Hyperscale data centers, operated by key companies known as hyperscalers, have emerged as the primary providers of digital infrastructure, encompassing cloud services and supporting the seamless functioning of the internet. As a result, these hyperscalers play a critical role in meeting the growing demands for computing power, storage, and network capabilities in today’s data-driven world.
Hyperscale refers to a computing infrastructure’s ability to rapidly and efficiently scale with demand, especially in data centers and cloud platforms, by adding resources like servers and storage. Key hyperscalers include AWS, Microsoft Azure, Google Cloud, Meta Platforms, Apple, and TikTok.
Dgtl Infra delves deeper into the concept of hyperscale, its implications for data centers, and the companies driving this trend, including hyperscale cloud providers and large internet companies. Continue reading to uncover the innovations behind hyperscale servers and how they are reshaping the future of cloud computing and internet services.
What is Hyperscale?
Hyperscale refers to the ability of a computing infrastructure to expand rapidly and efficiently in response to increasing demand. This concept is often associated with data centers and cloud computing platforms, where systems are designed to seamlessly scale out by adding more resources, such as servers, storage, and networking components, as needed.
Hyperscale data centers have become the backbone of many cloud service providers (CSPs). Companies like Amazon Web Services (AWS), Microsoft Azure, and Google Cloud Platform (GCP) leverage hyperscale computing to offer businesses and individuals affordable, on-demand computing resources.
Hyperscale computing has also attracted the interest of enterprises, particularly large internet companies, such as Meta Platforms (Facebook). As big data, artificial intelligence (AI), and the Internet of Things (IoT) continues to grow, many hyperscalers are investing in scalable, flexible, and efficient hyperscale data centers.
Hyperscale Data Centers
Hyperscale data centers, also known as cloud data centers, are massive, centralized, highly-efficient, and custom-built computing facilities that are operated by a single company. These facilities support primarily cloud service providers (CSPs) and large internet companies with enormous compute, storage, and networking requirements.
The key characteristics and benefits of hyperscale data centers are size and scalability, distributed computing, energy efficiency, high computing power, automation, and cost-effectiveness.
1) Size and Scalability
Hyperscale data centers are characterized by their immense size and the ability to seamlessly scale infrastructure in response to demand. These facilities often house tens of thousands or even millions of servers, storage devices, and networking components, providing vast amounts of computing resources. Furthermore, through the use of server virtualization, hyperscale data centers can operate millions of virtual machines (VMs), enhancing their computing capabilities to support numerous customers and applications.
Hyperscale data centers are typically built with scalable power capacity for requirements of 5 to 100 megawatts (MW) across 50,000 to 1+ million square feet of space. As such, these facilities can cost upwards of $1 billion to construct.
To scale capacity efficiently, hyperscale data centers employ a modular design approach, utilizing standardized, pre-fabricated components for rapid deployment, expansion, and simplified maintenance without compromising performance. This design strategy enhances flexibility, allowing individual components, such as power and cooling modules, to be replaced or upgraded without disrupting the entire system.
READ MORE: Modular Data Centers – Prefabricated Containers and Modules
2) Distributed Computing
Distributed computing is a key characteristic of hyperscale data centers, where computing tasks are broken down into smaller, manageable tasks that are processed simultaneously across multiple servers or nodes to ensure efficient and reliable processing. These servers or nodes are interconnected by high-speed networking switches, forming a pool of resources that enables enterprises to build, manage, and deploy applications anywhere, from large cloud data centers to smaller edge locations.
Cloud service providers (CSPs), such as Amazon Web Services (AWS), play a crucial role in enabling distributed computing by constructing and maintaining cloud regions and availability zones (AZs) across the globe. Simultaneously, edge computing services, such as AWS Outposts, extend CSP infrastructure, services, APIs, and tools to virtually any data center, colocation facility, or on-premises location.
Distributed computing enhances performance for a wide range of applications and workloads, as well as provides redundancy and fault tolerance. In the event of a hardware failure, tasks can be automatically redistributed to other servers or nodes, minimizing the impact on the overall system.
3) Energy Efficiency
Energy efficiency is a crucial consideration for hyperscale data centers, as their enormous size and power requirements can lead to significant environmental impact and operational costs. Dynamic provisioning of resources, advanced cooling systems (e.g., liquid cooling, free cooling, or adiabatic cooling), energy-efficient servers, and the use of renewable energy sources are all employed to minimize energy consumption and reduce the overall carbon footprint of these facilities.
Energy consumption for data centers is typically measured using an efficiency metric known as power usage effectiveness (PUE). Specifically, PUE is the ratio of the total amount of electricity consumed by a data center, including power distribution and cooling, to the amount of electricity delivered to its IT equipment – with a metric closer to 1.0 being optimal.
As shown below, hyperscale data center operators like Microsoft Azure, Google Cloud, Alibaba Cloud, and Meta Platforms (Facebook) all report a substantially lower PUE – in the range of 1.05 to 1.25 – as compared to the average data center, which achieves a PUE of 1.55, according to the Uptime Institute’s 2022 Global Data Center Survey.
Power Usage Effectiveness (PUE) – Hyperscale vs Average Data Center
|Data Center Operator||Power Usage Effectiveness (PUE)|
|Meta Platforms (Facebook)||1.09|
|Data Center – Average||1.55|
Major hyperscalers Amazon, Microsoft, Google, and Meta Platforms (Facebook) have all undertaken extensive renewable energy projects, utilizing solar and wind power to reduce their carbon footprints and promote sustainable operations. These initiatives involve investing in and constructing large-scale solar and wind farms, often with capacities of several hundred megawatts, as well as purchasing renewable energy to power their data centers.
4) High Computing Power
Hyperscale data centers are designed to deliver immense computing power, capable of processing and storing vast amounts of data at high speeds. This is achieved through the use of powerful servers, high-speed networking, and specialized hardware, such as solid-state drives (SSDs) and graphics processing units (GPUs).
High computing power enables hyperscale data centers to support a wide range of demanding applications, including artificial intelligence (AI) and machine learning (ML) workloads, big data analytics, augmented reality (AR) & virtual reality (VR), and video rendering.
To support these demanding applications, hyperscale data centers often operate at power densification levels of 10 kW per rack to 14 kW per rack, which is higher than traditional data centers. Hyperscale data centers support more complex IT requirements and benefit from economies of scale to implement sophisticated infrastructure and power-intensive, high-density computing.
Automation is a core characteristic of hyperscale data centers, enabling them to efficiently manage and optimize their vast infrastructure. This includes automated monitoring, maintenance, and resource allocation, as well as the use of artificial intelligence (AI) and machine learning (ML) to predict and respond to potential issues. Automation helps to reduce the operational complexity and human intervention required to manage these large-scale facilities.
Software-defined networking (SDN) is an example of how automation is used in hyperscale data centers, particularly because it enables centralized management and configuration of network behavior through software. SDN allows operators to swiftly provision and adjust network resources in response to changing demands, ensuring efficient resource utilization and streamlined management in large-scale computing environments.
READ MORE: Software-Defined Networking (SDN) Explained
Hyperscale data centers are cost-effective due to their ability to leverage economies of scale, modular design, and energy-efficient technologies, significantly reducing both capital expenditures and operational expenditures. The large scale of these facilities allows them to procure hardware, software, and other resources at lower costs, while the modular design approach and use of standardized components also contribute to cost savings by simplifying deployment, maintenance, and upgrades.
Additionally, hyperscale data centers employ advanced cooling systems, energy-efficient hardware, and sophisticated power management techniques, reducing energy consumption and lowering operational costs over time. Finally, hyperscale data centers are often located on the outskirts of major cities, providing cost advantages in land and power procurement.
Hyperscale companies, also known as hyperscalers, include both cloud service providers (CSPs) and large internet companies with enormous compute, storage, and networking requirements. Currently, there are over 25 companies worldwide that are hyperscalers, which include hyperscale cloud providers, such as Amazon Web Services (AWS), and large internet companies, like Meta Platforms (Facebook). Collectively, these hyperscale companies operate more than 500 hyperscale data centers.
Hyperscale Cloud Providers
Hyperscale cloud providers offer a wide range of services that cater to various computing needs, such as Infrastructure as a Service (IaaS), Platform as a Service (PaaS), and Software as a Service (SaaS). These cloud service models enable customers to access scalable and cost-effective computing resources, storage, and networking capabilities.
The most prominent hyperscale cloud providers are Amazon Web Services (AWS), Microsoft Azure, Google Cloud Platform (GCP), and Oracle Cloud, which are highlighted below:
Amazon Web Services (AWS)
Amazon Web Services (AWS), the cloud computing service of Amazon.com, is the largest hyperscale cloud provider globally. AWS has 31 cloud regions and 99 availability zones in operation, with plans to launch 5 more cloud regions and 15 more availability zones.
Globally, the data center portfolio of AWS totals 33.5 million square feet, of which the company leases 18.0 million square feet (54% of total) and owns 15.4 million square feet (46% of total).
READ MORE: Amazon Web Services (AWS) Data Center Locations
Microsoft Azure, the cloud computing service of Microsoft, is the second largest hyperscale cloud provider globally. Azure has 60+ cloud regions and 120+ availability zones in operation.
Globally, the data center portfolio of Microsoft Azure totals over 200 facilities, which are linked together by more than 175,000 miles of fiber optic lines.
READ MORE: Microsoft Azure’s Data Center Locations
Google Cloud Platform (GCP)
Google Cloud Platform (GCP), part of Alphabet Inc, is the third largest hyperscale cloud provider globally.
There are currently 36 cloud regions and 109 availability zones in operation by Google Cloud. These cloud regions and availability zones are situated throughout the United States, Americas, Europe, and Asia Pacific.
READ MORE: Google Cloud’s Data Center Locations
Oracle Corporation’s cloud computing unit, known as Oracle Cloud Infrastructure (OCI), is a key hyperscale cloud provider globally.
There are presently 41 cloud regions and 50+ availability zones in operation by Oracle Cloud. These cloud regions and availability zones are located throughout the United States, Canada, Europe, the Middle East, & Africa (EMEA), Latin America, and Asia Pacific.
READ MORE: Oracle Cloud’s Data Center Locations
Large Internet Companies
Large internet companies utilize hyperscale data centers to deliver their products and services to end users. These hyperscale companies include Meta Platforms (Facebook), Apple, TikTok (ByteDance), Uber, Twitter, Salesforce, NVIDIA, and LinkedIn.
Meta Platforms (Facebook)
Meta Platforms (NASDAQ: META) owns and operates 21 hyperscale data center campuses globally, comprising over 50 million square feet. Geographically, these hyperscale data centers are located throughout the United States, as well as parts of Europe (Denmark, Ireland, Sweden) and Asia Pacific (Singapore). Additionally, Meta leases further data center capacity, in additional locations, from third-party operators like Digital Realty and CyrusOne.
Meta designs and builds its hyperscale data centers and technical infrastructure so that it can serve its products, which include Facebook, Instagram, Messenger, and WhatsApp.
READ MORE: Meta / Facebook’s Data Center Locations
Apple (NASDAQ: AAPL) operates eight hyperscale data centers located in the United States, Denmark, and China. At full build-out, these facilities will collectively comprise approximately 10 million square feet. Additionally, two more hyperscale data centers are under development by Apple, in the United States and Europe, which would bring the company’s total portfolio to 10 data centers upon completion.
Apple’s hyperscale data centers help power the company’s growth in services, including iCloud, App Store, Apple Maps, Apple Music, Apple Pay, Apple TV+, iMessage, and Siri.
READ MORE: Apple’s Data Center Locations
Other Large Internet Companies
Many large internet companies do not build-out their own infrastructure. Instead, they outsource and lease hyperscale data center capacity from third-party service providers such as Digital Realty, CyrusOne, and CloudHQ. Examples of the large internet companies that lease the majority of their hyperscale data center capacity include TikTok (ByteDance), Uber, Twitter, Salesforce, NVIDIA, and LinkedIn.
Hyperscale servers are high-performance, energy-efficient computing systems specifically designed to support the massive processing and storage requirements of hyperscale data centers. These servers are built with a focus on scalability, modularity, and cost-effectiveness, and are purchased in volume by the hyperscalers, allowing them to negotiate better prices with suppliers.
Hyperscale data centers account for more than 50% of all installed data center servers, and according to NVIDIA, approximately 9 million hyperscale servers are deployed each year.
Characteristics of Hyperscale Servers
The key characteristics of hyperscale servers are high-density computing, cost-optimization strategies, and energy efficiency.
1) High-Density Computing
Hyperscale servers are characterized by high-density computing, where a large number of powerful processors (CPUs), memory, and storage components are packed into a compact form factor. This design approach enables hyperscale data centers to accommodate more computing resources in a smaller footprint, providing the immense processing power needed to support demanding applications and workloads.
2) Cost-Optimization Strategies
Hyperscale data center operators employ various tactics to optimize server costs, including custom hardware designs, standardization, and the use of refurbished equipment.
- Custom Hardware Designs: hyperscale data center operators often design their own server hardware tailored to their specific needs. This allows them to optimize server performance and efficiency, and eliminate unnecessary features that add to the cost. For example, AWS Graviton processors are designed to deliver cost-optimized performance for cloud workloads running in Amazon EC2
- Standardization: using standardized and modular server designs, hyperscale data center operators can simplify their infrastructure and reduce management overhead. The modular design of hyperscale servers also makes it easier to maintain, replace, and upgrade components as needed, helping to minimize overall costs
- Refurbished Hardware: some hyperscale data center operators purchase refurbished or used servers that still meet their performance requirements. This can help to further reduce costs, especially for non-critical workloads or secondary data centers
3) Energy Efficiency
Hyperscale servers are designed with energy-efficient components, such as high-efficiency power supply units (PSUs), voltage regulator modules (VRMs), and advanced cooling systems. These components help to reduce power consumption and improve overall efficiency for hyperscalers.
Hyperscale Server Architectures
Hyperscale server architectures are being designed by initiatives including the Open Compute Project (OCP) and Microsoft’s Project Olympus.
Open Compute Project (OCP)
The Open Compute Project (OCP) is an initiative launched by Facebook (now Meta Platforms), Intel, and Rackspace, among others, in 2011, with the aim of developing open-source hardware designs for data centers. The project focuses on creating highly efficient, cost-effective, and scalable server architectures that can be easily customized and adapted to various use cases.
OCP server designs prioritize energy efficiency, modularity, and simplified maintenance. Organizations can leverage OCP’s standardized server designs to build and deploy hyperscale data center infrastructure more efficiently.
As an example, Meta Platforms has developed the Open Rack enclosure, which is a 21-inch-wide rack design that allows for better airflow and more efficient power distribution compared to the traditional 19-inch-wide rack.
Project Olympus is a server architecture initiative by Microsoft, contributing to the Open Compute Project (OCP). Launched in 2016, Project Olympus aims to create an open-source, modular, and flexible hardware design for hyperscale data centers. The project focuses on developing a universal motherboard and chassis design that can support various configurations of processors, memory, storage, and networking components.
By providing a customizable and adaptable server architecture, Project Olympus enables hyperscalers to optimize their infrastructure for specific workloads and easily scale their data center capacity as needed.