Skip to content
Eudora Technology
Home  / Insights
Cloud Technology

An Introduction to Cloud Technology

The data center of CNPC in Changping district.
Photo: Data Center of CNPC by Charlie fong, CC BY-SA 4.0, via Wikimedia Commons

Cloud computing is renting computers, storage and software by the hour from someone whose entire business is keeping them running. You get an API and a console instead of a purchase order and a delivery van, and you pay for what you switch on. That is the whole idea. Everything else, the service names, the certifications, the diagrams with little clouds on them, is detail layered on top.

Fotografías tomadas un día domingo en la ciudad universitaria de la Universidad Nacional de Córdoba
A research data centre: the physical thing you stop buying when you move to cloud.Photo: Data Center (UNC) by Ana Las Heras, CC BY-SA 4.0, via Wikimedia Commons

What the cloud actually is

The US National Institute of Standards and Technology published the definition everyone still quotes, Special Publication 800-145, back in 2011. It has aged well because it describes behaviour rather than products. NIST lists five essential characteristics: on-demand self-service, broad network access, resource pooling, rapid elasticity and measured service. If a supplier makes you raise a ticket and wait three days for a server, whatever they are selling is not cloud.

NIST also names four deployment models: private, community, public and hybrid. Most small and mid-sized businesses use public cloud for new systems and keep one or two older things on their own hardware, which is a hybrid estate whether anyone calls it that or not. Flexera’s 2026 State of the Cloud research found 73% of organisations operating hybrid, up three percentage points year on year, so the messy middle is the normal state, not a transitional phase.

The five characteristics, applied as a test
  • Can you provision it yourself in minutes, with no human in the loop?
  • Can you reach it over standard networks from the devices your team already uses?
  • Is the underlying hardware shared and abstracted away from you?
  • Can you scale up for a campaign and back down afterwards?
  • Is usage metered, with a bill you can attribute to a team or a project?

The three service models in plain language

The three service models decide how much work stays with you. Infrastructure as a service hands you a virtual machine and a network, and you still patch the operating system. Platform as a service hands you a managed database or application runtime, and the provider patches underneath you. Software as a service hands you a finished application, and your job is configuration, data and user accounts.

ModelYou look afterProvider looks afterTypical use
IaaSOS patching, runtime, application, data, accessHardware, hypervisor, network, physical securityLift-and-shift of existing servers, bespoke workloads
PaaSApplication code, schema, data, accessOS, runtime, managed database, scaling, backups of the serviceNew applications, managed databases, container platforms
SaaSData, user accounts, configuration, how you use itThe entire stack including the applicationEmail, finance, CRM, helpdesk, file sharing
Responsibility moves to the provider as you move down the list. Your accountability for data never moves.

That table is the single most useful thing to internalise, because it predicts where your effort goes and who is at fault when something breaks. We go deeper in our guide to the service models, including how each one is priced.

How the bill is put together

Cloud invoices have three main shapes: compute time, stored data, and data movement. The figures below were read from vendor price lists in October 2026 in their US regions, and they will drift, so treat them as orders of magnitude rather than quotes.

Line on the billPublished rateWhat it means in practice
General-purpose VM, 4 vCPU and 16 GB (Azure D4s v5, Linux, East US)$0.192 per hourAbout $140 a month if you leave it running all month
The same VM on interruptible Spot capacity$0.040531 per hourRoughly a fifth of the price, if the workload can be evicted
Object storage, hot tier (Amazon S3 Standard, first 50 TB, N. Virginia)$0.023 per GB-month1 TB of active files costs about $23 a month
Archive tier (S3 Intelligent-Tiering Deep Archive Access)$0.00099 per GB-month1 TB of cold archive costs about $1 a month, with retrieval fees
Internet egress after the free allowance (Azure, from North America or Europe)$0.087 per GBServing 1 TB of downloads a month costs roughly $87
Three-year commitment (AWS Savings Plans or Reserved Instances)Up to 72% off on-demandThe largest single discount available, in exchange for a promise
List prices read in October 2026 from AWS, Azure and their published price lists. Regions and commitments change the numbers.

Notice what is missing: there is no line for the three days you spent sizing a server, the capital you tied up in hardware you half-use, or the weekend you lost to a failed RAID controller. Those costs are real, they just do not appear on an invoice, which is why like-for-like comparisons with on-premises kit are so often argued and so rarely settled.

Data Center của CMC Telecom
Shared, pooled hardware is what makes per-hour pricing possible.Photo: Data Center của CMC Telecom (2) by Daoducquan, CC BY-SA 4.0, via Wikimedia Commons

Where cloud saves money, and where it quietly does not

Cloud wins clearly when demand is uneven, when you need to be in several countries, when the alternative is buying hardware you will outgrow, and when you would rather pay for a managed database than learn to tune one. It wins quietly on security patching, because managed services get patched whether or not anyone on your team remembers.

It loses when a workload is flat, predictable and already paid for. A steady database on owned hardware with three years left in it is not a cloud candidate on cost grounds alone. Flexera’s 2026 research estimated 29% of IaaS and PaaS spending is wasted, and reported that 71% of organisations now run a cloud centre of excellence to get that under control. The waste is not the provider’s fault. It is idle instances, oversized disks, forgotten test environments and data nobody deleted.

Flexera 2026 State of the Cloud: three numbers worth knowing
Organisations running hybrid estates73%
Organisations with a cloud centre of excellence71%
Estimated wasted IaaS and PaaS spend29%
Source: Flexera, 2026 State of the Cloud Report, as summarised on Flexera’s FinOps blog.

The practical conclusion is unglamorous. Cloud is cheaper than owning hardware only if somebody owns the bill. Set a monthly review, tag everything by project, and delete what nobody can explain. Our walkthrough of how to cut cloud costs without losing performance is the next thing to read if your invoice has been creeping.

What reliability you are really buying

Providers sell nines, but they sell them conditionally. Amazon’s Compute Service Level Agreement commits to a 99.99% Monthly Uptime Percentage at region level only when your instances run concurrently across two or more Availability Zones. A single instance gets 99.5%. Miss the target and you receive a service credit, starting at 10% of the bill, rising to 100% below 95.0% uptime. A credit is not compensation for lost trade.

Downtime is expensive enough to plan for properly. The Uptime Institute’s 2026 annual outage analysis reported that 57% of respondents said their most recent major outage cost more than $100,000, and for the second year running one in five put the cost above $1 million. Those are enterprise numbers, but the shape holds at smaller scale: the cost of being down for a day is almost always larger than the cost of the second Availability Zone you did not deploy.

Cost of the most recent major outage (Uptime Institute, 2026)
More than $100,00057% of respondents
More than $1 million1 in 5 respondents
Source: Uptime Institute, Annual outage analysis 2026, based on its 2025 annual survey.
Data center infrastructure in the United States
Someone else’s night shift: part of what the monthly bill actually covers.Photo: Data center infrastructure in the United States by DOE / National Renewable Energy Laboratory (NREL), Public domain, via Wikimedia Commons

A sensible first 30 days

You do not need a strategy document to start. You need one real workload, one budget alarm and a rule about who can create things.

  1. Pick one workload that is annoying you. A file share people cannot reach from home, a staging site, a backup target. Not the finance system.
  2. Open one account and turn on billing alerts first. Before any resource exists, set an alert at your intended monthly spend and another at double it.
  3. Use managed services where you can. A managed database costs more per hour and less per year than a VM you patch yourself.
  4. Tag everything with a project and an owner. Untagged resources are how 29% waste happens.
  5. Decide your recovery targets before you build. How much data can you lose, how long can you be down? Write the numbers down and design to them.
  6. Review the bill in month two, line by line. This is the single habit that separates teams who like their cloud from teams who resent it.

If you want a second pair of eyes on any of that, our cloud solutions work is delivered remotely for clients worldwide: assessment, build, migration and the monthly cost review. Clients who want hands-on-site support in Sri Lanka are looked after through eudora.lk. Two more posts worth reading before you commit: on-premises versus cloud for the money argument, and our provider comparison for the shortlist.

Frequently asked questions

Is the cloud cheaper than buying our own server?

Sometimes. For uneven or growing workloads, usually yes, because you stop paying for peak capacity you use twice a year. For a flat workload on hardware you already own, often no. The honest comparison includes power, cooling, replacement cycles, backups and the hours your team spends on maintenance, not just the hardware invoice.

What is the difference between cloud and hosting?

Self-service and metering. Traditional hosting gives you a fixed server for a fixed monthly fee, provisioned by a human. Cloud lets you create and destroy resources through an API in minutes and bills you by the second, hour or GB. NIST’s SP 800-145 lists on-demand self-service and measured service as essential characteristics for this reason.

Is our data safe in a public cloud?

The platform is almost certainly better protected than a cupboard in your office, but security is shared. The provider secures the facility, the hypervisor and the managed service. You still own identity, access rules, configuration, encryption choices and backups. Most incidents we see come from a misconfigured bucket or a reused password, not from the provider.

How long does a first migration take?

For a single application with a database, plan two to six weeks of elapsed time, most of it testing rather than copying. The copy is the easy part. Agreeing acceptable downtime, proving the restore works and updating DNS and integrations is where the days go.

Do we need an in-house cloud engineer?

Not at the start. You need someone accountable for the bill and someone who can read a runbook. We work with clients remotely across time zones, which keeps a small team from having to hire a specialist before the workload justifies one.

Thinking about a first move to cloud and want a plain assessment rather than a sales pitch? Tell us what you run today. Get in touch with Eudora Technology to talk about your project.

Sources

  1. The NIST Definition of Cloud Computing (SP 800-145) US National Institute of Standards and Technology · September 2011
  2. Q2 Cloud Market Passes $143 Billion; Highest Growth Rate in Eight Years Synergy Research Group · 30 July 2026
  3. The new era of cloud: what 2026 data tells us about spend, scale and strategy Flexera · 2026
  4. Amazon Compute Service Level Agreement Amazon Web Services · accessed October 2026
  5. Annual outage analysis 2026 Uptime Institute · 2026
  6. Bandwidth pricing (data transfer) Microsoft Azure · accessed October 2026
Keep reading

Related insights