Where is the cloud? It’s a simple question, but one that relatively few people stop to ask. This is despite the fact that cloud computing plays a key role in essentially every aspect of our modern digital lives.
Contrary to popular belief, the cloud doesn’t just exist out there in the ether. It’s true that end users can access it from almost anywhere, but that doesn’t mean that the cloud itself is everywhere.
In reality, the cloud is made up of interconnected servers and other infrastructure running inside physical data centers. Specifically, cloud providers rely on one particular kind of data center to provide the foundation for their services: hyperscale data centers.
What makes hyperscale data centers different from traditional data centers?
To put it simply, hyperscale data centers are unique because of how truly massive they are. Most hyperscale facilities are 10,000 square feet or more, providing ample space to host thousands of servers. They also need a reliable local grid to provide the energy to keep all those servers running. Companies that use hyperscale data centers may need hundreds of megawatts of power in a single deployment, particularly as AI has changed the landscape.
Running so many servers in one facility inevitably creates a lot of heat. Therefore, hyperscale data centers often require advanced cooling capabilities such as direct-to-chip liquid cooling, particularly to support high-density workloads like AI training.
Compared to other types of data centers, it makes sense that consumers and even most enterprise IT professionals don’t spend much time thinking about hyperscale data centers and why they’re important. After all, the vast majority of organizations never have a reason to use hyperscale data centers—at least not directly.
Who uses hyperscale data centers?
Hyperscale data centers provide colossal amounts of space and power to support the largest workloads from some of the largest digital service providers on the planet. Of course, this includes the cloud providers, but also major SaaS and content providers. This select group of companies makes up essentially the entire global market for hyperscale data centers. They’re commonly referred to as “hyperscalers.”
In some cases, these hyperscalers invest substantial amounts of money to secure land and build data centers for themselves. This can help give them a sense of confidence that they’ll get the exact facility that meets their needs. In other cases, they may decide to tap into the resources and expertise of a global colocation provider instead. Hyperscalers are technologically savvy enough to design and build their own data centers, but it doesn’t make sense for them to do so in every situation.
By working with a colocation partner like Equinix, hyperscalers can access a specialized wholesale service that’s sized to their exact needs. They still take a hands-on role in designing and operating the hyperscale deployment they’re looking for. The only difference is that they’ll let someone else worry about buying the land, managing the construction process, procuring the energy and meeting efficiency and sustainability targets.
Where are hyperscale data centers typically built?
Since hyperscale data centers require so much space and power all in one location, it’s very common to find them in more remote locations, where real estate and energy tend to be affordable compared to major metro areas. In these locations, it’s also easier to find large parcels of land with grid capacity on which to build an interconnected campus with multiple hyperscale facilities.
It’s true that building hyperscale data centers in rural areas could lead to higher latency, since data traffic would need to cover longer distances to reach the facility. However, this typically isn’t an issue, since service providers tend to use hyperscale data centers to support workloads that are less sensitive to latency. As an example, let’s consider how hyperscale data centers enable certain AI workloads.
AI model training and model inference workloads have different infrastructure requirements, and it’s best to use different kinds of data centers to support them. Model training involves processing huge amounts of data to establish pattern recognition and thus requires a lot of compute capacity. On the other hand, training doesn’t involve real-time datasets, so network latency isn’t a major concern. This combination of factors makes hyperscale data centers a perfect fit to host model training workloads, especially for large language models (LLMs) with particularly vast training datasets.
How to make the most of hyperscale data centers
Just because hyperscale data centers tend to be remote doesn’t mean they should be isolated. In fact, ensuring connectivity is essential to get the full value of hyperscale data centers.
To go back to our AI example, model training is only one aspect of the distributed AI workflow. Model inference workloads are smaller and more sensitive to latency, so they’re often hosted in strategically placed colocation data centers that offer closer proximity to end users. It’s essential to have high-performance network infrastructure to move data between training and inference locations on an ongoing basis.
While it’s true that enterprises typically don’t use hyperscale data centers directly, they’re often the intended customers for the services hosted there. For instance, enterprises that use LLMs from a service provider would indirectly benefit from model training that happens in hyperscale data centers.
Enterprises rarely connect to services from inside the hyperscale data center. Instead, the service provider sets up a network extension from the hyperscale facility to wherever the enterprise customer is located. Building a network node inside a nearby colocation data center is a quick and easy way to do this, since there’s a good chance that many of the enterprise customers are already deployed there.
This is why service providers can get the most value from acquiring hyperscale services on a global platform that also includes colocation and digital infrastructure services. For instance, Equinix offers a comprehensive platform to meet the infrastructure needs of large service providers and their customers:
- Equinix IBX® colocation data centers are available in 70+ global markets worldwide.
- Equinix xScale® hyperscale data centers are available in proximity to many of these colocation markets. In addition, we recently announced a joint venture that aims to raise more than $15 billion in capital to fund new xScale data centers in the United States.
- Our portfolio of interconnection services includes Equinix Fabric®, which makes it easy for service providers to connect their distributed infrastructure and reach customers across the globe.
To learn more, read the IDC Vendor Profile: Equinix Experiences Strong Growth Driven by AI, Hyperscale, and Digital Infrastructure.[1]
[1] Courtney Munroe and Avinash Naga, Equinix Experiences Strong Growth Driven by AI, Hyperscale, and Digital Infrastructure, IDC Vendor Profile, May 2024, IDC #US50186623.
Subscribe to the Equinix Blog