organizations and their customers and supply chains. The size and complexity of these facilities vary with their functions in the business ecosystem. Much like the well-known depictions of the evolution of horses from tiny brush creatures to mighty stallions, the overall category cries out to be broken down along multiple dimensions.
This post provides a naturalist’s field guide to data center types and features.
Enterprise Data Centers
Enterprise data centers are operated by a single organization, either through direct ownership or a leasing arrangement, to support its internal information technology (IT) operations. These are the traditional and longstanding center types, often running on a private cloud. They have customized infrastructure to meet one company’s needs, giving it full control over security, compliance standards, and system optimization.
Organizations that can afford to operate an enterprise data center benefit from the ability to fine-tune the environment, ensuring governance over sensitive data, and adhering to industry-specific regulations. They pose challenges such as the need for dedicated staffing and robust backup solutions to mitigate downtime, as well as generally high costs not spread across other users.
Enterprise data centers can be either on-premises or off-premises of the organization they serve. On-premises centers offer greater assurance of reliability and security. In contrast, off-premises centers can be a cost-effective alternative for organizations with less stringent security and speed demands, reducing upfront infrastructure and real estate costs, and alleviating the workload for in-house IT staff.
These enterprise centers can be very modest in scale. One private research university, for example, runs a 12,000-square-foot facility—6,000 square feet of data center space and 6,000 square feet of infrastructure space—on 450 KW of power and cooling. The center costs $12.4 million and hosts the institution’s private cloud that supports campus servers.
There are enterprises, and then there are enterprises—how about the NSA? The National Security Agency operates a high security center powered by 65 megawatts of electricity in Camp Williams near Bluffdale, Utah. The $1.5 billion campus encompasses one million square feet with a 100,000-square-foot mission critical data center. The other 900,000 square feet in the campus is used for technical support and administrative space. This center facilitates the NSA’s efforts, as the executive agent for the Office of the Director of National Intelligence, to monitor, strengthen and protect the nation with regard to digital data.
Colocation Data Centers
Since operating an enterprise data center is costly and requires dedicated expertise, many users turn to the colocation data center model. Colocation offers storage and support for computation essentially as a service. These third-party facilities lease space to businesses for their servers, often at a lower initial and potentially reduced operational cost compared to on-premises data centers. The centers host multiple businesses as tenants, each renting the amount of space according to its specific needs—either a full cabinet or often just a fraction of a cabinet.
The centers also supply essential infrastructure, including power, cooling, maintenance services and network connectivity. For organizations lacking around-the-clock expert staff or seeking to minimize upfront investment, colocation can be a practical choice. Additionally, these facilities can offer scalability to meet market demand, enabling businesses to expand their operations as space and resources allow in that specific colocation center.
Colocation can be specifically attractive to e-commerce and financial services companies through faster load times and handling increased traffic and demand due to the site owner’s specialized knowledge of the IT industry. This model can allow businesses to adapt to ever-changing demands in the market and scalability if the site has additional space to lease. Health care providers and governmental bodies can depend on these multi-tenant data centers if they lack the expertise in developing and operating their own enterprise sites.
A composite example of a colocation data center is what we called the Blackacre Data Center in northern California. The site houses two data center structures, a three-story main building consisting of 350,000 square feet and a 15,000-square-foot one-story auxiliary building. For an inside look at exactly what goes on in Blackacre Data Center, see our recent post, Anatomy of a Data Center.
Hyperscale Data Centers
Hyperscale data centers are vast, highly scalable facilities—often exceeding a thousand cabinets, 10,000 servers, and a million square feet. By one 2024 census there were about a thousand centers worldwide in this category, half of them in the United States. Most are owned or operated for the benefit of cloud industry giants, and similar enterprises in China and elsewhere.
These facilities commonly function as enterprise data centers, operated solely by the provider for their own needs, but they can sometimes function as colocation data centers. Regardless of the ownership and operation structure, they are designed to support massive workloads and bring to bear enormous infrastructure capabilities—and enormous infrastructure demands including power and cooling.
Hyperscale facilities are engineered for resilience, featuring robust backup systems and automated failover mechanisms that activate immediately in the event of equipment failure or power loss. While these facilities can be purchased or leased, they are typically custom-built and specifically designed to meet the unique operational needs of the owner and located in rural environments.
It is difficult to convey a sense of the immense scale of these facilities. The world’s largest hyperscaler, a $3 billion project, is located in China’s Inner Mongolia region. This facility consumes 150 MW of power and spans almost 11 million square feet, or the size of 165 regulation U.S. football fields.
A Reno hyperscale is also the single largest colocation data center facility in the world, occupying 1.3 million square feet drawing up to 130 MW of power. It is the first facility to be built within a new data center campus targeting up to 7.2 million square feet of data center space and 650 MW of renewable power.
A hyperscale facility in Mesa, Ariz., occupies 1.3 million square feet, consumes 50 megawatts of renewable power, and cost a reported $2 billion. And another center under development in Louisiana occupies land that would constitute a sizable chunk of Manhattan.
Edge Data Centers
Some cutting-edge applications, such as autonomous vehicles, cloud gaming and instant grocery delivery, require data to be hosted close to users to achieve high-speed network functionality. Edge data centers offer a relatively new solution to these challenges. They are often smaller, decentralized facilities located close to the “edge” of a larger telecommunications and IT network.
The unique location and footprint of these centers allow organizations to reduce latency and optimize bandwidth for new applications. Further, because edge data centers lower network bandwidth usage, they can significantly reduce the cost of data transmission and routing—lessening the required amount of expensive circuits and interconnection hubs. Edge data centers are often deployed in various urban locations, such as at a telecommunications central office or at the base of cell towers throughout a major city.
One services provider recently launched a number of edge data centers designed to extend their infrastructure closer to large urban populations. This venture allows visual effects, video and other applications that require single-digit millisecond latency or local data processing to achieve their goals by bringing infrastructure closer to the end user.
Evolutionary Trends
Data centers are categorized not only by size but also by stability of uptime. A Tier Classification System runs from Tier 1 with 99.671% uptime and a maximum of 28.8 hours of annual outages to Tier 4, for the most mission-critical applications, with 99.995% and only 26.3 minutes of permitted outage. Getting that last bit of assured uptime is very expensive and demands considerable redundancy of all systems.
An alternative for enterprise and edge centers is the modular data center, which can be added or relocated to existing facilities. These prefabricated facilities can finesse issues with real estate acquisition and permitting and allow users to fill interim needs while larger permanent facilities are being developed.
Data centers are running up against economic and land use constraints in their traditional habitats. The increasing size of each center accentuates this challenge; from 2022 to 2024, the average data center parcel grew by 144% to 224 acres. As of October 2024, the weighted average cost of data center land stands at $244,000 per acre. Moreover, the ability of the centers or the local utility grid to support the prodigious power requirements of the data center population is leading to constraints and delays on entitlements and approvals of centers in overheated markets. Data centers are engaged in active management of community, government and public relationships to address the externalities associated with the facilities.
The headwinds in some locations are leading the data center investment and proprietor community to look far beyond the original settings. Thanks to improvements in processing speed and telecommunications capabilities, the largest centers are being situated anywhere in the world that they can find sufficient power, accommodating land acquisition and permitting conditions, and workforces and supply chains suitable for large scale data processing.
Conclusion
Multiple species can thrive in distinct parts of the biosphere. Data centers exist in a diverse range of sizes, locations, operation models, and ownership structures, each tailored to meet a particular set of demands. Selecting the right location, size and model is one of the greatest strategic decisions any organization can face.
RELATED ARTICLES
Something Biden and Trump Can Agree On?—Data Centers, AI and Energy Generation