Data centers, on-premises infrastructure, and the cloud

Since Module 2, you have known that software can run on remote machines.

Since Module 4, you have known that those machines are often serversConcept · lights on your mapserverThe role that receives requests and returns responses, played here by real hardware, usually a screenless machine installed in a data-center rack. The remote machines you have known since Module 2, now seen physically., receiving requests and returning responses.

Now we can ask an embarrassingly concrete question: where are those servers physically located?

The answer is often a data center.

What is a data center?

A data centerConcept · lights on your mapdata centerA facility designed to house and operate computing equipment: racks of servers, storage, and networking gear, engineered around reliable electricity, cooling, connectivity, and physical security. The building around the computer is part of the technical system. is a facility designed to house and operate computing equipment.

Inside, rows of computers are commonly installed in tall metal structures called racks, which contain servers, storage devices, networking equipment, and other hardware needed to run software systems.

A data center may belong to one company, be shared by several organizations, or be operated by a cloud provider serving thousands of customers.

Some resemble enormous industrial buildings. Others occupy a floor, room, or specialized section of a larger facility.

Regardless of size, they are engineered around the same handful of critical needs. Reliable electricity, cooling, network connectivity, physical security, fire protection, hardware maintenance, and backup systems all have to be in place.

A server cannot answer requests if it has no power, overheats, loses its network connection, or is physically damaged.

The building around the computer is therefore part of the technical system.

Electricity

Servers require continuous electrical power.

A large data center can consume enormous amounts of electricity because thousands of processors, memory modules, storage devices, and network components may be operating simultaneously.

A brief power interruption could stop applications, interrupt database writes, and disconnect customers.

Data centers therefore commonly use several layers of protection, such as multiple power connections, battery systems, backup generators, redundant electrical equipment, and monitoring for electrical failures.

The exact design varies, but the goal is straightforward: a single power problem should not automatically shut down the entire facility.

This is an early example of redundancy, having additional components available when one component fails, and you will encounter that principle repeatedly throughout the infrastructure modules.

Cooling

Computers turn electrical energy into heat.

One laptop can become noticeably warm. Now imagine thousands of much more powerful machines operating closely together.

Without cooling, the equipment could overheat, slow down, become unreliable, or shut itself off to prevent damage.

Data centers use cooling systems to remove that heat and maintain safe operating conditions.

Depending on the facility, this may involve air conditioning, carefully managed airflow, cold and hot aisles, liquid cooling, outside air, water-based systems, and temperature and humidity monitoring.

A substantial portion of data-center engineering is therefore not traditional software engineering. It involves electricity, mechanical systems, buildings, and environmental control.

The software world depends on physical infrastructure.

Connectivity

And servers must also communicate with users and other systems.

Data centers connect to high-capacity networks that carry information across cities, countries, and continents.

Networking equipment inside the facility directs traffic between servers, storage systems, other data centers, internet providers, cloud services, and customer devices.

When you tap Pay, your request may travel through local networks, internet providers, routing equipment, and several remote systems before reaching the backend.

Then the response travels back through networks to your device.

The “cloud” may feel invisible, but the data still moves through physical cables, routers, switches, radio links, and buildings.

Where did your request go?

In Mini-capstone A, you traced a request from the browser to a backend and then to other services, and those remote systems ultimately ran on physical machines somewhere.

The coffee application’s backend may run in one data center, its database may run in another section of the same facility or in a separate facility, and its payment provider may operate in entirely different data centers.

A single user action can therefore involve several physical locations:

Not every request travels to one distant central building.

Some content may be served from a nearby edge location or content delivery network. Some work may happen directly on the user’s device. Large services may operate across many data centers.

The important mental shift is: remote software still runs somewhere physical.

What is on-premises infrastructure?

Before public cloud platforms became widespread, companies commonly purchased and operated their own computing infrastructure.

This approach is called on-premisesConcept · lights on your mapon-premisesHardware operated primarily for one organization’s own use, whether in its building, its own data center, or a colocation facility. The alternative to renting cloud infrastructure on demand; the practical question is which fits each workload., often shortened to “on-prem.”

On-premises infrastructure is hardware operated primarily for one organization’s own use, not rented as a public cloud service.

The equipment might be located in a server room inside the company’s building, in a company-owned data center, in a rented data-center facility, or in several private locations.

The organization is responsible for operating the environment.

That may include purchasing servers, installing equipment, configuring networks, replacing failed hardware, and applying updates. It may also include planning capacity, managing physical access, maintaining backups, providing power and cooling, and responding to outages.

The company gains control, but it also assumes the operational responsibility.

What is colocation?

A company does not need to own an entire building to own its servers.

It can place its equipment inside a facility operated by another company, an arrangement called colocation, often shortened to colo.

The colocation provider supplies facilities such as rack space, electricity, cooling, physical security, and network connectivity.

The customer still owns or controls the servers placed there.

A simplified distinction is: with on-premises or colocation infrastructure, the organization operates its own hardware environment. With public cloud infrastructure, the cloud provider owns the underlying hardware and rents computing capabilities to customers.

Colocation sits between the image of “servers down the hall” and the fully rented public cloud.

The capacity problem

Operating physical servers creates a planning challenge.

A company must decide how much computing capacity to buy before it knows exactly how much it will need.

Suppose the coffee application expects 100,000 users.

If the company buys too little capacity, a successful launch could overwhelm the servers, while if it buys far more than necessary, expensive equipment may remain underused. This is called capacity planning.

Physical infrastructure may require upfront purchasing, shipping, installation, and configuration, along with long procurement processes, space for future growth, and replacement as hardware ages.

The organization must make decisions today based on uncertain future demand.

What is cloud computing?

Cloud computingConcept · lights on your mapcloud computingThe delivery of computing resources and managed services over a network from provider-operated data centers: processing, storage, networking, databases, and more, created on demand and billed by usage. Not the absence of computers: a different ownership and operating model for them. Major providers: AWS, Microsoft Azure, Google Cloud. is the delivery of computing resources and services over a network, commonly through large providers that operate data centers on behalf of customers.

Instead of purchasing every physical server, a company can request computing capacity from a cloud provider.

It might rent processing power, memory, storage, networking, databases, message queues, analytics systems, security services, artificial-intelligence infrastructure, and many other managed capabilities.

These resources can often be created through a website, command-line tool, configuration file, or API.

The physical machines still exist. The cloud changes who owns them, who operates them, and how customers access their capabilities.

A useful first mental model is: cloud computing turns infrastructure into an on-demand service.

The cloud is not magic

The word cloud can make computing sound as though it happens nowhere in particular, but in reality, cloud providers operate enormous networks of physical data centers.

The provider manages buildings, power, cooling, physical servers, storage hardware, networking equipment, hardware replacement, and much of the foundational infrastructure.

The customer rents resources built on top of that machinery.

The cloud is therefore not the absence of computers but a different ownership and operating model for them.

Major cloud providers

Three names appear frequently in technical conversations:

• Amazon Web Services, or AWS
• Microsoft Azure
• Google Cloud

These companies operate data centers around the world and offer large catalogs of infrastructure and software services.

Other cloud providers also exist, including specialized and regional providers.

When an engineer says “we run on AWS,” they generally mean that some important portion of the organization’s infrastructure uses services operated by Amazon Web Services.

That could include backend servers, databases, file storage, networks, analytics systems, monitoring, authentication services, and development tools.

It does not necessarily mean that every part of the company uses AWS. A company may use several cloud providers or combine cloud services with private infrastructure.

The shared responsibility model

When a company uses the cloud, it does not give the provider responsibility for everything.

Responsibility is divided between the provider and the customer, an arrangement commonly called the shared responsibility model.

The cloud provider typically protects and operates the underlying infrastructure, including the physical buildings, power and cooling, physical servers, foundational networks, and hardware maintenance.

The customer typically remains responsible for areas such as:

• Application code
• User permissions
• Data
• Security configurations
• Network rules
• Which services are exposed publicly
• Software running inside rented machines
• Backups and retention, depending on the service
• Compliance with its own obligations

For example, if a company accidentally makes a private database publicly accessible, the database may be running correctly inside the provider’s facility. But the customer’s configuration caused the exposure.

A useful principle is: the provider secures the cloud infrastructure. The customer must securely use what it rents.

The exact boundary changes depending on the service. The more management the provider performs, the less infrastructure work remains for the customer, but the customer never gives up responsibility for its product and data entirely.

Why companies use the cloud

Cloud computing provides several major advantages.

Faster provisioning

A team can often create a new server or database in minutes, without purchasing and installing physical hardware.

Flexible capacity

Resources can be added or removed as demand changes.

Lower upfront investment

The company may avoid purchasing large amounts of hardware before the product has proven its demand.

Global infrastructure

Cloud providers operate facilities in multiple geographic areas, allowing products to run closer to users or satisfy regional requirements.

Managed services

Teams can rent databases, queues, storage, analytics tools, and other capabilities that the provider helps operate.

Automation

Infrastructure can be created and configured through code and APIs, not only through manual hardware installation, and these benefits allow teams to focus more engineering effort on their product and less on operating physical facilities.

Elasticity

One of those advantages is also one of cloud computing’s central ideas. Flexible capacity has a proper name, elasticity, which means increasing or decreasing computing resources as demand changes.

Imagine that the coffee application normally receives 1,000 orders per hour but expects a major promotion to produce ten times that traffic.

In a cloud environment, the team may be able to add more computing capacity during the promotion and reduce it afterward, which is easier than physically buying, installing, and later removing servers for a temporary event.

The mental model is:

• Low demand → fewer resources
• High demand → more resources

However, elasticity is not automatically perfect. The application must be designed and configured to use additional capacity effectively. Renting scalable infrastructure does not guarantee that the software itself can scale.

Paying for usage

Cloud services commonly use consumption-based or subscription pricing.

A customer may pay according to how long computing resources run, how much storage is used, how many database operations occur, how much data travels across networks, how many API requests are processed, and which service tier is selected.

This can reduce upfront spending, but it also creates ongoing operating costs.

A forgotten server, inefficient query, excessive logging system, or unexpected traffic spike can increase the bill.

Cloud-cost management is therefore an engineering and business discipline.

Teams may ask:

• Which resources are running?
• Are they being used?
• Can they be reduced when idle?
• Which feature caused the cost increase?
• Are we paying for unnecessary data transfer?
• Would a managed service or owned system be cheaper at our scale?

The cloud turns infrastructure spending from a large purchase into a continuing meter.

The cloud creates dependencies

Using a cloud provider creates one of the product’s deepest dependencies. If an important cloud service becomes unavailable, applications relying on it may fail even when their own code has not changed.

A cloud outage can affect websites, mobile applications, databases, internal tools, authentication systems, payment flows, analytics, and development environments.

Organizations reduce this risk using techniques such as redundancy, multiple facilities, backups, disaster-recovery plans, graceful degradation, and, in selected cases, multiple cloud providers.

However, resilience across providers is expensive and complex.

The same tradeoff from Module 5 appears again: renting gives speed and capability, but creates dependency on the provider’s systems, pricing, contracts, and availability.

On-premises versus cloud is not all-or-nothing

Organizations do not always choose one model for everything.

A hybrid cloud approach combines private or on-premises infrastructure with public cloud services.

A company might keep a sensitive legacy database in its own environment while using cloud services for analytics and customer-facing applications.

A multi-cloud approach uses more than one public cloud provider.

A company might use one provider for its main application and another for a specialized data or AI service.

These approaches may provide flexibility or satisfy specific requirements, but they also create additional operational complexity.

The practical question is rarely: “is cloud always better than on-prem?”

It is more often: “which infrastructure model best fits this workload, risk, cost, regulation, and team?”

Private cloud

You may also hear the term private cloud.

A private cloud uses cloud-like automation and self-service capabilities for infrastructure dedicated to one organization.

The organization may operate it itself or use a provider, but the hardware is not shared through the same public service model as a typical public cloud.

Private cloud can provide greater control, but the organization still assumes more responsibility for operating the underlying environment.

The term shows that “cloud” describes more than who owns one computer. It also describes how computing resources are delivered and managed.

Regions and locations

Cloud providers organize infrastructure across geographic areas.

A broad geographic area is commonly called a region.

For example, a provider may operate regions in Northern Virginia, California, Europe, Asia, and South America.

Inside a region, providers commonly operate several separate facilities or groups of facilities so that one local failure does not necessarily affect everything.

The exact terminology differs by provider, and you will examine it more closely later.

For now, understand that choosing where infrastructure runs can affect speed for users, availability, cost, legal requirements, data residency, and disaster recovery.

The cloud is global, but each resource still exists in particular physical locations.

Seeing the coffee application physically

The coffee product may look like one application on Erik’s phone.

Physically, it could involve frontend code running on Erik’s device, static files served from a nearby edge location, backend services running in a cloud region, and a managed database running on cloud infrastructure. It could also involve backups stored in another location, payment processing running in another company’s cloud environment, analytics data stored in a warehouse, and internal tools used from company offices.

The product is distributed across devices, networks, facilities, providers, and regions.

The user sees one interface. Underneath, the product is a coordinated collection of physical and rented computing resources.

The mental model to remember

A data center is a facility designed to operate servers, storage, and networking equipment with reliable power, cooling, connectivity, and security.

On-premises, or on-prem, infrastructure is operated primarily for one organization’s own use.

Colocation means placing organization-controlled hardware inside another company’s data-center facility.

Cloud computing delivers computing resources and managed services on demand from provider-operated infrastructure.

AWS, Microsoft Azure, and Google Cloud are major public cloud providers.

Elasticity means increasing or decreasing resources as demand changes.

The shared responsibility model divides responsibility between the cloud provider and the customer.

A hybrid cloud combines private infrastructure with public cloud services.

A multi-cloud strategy uses services from more than one public cloud provider.

You should now understand that the cloud is not an imaginary place. It is a service model built on physical data centers, servers, storage devices, networks, and people operating them.

The data center is the stage.