More Detail and Sources:

Hyperscale AI Data Centers

As the name suggests, Hyperscale Data Centers are real enterprise Data Centers. They are vast facilities, often operated by hyperscale customers, usually tech giants like Amazon, Microsoft, and Google. They can contain thousands to millions of servers and other IT resources, making them ideal for massive workloads that require high-speed connectivity and extensive processing power.

Hyperscale Data Centers are designed for horizontal scalability to accommodate increased demand. This scalability makes them highly flexible and efficient, as resources can be added or removed dynamically based on the workload. These data centers are built to grow fast. If the company needs more power or storage, they can add it without shutting anything down. That’s why they’re called “hyperscale”—they can scale up or down very quickly.

The global market of Hyperscale Data Centers is expected to grow at a CAGR of 16.7%, from $35.72 billion in year 2022 to $41.69 billion in year 2023. This demonstrates the rapidly growing demand for high-capacity data centers capable of supporting extensive computational workloads, such as big data analysis, cloud services, and machine learning algorithms.

Exploring the differences 4 types of Data Centers source: 

Hyperscale Data Centers represent data storage on a gigantic scale. According to the Independent Data Council (IDC) definition of a hyperscale database, as reported by VIAVI Solutions (link resides outside ibm.com), to be considered a true Hyperscale Data Center, it must contain at least 5,000 servers and occupy at least 10,000 square feet of physical space. There is no hard-and-fast guideline concerning energy usage, although most Hyperscale Data Centers use somewhere between 100 megawatts (MWs) and 300 MWs.

AI Data Centers

Built specifically for AI and machine learning (ML) applications, these facilities are optimized for processing large volumes of data, running deep learning models and supporting AI-driven tasks like natural language processing and image recognition, among others.

They rely heavily on GPUs (Graphics Processing Units), TPUs (Tensor Processing Units) and other specialized AI accelerators. GPUs can process thousands of tasks simultaneously, making them ideal for AI training and inference. The growth of artificial intelligence, which uses larger and more complex chips needing far more power, only accelerates the power demand.

AI computations generate significantly more heat due to the high processing power of GPUs. As a result, AI Data Centers require advanced cooling solutions, such as liquid cooling to maintain efficiency and prevent overheating.

Components that go into building a Hyperscale Data Center

• A tract of land that can support a structure of at least 10,000 square feet.
• Development costs to survey the land, clear the site and prepare it for construction.
• One structure with at least 10,000 square feet of floor space. It should be built sturdily enough to support normal operations as well as withstand  normal building “wear and tear,” plus any relevant local conditions, such as extreme weather situations or even geological occurrences.
• Fire-safety  equipment that is sufficient for protecting the building and its operators, such as sprinkler systems and extra extinguishers, as well as the development and instruction of fire-safety Standard Operating Procedures (SOPs).
• A parking lot of sufficient size to support the number of vehicles used by data center operators, as well as ample parking room for any emergency vehicles that might be needed.
• Cooling  system equipment to offset the immense heat generated by 5,000 servers operating 24 hours per day.
• Specialized water lines and piping to support the cooling system.
• Dedicated and reinforced power lines and equipment to safely handle massive electrical loads.
• Back-up power systems in case of mainline outages.
• At least 5,000 servers.
• Metal (or wooden) racks for housing 5,000 servers.
• Associated IT equipment.
• Reinforced cabling to connect 5,000 servers.
• Networking equipment to connect 5,000 servers.
• Telecommunications (telecom) equipment.
• Specialized firewalls and other protocols for enhancing the cybersecurity of Data Centers.

Keep in mind that this list is in no way comprehensive and doesn’t represent the full costs associated with building a Hyperscale Data Center. For example, it doesn’t include the primary asset needed for such activities: electricity.

Hyperscale vs. Colocation source: IBM

Who is building Hyper-scale?

Hyperscale Data Centers are designed to support very large-scale IT infrastructure. According to Synergy Research Group, as of 2021 there were only 700 Hyperscale Data Centers in existence—twice as many as five years prior.  As of 2024 there were 1,136. While this may be a small percentage compared to the number of Data Centers across the globe (there are more than 7 million data centers worldwide), Hyperscale Data Centers are on the rise. Interesting fact: Amazon, Microsoft and Google account for more than half of all Hyperscale Data Centers.

Like Enterprise Data Centers, Hyperscale Data Centers are owned and operated by the company they support—just on a much larger scale for cloud computing platforms and big data storage. A typical Hyperscale Data Center has at least 5,000 servers, 500 cabinets and 10,000 square feet of floor space.

Understanding differences between 5 common types of Data Centers source: Data Center Frontier 

Hyperscale Data Centers are Hungrier

The new Hyperscale Data Centers are a different animal from their predecessors. A decade ago, an average Data Center used 20 megawatt hours of power a month, about 20 times the use of an average home. The newest generation of large-scale Data Center campuses that support artificial intelligence with its large energy-consuming chips can use 100 or more megawatt hours a month, or five times their predecessors. The Lawrence Berkeley National Laboratory’s recent report said that Data Centers used 4.4 percent of the nation’s electricity in 2022 and would use 6.7 to 12 percent by 2028.

Some utilities have created special contracts for data center companies. This can lead to a realignment of all customers’ rates. Because of their special deals, data center companies may get a break, but it is difficult to tease out if they do.

Thirsty for power and water AI crunching data centers spreadsource: Stanford University