Data centers (of all sizes) can track their origins back to the important concept of virtualization. Virtualization uses software to create an abstraction layer over computer hardware that enables the division of a single computer’s hardware components into multiple virtual machines (VMs). Each VM runs its own OS and behaves as an independent computer, even though it’s running on just a portion of the actual underlying computer hardware. In this way, virtualization enables cloud computing.

A hyperscale data center differs primarily from traditional data centers by virtue of its sheer size. A hyperscale data center requires a physical site large enough to house all associated equipment—including at least 5,000 servers and quite possibly miles of connection equipment. As such, hyperscale data centers can easily encompass millions of square feet of space.

Redundancy is another important aspect of hyperscale data centers, and it simply means to provide backup measures and/or devices that can kick in automatically should equipment fail or power be lost. Redundancy is especially critical for hyperscale data centers because these systems are often running automatically—in the background, around the clock and with little direct supervision.

An organization wishing to operate a hyperscale data center will have no shortage of decisions to make as it determines the best course of action to follow. The first of these questions may be: “Should we build or rent?” Building even a modest data center is going to entail some level of investment. Constructing a hyperscale data center requires a financial commitment that’s even more serious.

Many companies opt to go another way by choosing a colocation data center—a data center whose owners rent out facilities and server space to other businesses. The appeal of this method should be immediately apparent: Renting out space for hardware demands a significantly smaller investment than building an entire structure to house equipment, at least in terms of up-front monies paid.

Comparing the two basic options, it’s clear each has its advantages and disadvantages. Building a hyperscale data center is typically expensive and labor-intensive, but it also offers hyperscale facilities that are custom-built for that company, with all of its adjustable aspects suitably optimized.

Meanwhile, renting out space in a colocation data center offers more options for mobility and requires infinitely less investment. But it’s going to be very unlikely that a colocated data center will have been designed to that client’s ideal specifications.

 Some organizations are seeking yet another option—one that assures them that as they continue to move forward, they will be able to scale upward without having to purchase additional and costly storage equipment for their private-cloud-based system. For many of these companies, the right answer involves migrating away from a privately owned system and relocating their operations to a public cloud environment, such as they find with Software-as-a-Service (SaaS)  apps like Microsoft 365 or Google Suite.

There are other variations like modular data centers, which are pre-engineered facilities that are designed for use as data centers. Not only are modular data centers pre-engineered, they’re also pre-piped and equipped with necessary cooling equipment. Modular data centers are ideal for organizations that want to experiment with data center functionality in a limited way before making huge investments, as well as for those companies that need to implement a reliable data center solution quickly.

Data centers have been used since the 1940s, back when a single, mammoth computer filled entire department spaces. As computers became more size-efficient through the years, the on-premises physical space allotted them also evolved. Then, in the 1990s, the first burst of microcomputers was seen, radically reducing the amount of size needed for IT operations. It wasn’t long before “server rooms” were being referred to instead as “data centers.”

In terms of important advancements, the first hyperscale data center is often considered to have been launched in 2006 by Google in The Dalles, OR (near Portland). This hyperscale facility currently occupies 1.3 million square feet of space and employs a staff of approximately 200 data center operators. The term “currently” is used because Google is now pursuing plans to scale out this “energy campus” with a USD 600 million expansion project that will add a fifth building of 290,000 square feet. Showing its further commitment to this technology, Google now operates 14 data center facilities in the US—as well as 6 in Europe, 3 in Asia and 1 in Chile.

At present, the world’s largest hyperscale facilities (link resides outside ibm.com) exist within the Inner Mongolia region of China. That’s where China Telecom operates a hyperscale data center that’s roughly 10.7 million square feet. Put another way, the largest data center is the size of 165 regulation US football fields—all conjoined in a space 11 football fields wide and 15 football fields long. Not surprisingly, this humongous hyperscale data center (which cost USD 3 billion to construct) is outfitted with everything it needs functionally and even boasts residence facilities for the hyperscale data center operators who work there.

Ranking the leading hyperscale companies can be problematic. First, there’s the relatively easy issue of defining what criteria will be used. Theoretically, you could focus on the number of hyperscale data centers owned or built, but that’s not an entirely accurate way to judge a company’s delivery of cloud services, since numerous outfits (including Apple) lease some of the data centers they use from other service providers.

That leaves only one viable means of comparison: percent of market share. This is ultimately the best indicator of which hyperscale companies are truly driving the market. Even if this method is slightly imperfect because the hyperscale computing market is never static—it’s always in a state of flux.

However, there is ample trend evidence to suggest that an upper tier of three hyperscale providers has currently established a solid hold on the largest segment of this market:

Presently, AWS is the largest provider of hyperscale cloud services, with a commanding AWS market share (link resides outside ibm.com) of roughly 32%. AWS operates in 32 cloud regions and 102 availability zones, with a total space of 33.5 million square feet. AWS is known for its expertise in automation, database management and data analytics.

Microsoft’s popular hyperscale platform Azure currently garners approximately 23% (link resides outside ibm.com) of the hyperscale market. Azure operates in 62 cloud regions, with 120 availability zones. Not surprisingly, Microsoft Azure works especially well in conjunction with Microsoft software for enterprise data centers.

Known originally and primarily for its vast expertise in data handling, Google Cloud controls roughly 10% (link resides outside ibm.com) of the hyperscale market and operates in 39 cloud regions, with 118 availability zones. In addition to data handling and data processing, GCP attracts businesses for its strengths in artificial intelligence (AI) and advanced analytics.

• Alibaba Cloud: Although not as large as AWS or other top providers, Alibaba does possess a strong hyperscale market share within the Asia-Pacific region. Its notable offerings include infrastructure-related products and AI services.

• Apple: Apple uses a hybrid model for its cloud services. Apple owns eight data centers in the US, Europe and China, with plans to build more. In addition, Apple has multi-year lease arrangements for cloud-computing services from providers like AWS and GCP.

• IBM Cloud: Long synonymous with technology expansion, IBM Cloud has deep experience working with enterprise data centers and delivering services in a wide range of related areas. Lately, the company’s pioneering work with AI is bringing it renewed visibility.

• Meta Platforms: The parent company behind popular online platforms like Facebook and Instagram, Meta Platforms operates 21 hyperscale data centers around the world whose total space exceeds 50 million square feet.

• Oracle Cloud Infrastructure (OCI): OCI has positioned itself as a lower-cost alternative to AWS, offering similar services at a significantly reduced price point. OCI excels at enabling the creation of cloud-native apps and easy migration of mission-critical workloads.

Power usage is the most pressing issue surrounding hyperscale data centers. The massive computing power needed to operate hyperscale data centers comes from a massive amount of electric power. Like the rise of cryptocurrencies and bitcoin mining, hyperscale data centers are a fairly recent technological development whose unusually large appetite for power places them at least somewhat at odds with ecological sustainability goals (although some companies are finding ways to mitigate or even eliminate such problems).

In addition to the thousands of computer servers being run in constant operation, there are also the electrical needs of power and networking equipment (such as transformers and routers) to consider, in addition to critically important cooling systems that keep hardware from reaching thermal overload situations. And that doesn’t even begin to address the costs of constructing and running an enormous structure to contain this beehive of activity.

That’s why energy efficiency is so crucial to effectively running hyperscale data centers. If one server is not working at peak efficiency, that might be considered of negligible importance. But if hyperscale facilities contain thousands of servers that aren’t working efficiently, that’s going to be considered a much larger and more expensive problem.

Depending on their size, hyperscale data centers may require megawatts of power or even gigawatts. There’s so much variance among hyperscale data centers that nailing down an average energy usage amount can be difficult. These guidelines of different-sized data centers are not official definitions, but have been unofficially ratified over time, and can assist our general understanding of these facilities:

• Micro data center: The smallest recognized data center is generally used by single companies or for remote offices. Micro data centers usually exhibit a capacity of 10 server racks or less, which works out to a total capacity of approximately 140 servers. Micro data centers typically occupy less than 5,000 square feet of space. Energy draw: Under 100-150kW.

• Small data center: Small data centers typically require between 5,000–20,000 square feet of space and may host anywhere from 500 to 2,000 servers. Energy draw: 1–5MW.

• Average data center: The average onsite data center typically has between 2,000 and 5,000 servers. Likewise, its square footage could vary from between 20,000 square feet and 100,000 square feet. Energy draw: Around 100MW.

• Hyperscale data centers: The IDC definition of a hyperscaler (link resides outside ibm.com) holds that to be considered a true hyperscale data center, the facility should contain at least 5,000 servers and occupy at least 10,000 square feet of physical space. Energy draw: Over 100MW.

Energy is more expensive depending on its location, and companies often look to undeveloped/underdeveloped countries or regional areas as potential building sites for their hyperscale data centers because the pricing of electricity is more attractive within those economies.

But that’s only one criterion to be considered. Finding an area with an affordable energy source is key, as is observing local sustainability. But so is locating a site that likely won’t be subject to constant and hazardous weather situations that could imperil a company’s mission-critical functions through power outages and downtime.

Companies are trying to balance their corporate initiatives with sustainability goals so that hyperscale data centers are affordable to run and leave as light a carbon footprint as possible. The drive to reduce power usage is even pushing companies to push for renewable energy solutions for powering their hyperscale data centers. Most of the major businesses involved in hyperscale computing have investigated or are now exploring renewable energy options like solar power and wind power as a way to offset their considerable power usage.

In some cases, cloud service providers have even pledged to ensure that their data centers offer complete sustainability. Most impressive is Apple (link resides outside ibm.com), which has supported sustainability with actual constructive action since 2014. That’s when the company mandated that all of its data centers become powered completely by renewable energy. For a decade now, Apple’s data centers—all of them—have been run on various combinations of biogas fuel cells, hydropower, solar power and wind power.