Cloud IT services explained: what they include and who manages what

Cloud IT services can mean the technology you rent, the software you use, or the people paid to keep it running. A virtual machine, a managed database, and an outsourced help desk solve different problems, even when a proposal puts all three under the same heading.

To compare offers, establish two things: what capability you receive and who operates each part of it. This guide explains the main service categories, the responsibilities that stay with your team, and the questions to settle before you commit.

What is a cloud service?

A cloud service is a computing capability delivered over a network through cloud infrastructure that pools computing resources, supports on-demand access, and measures use. Examples include processing capacity, object storage, a database, an AI endpoint, or a finished application. Cloud computing describes the delivery model; a cloud service is the capability you actually use.

You can access cloud services through a browser, an application, a command-line tool, or an API. The physical infrastructure still occupies real locations, including data centers. Our guide to where cloud data is stored explains why those locations matter.

What makes it cloud rather than simply hosted?

NIST SP 800-145 identifies five characteristics: self-service provisioning, access through standard networks, pooled resources, rapid expansion or release of capacity, and measured usage. Its companion evaluation guide helps distinguish cloud services from other computing arrangements.

Remote access alone is insufficient. Equally, cloud does not necessarily mean public internet access or consumption-based billing. Private networks and subscriptions can fit the model. Elasticity also has practical limits: capacity, quotas, configuration, and the service design determine what can scale and how quickly.

Four meanings of cloud IT services

These terms describe different purchases. A provider may offer several, so check the scope of the individual service.

• Cloud service: a technical capability, such as a GPU instance, storage bucket, or online application.
• Managed cloud service: a cloud product or operational arrangement in which the provider runs specified parts of the stack. A managed database removes some database administration work, but its backup, tuning, and recovery provisions still need checking.
• Managed IT service: ongoing administration of an agreed part of your environment. A managed service provider, or MSP, might monitor systems, patch software, support users, or handle incidents.
• Professional service: a defined project, such as migration, architecture design, integration, security review, or training.

For example, a team can rent cloud servers, hire an MSP to maintain their operating systems, and commission a consultant to migrate an application. These are three separate commitments. Buying the servers does not include the other two unless the agreement says so. An MSP can operate another company's infrastructure; it does not need to own a cloud platform.

Eight types of cloud services and IT support

Cloud computing services and operational support address the following common needs. The management boundaries are typical examples, not promises about every product. Pricing units are also examples: the applicable rate card and contract determine the bill.

1. Compute services

What you receive: processing power through virtual machines, CPU or GPU instances, containers, bare-metal servers, or batch environments. Typical uses include application development, web applications, rendering, simulation, and model training.

Who manages what: the cloud provider manages the hardware and provisioning platform. With a VM, your team usually manages the guest operating system, installed software, access, data, and application health. Containers and managed batch products change that boundary, so check the available controls.

Pricing: instance runtime, GPU runtime, or allocated CPU and memory, with possible storage and network charges.

Hivenet fit:Compute with Hivenet offers GPU and CPU instances with per-second billing. Its VM and container options expose different levels of control: VMs provide OS access, while containers use templates and have OS-level restrictions. Choose according to the software you need to operate.

2. Data storage and backup services

What you receive: object, file, or block storage, archives, or a backup application. Common uses include application files, datasets, media, and recovery copies.

Who manages what: the provider operates the storage system. Your team decides what to save, who can access it, how long to retain it, and how to recover it. Verify whether versioning, backup scheduling, deletion protection, or restore assistance is actually included.

Pricing: stored capacity over time, requests, retrieval, transfer, or a subscription. Minimum retention periods can also affect cost.

Hivenet fit:Hivenet Object Storage serves S3-compatible application and dataset workflows. Store with Hivenet is a separate product for personal files and photo backup. Avoid applying one product's storage architecture or features to the other.

Storage capacity alone is not a disaster recovery plan. Synchronization and backup have different purposes; test whether you can restore the version you need after an accidental change or deletion.

3. Networking services

What you receive: capabilities such as virtual networks, addresses, firewalls, DNS, load balancing, private connections, or content delivery. These connect applications, control access, and move traffic between systems.

Who manages what: the provider operates the underlying network and the networking product. Customers configure the routes, domain records, access rules, certificates, or other settings exposed by that product.

Pricing: transfer volume, provisioned capacity, gateway or load-balancer runtime, and requests, depending on the service.

Hivenet fit: instance connectivity is part of the Compute workflow. If your design requires a particular private connection, network appliance, or global delivery service, confirm that requirement separately. Access to a compute instance does not establish the availability of a full networking catalog.

4. Application and database platforms

What you receive: managed application runtimes, databases, caches, container platforms, or serverless computing functions. These can suit teams that want to deploy code or query data with less platform maintenance.

Who manages what: the provider runs the layers included in the product, such as the database engine or runtime. Customers still manage application logic, schemas, queries, credentials, and exposed configuration. Recovery and scaling options vary.

Pricing: allocated capacity, runtime, storage, requests, function executions, or consumption units.

Hivenet fit: suitable applications and databases can run on Compute instances. That is a self-managed deployment unless an additional agreement assigns administration elsewhere. Running a database on a VM does not make it a managed database service.

5. AI and data services

What you receive: infrastructure for training and analysis, managed model endpoints, or data-processing tools. The right choice depends on whether you need control of the software stack or an interface that runs an agreed task.

Who manages what: on rented compute, customers operate their workload software. With a managed endpoint, the provider operates the serving components it includes. Customers still handle application integration, input data, access, evaluation, and review of model outputs.

Pricing: GPU runtime, endpoint capacity, requests, tokens, or processed data. These units are not interchangeable.

Hivenet fit: Compute supports customer-operated workloads. Hivenet Inference API provides managed, OpenAI-compatible endpoints with dedicated replica capacity billed by runtime. Private AI is a guided route for scoping model, data, infrastructure, and support requirements. Confirm the responsibilities for the proposed deployment.

6. Identity and security services

What you receive: capabilities such as directories, access management, secrets storage, logging, threat detection, or an agreed security operations service. These support access control and investigation across applications and infrastructure.

Who manages what: the provider runs the supplied product or contracted operation. Customers define appropriate permissions, maintain user access, protect credentials, and decide how alerts should be handled. A monitoring tool does not necessarily include someone responding to it.

Pricing: users, protected resources, ingested data, retained logs, or an operational retainer.

Hivenet fit: Compute includes organization roles and access controls. Treat these as controls for that product, not evidence of a company-wide identity platform or an outsourced security operations center.

7. Finished cloud applications

What you receive: usable software applications for tasks such as email, accounting, customer relationship management, collaboration, or file sharing. This is the familiar software-as-a-service, or SaaS, model.

Who manages what: the provider runs and updates the application and its supporting infrastructure. Customers manage their users, settings, data, sharing choices, devices, and connected applications.

Pricing: user subscriptions, feature tiers, storage allowances, or usage.

Hivenet fit:Store with Hivenet provides personal file storage and photo backup; Send with Hivenet supports file transfer. Evaluate each against the task you need, rather than treating them as a complete business software suite.

8. Managed IT operations

What you receive: ongoing work such as monitoring, patching, backup administration, help-desk support, cost management, or incident response. This can cover cloud and on-premises systems.

Who manages what: the MSP performs the tasks and coverage hours specified in the agreement. The customer retains decisions about business priorities, acceptable access, budgets, and oversight. Escalation contacts and responsibilities need named owners.

Pricing: per user, device, managed resource, monthly retainer, or an agreed support allowance.

Hivenet fit: operating its cloud products does not mean Hivenet administers a customer's entire IT environment. Product support should not be assumed to include employee devices, an end-user help desk, or full security operations. Those require an explicit scope.

Who manages each layer of the cloud computing stack?

Infrastructure as a service (IaaS) generally leaves more operating work with the customer. Platform as a service (PaaS) transfers more of the runtime or platform work. SaaS transfers operation of the application itself. Microsoft Azure shared-responsibility guidance illustrates how that boundary changes across the three models.

AWS's explanation gives a useful concrete distinction: a customer operating a virtual machine manages its guest OS and installed applications, whereas an abstracted storage service leaves those underlying layers with the provider. Permissions and the customer's use of data still need attention.

Make the boundary operational. For each application, name who patches software, reviews access, watches alerts, tests restores, handles an incident, and approves spending. If two suppliers are involved, specify who coordinates them. A platform availability commitment does not establish that your application is configured correctly or that someone will restore it.

How cloud resource allocation works in practice

Consider a team creating a development environment. Someone requests capacity through a console or API. The platform checks identity, permissions, quotas, and availability, then allocates cloud resources and exposes an access method. The team installs or connects its application, runs the work, and monitors performance and cost.

At the end, the team exports what it needs and stops or deletes the resources according to the product's rules. Stopping compute, retaining a disk, and deleting data are distinct actions. Confirm which charges continue and how long retained data remains available. A service's lifecycle is part of the purchase, not an administrative detail to discover afterward.

Questions to ask cloud service providers

Use these questions to turn a feature list into a service you can operate:

1. What exactly are we buying? Identify the capability, its limits, and whether the offer is a product, ongoing operations, or a project.
2. Who does each task? Record the provider's responsibilities, the skills your team needs, and any operational gaps.
3. Where will the workload and data run? Check the actual service location, available capacity, network needs, and relevant contractual terms.
4. How will access be controlled? Review roles, credentials, logging, staff departures, and access to sensitive data by support personnel.
5. What will the full workload cost? Include runtime, storage, requests, transfer, support, commitments, migration, and your team's time. Test a realistic usage pattern.
6. What happens during failure? Confirm monitoring, incident contacts, coverage hours, backup responsibilities, restore procedures, and the scope of any service-level agreement for service availability.
7. Can the service handle our workload? Test performance, quotas, concurrency, capacity changes, and integrations with existing systems instead of relying on a category label.
8. How do we leave? Test exports and restore elsewhere. Check deletion rules, termination notice, retention after closure, and the work required to replace proprietary features.

Benefits of cloud services and their trade-offs

Cloud services can reduce hardware purchasing and transfer some maintenance work. They also introduce recurring costs, provider dependencies, network requirements, and service limits. The useful comparison is the cost and effort of running a specific workload to the required standard.

Where Hivenet fits in a cloud IT plan

Choose by the responsibility you want to retain. Compute with Hivenet is relevant when your team wants to operate workload software on rented processing capacity. Inference API is relevant when you want a managed model endpoint. Object Storage serves application data workflows, while Store and Send address file storage and transfer for users.

For an AI application, that might mean storing input files in object storage and calling a managed endpoint from your own application. Your team still owns the integration, access decisions, and output evaluation. If you instead need a custom model-serving stack, renting compute changes both your control and your operational workload.

Compare those options against equivalent services. If the missing requirement is database administration, employee support, or a migration project, specify it separately and establish who will deliver it.

Frequently asked questions

Are public and private clouds managed differently?

Public cloud services are available to a broad customer base, while private clouds serve one organization. A hybrid cloud connects distinct cloud environments. These deployment choices do not, by themselves, say who patches an application or manages its backups; check the service and operating agreement.

Do cloud services always use pay-as-you-go pricing?

No. Services may use subscriptions, prepaid credits, usage allowances, commitments, or consumption-based charges. Measured use and the commercial billing arrangement are related but distinct.

Does managed mean the provider handles everything?

No. It means specified work is included. Ask what happens to patching, backups, monitoring, recovery, application configuration, and user support. Each may sit with a different party.

Do we need an MSP as well as a cloud provider?

You may if your team cannot cover the operational tasks left by the chosen products. A managed product can remove some tasks; an MSP can take on an agreed set of remaining work. Start with the gaps rather than buying a broad label.

Do standard APIs make switching providers easy?

They can reduce integration changes, but migration still involves data transfer, permissions, feature differences, testing, and operational handover. Verify the functions you use and rehearse an export before depending on portability.