Loading audio...
Not All Data Centers Are the Same
With data centers increasingly in the headlines, it can be easy to assume they are all enormous facilities with extraordinary power demands. Coverage of large AI developments can reinforce that impression, but data centers also support the systems hospitals use to care for patients and the applications businesses rely on to operate.
Demand for this infrastructure was growing well before ChatGPT brought generative AI into the public spotlight in 2022. Today’s facilities serve different purposes, from housing equipment for regional businesses to training large AI models. Understanding those differences helps people make sense of the headlines and evaluate what a proposed facility could mean for their community.
The following sections explain the different types of data centers and offer examples of the work they support.
Multi-Tenant Colocation Data Center
Multi-tenant colocation is the model DataBank operates. In this shared environment, multiple organizations house their servers and networking equipment in a facility managed by a colocation provider. The provider supplies the power and cooling systems, along with physical security and digital security frames while customers generally manage their own equipment.
For businesses, colocation offers access to professionally managed infrastructure without the cost and complexity of building a dedicated data center. Customers lease space and power capacity that fit their requirements, with options to expand as their needs grow. This model serves organizations ranging from regional businesses to global enterprises.
Example: A colocation facility might house infrastructure for dozens of customers, including a regional bank running transaction systems and a healthcare organization supporting patient care applications. Each customer manages its equipment in designated space. The provider operates the facility and may offer connectivity services and on-site technical assistance, such as replacing a component.
Edge Data Center
An edge data center places computing infrastructure close to the people or systems it serves. These facilities are typically smaller and help reduce latency, the delay as data travels between locations. They can support applications that need a fast response.
Location is central to this model. Edge facilities can extend a provider’s computing footprint into areas where users would otherwise rely on more distant infrastructure.
Example: An edge facility located near a cell tower can bring computing capacity closer to mobile users. By processing certain workloads nearby instead of sending data to a distant facility, it can help improve response times for applications such as interactive gaming.
Enterprise Data Center
An enterprise data center is a private facility owned and operated by one organization to support its computing needs. It may be located on the company’s campus or at a separate site. These facilities may be less visible to the public when they are part of an existing corporate campus rather than a standalone development.
This model gives a company direct control over its infrastructure, along with responsibility for keeping it operating reliably. An enterprise data center can also be part of a larger technology strategy, with some systems housed internally and other workloads supported through colocation or cloud services.
Example: A large manufacturer might operate a data center on its campus to support engineering applications and production management systems. Its technology team maintains the equipment and plans capacity around the company’s needs, helping keep essential systems available as employees design products and oversee production.
Cloud Data Center
A cloud data center houses the physical infrastructure used to deliver computing services over the internet. Customers access resources such as storage or processing capacity without owning the underlying hardware, typically paying based on usage. Amazon Web Services, Microsoft Azure, and Google Cloud are well-known providers.
“Cloud” describes how customers access computing resources, while “hyperscale” (discussed next) describes the scale of the infrastructure supporting them. A cloud provider may operate large hyperscale campuses or lease space in colocation facilities. Although customers access the services remotely, they still depend on physical equipment housed in data centers.
Example: An online retailer preparing for a holiday shopping surge can use cloud services to increase computing capacity as traffic rises, then reduce it when demand falls. The cloud provider manages the underlying hardware, allowing the retailer to handle seasonal demand without purchasing and maintaining additional servers.
Hyperscale Data Center
A hyperscale data center supports computing operations at very large scale. Companies such as Amazon, Google, Microsoft, and Meta develop these facilities to support services used by millions of people. They may deliver cloud computing to business customers or power the company’s own platforms, such as search engines and social networks.
Hyperscale facilities are designed to accommodate enormous computing demands and expand as those demands grow. A campus may include multiple buildings developed in phases. Their substantial electricity requirements and physical size often attract public attention, although individual projects differ in design and resource use.
Example: A technology company might develop a multi-building campus to support its video streaming platform. As more people use the service, the company can add computing capacity to process and deliver video. The campus provides infrastructure for a global audience, rather than primarily serving businesses in the surrounding region.
AI Data Center
Large AI data center projects have attracted considerable attention, and their size and power demands can shape what people picture when they hear “data center.” Some of the most prominent developments support major technology companies training large language models, or LLMs, that power generative AI tools.
These facilities use specialized processors, often graphics processing units, or GPUs, to perform enormous numbers of calculations. Training a large model can require thousands of processors working together on extensive datasets. Concentrating that equipment creates high electricity demand and considerable heat, often requiring advanced cooling systems such as liquid cooling.
AI facilities also support running trained models, which can have different infrastructure requirements. These workloads may operate in dedicated facilities or colocation environments, so the scale and resource needs of an AI deployment depend on the work it performs.
Example: A major technology company might develop a dedicated AI facility to train its next large language model. Thousands of interconnected GPUs process training data, supported by high-capacity electrical systems and liquid cooling. This computing-intensive work requires a different infrastructure design than a facility primarily supporting everyday business applications.
Looking Beyond the Label
Data centers support everything from patient care to the development of new AI models. The largest projects may dominate the headlines, but they do not represent every facility or the work it performs.
For anyone evaluating a proposed development, the useful questions concern the project itself: What will it support? How much power and water will it require? How will it fit into its surroundings? Looking at those specifics helps communities weigh potential benefits and concerns with a clearer understanding of what is being proposed.
Want to better understand the different types of data centers and the work they support? DataBank Digest brings you analysis, insights, and the latest developments across digital infrastructure, energy, and growth.
