Tell Me Why

Cloud Server Pricing Explained: What a Small Business Actually Pays

The number next to a cloud server rarely matches the number on the invoice. A small-business owner shopping for a virtual machine sees an appealing hourly rate — a few cents an hour, thirty dollars a month — and assumes that figure represents the cost of running their application in the cloud. It doesn't. It represents the cost of the compute instance sitting idle, with no storage attached, no data leaving the network, no backups, no static IP address, and no database. Everything a real application actually needs sits outside that headline price.

This isn't a trick, exactly. Cloud providers price compute, storage, networking, and managed services as separate products because customers use them in wildly different combinations. But the practical effect is that a business owner comparing "$0.04 an hour" across three providers is comparing almost nothing, because the number that ends up on the monthly statement depends on decisions made after that comparison: how much storage gets attached, how much traffic the application generates, whether backups are turned on, and whether the workload runs continuously or only part of the time.

This guide walks through how cloud server pricing actually works, what tends to get left out of the sticker price, and how to build a realistic monthly budget before committing to a provider. Prices below were checked against official provider pages and pricing calculators between late August and early September 2026, mostly for US East / us-east-1-equivalent regions, and are shown in US dollars. Cloud pricing changes often — Hetzner alone raised cloud server prices twice in 2026 — so treat every figure here as a snapshot rather than a permanent number, and confirm current rates before budgeting.

What a Cloud Server Actually Is

A cloud server, more precisely called a virtual machine or compute instance, is a slice of a provider's physical hardware that behaves like a dedicated computer: it has a fixed number of virtual CPU cores, an amount of RAM, an operating system, and a network connection. It differs from three things people often confuse it with. Shared hosting puts a business's website on a server shared with many other customers, with no root access and little control. A VPS (virtual private server) is functionally the same thing as a cloud server — the term is mostly used by smaller providers — while a dedicated server hands over an entire physical machine. Serverless computing, offered by all three major providers under names like AWS Lambda or Azure Functions, charges per execution rather than per hour and removes the server from the picture entirely. Managed application hosting (Heroku-style platforms, or AWS Elastic Beanstalk) sits on top of compute instances but bundles deployment and scaling logic into the price.

This article focuses specifically on general-purpose virtual machines — the kind a small business would use to run a website, an internal tool, or a business application — because that's where the gap between advertised and actual pricing is largest.

How Cloud Server Pricing Works

Every major provider prices compute on an hourly (or per-second) basis, then bills monthly for total usage. A server left running continuously accumulates roughly 730 hours in an average month — 24 hours × 365.25 days ÷ 12 — which is the standard figure providers and calculators use for continuous-usage estimates. So a server priced at $0.05 per hour costs about $36.50 a month if it runs all the time, and less if it's stopped overnight or only used for testing.

That $36.50 is compute only. It says nothing about the disk the operating system lives on, the IP address a client's browser connects to, the bandwidth used sending data back to visitors, or the backups that protect the business if something goes wrong. Those are separate line items, and for a typical small-business server, they can add anywhere from 20% to well over 100% on top of the compute cost.

Hourly Versus Monthly Pricing

Providers publish hourly rates because usage is metered by the second or minute in most cases, which matters for workloads that don't run continuously — a staging server spun up for testing, a batch job that runs for six hours a night. For a server that runs 24/7, the hourly rate times 730 gives a reasonable monthly estimate for compute alone. Providers such as DigitalOcean, Vultr, and Linode also publish flat monthly prices for their standard plans and cap hourly billing so a continuously running instance never exceeds the listed monthly figure — useful for predictable budgeting, since there's no risk of a longer month pushing the bill higher.

Compute Price Versus Total Cost

This is the central distinction the rest of this article works from. The compute price is what a provider charges for CPU and RAM, per hour or per month. The total cost is compute plus storage, plus data transfer, plus any IP address charges, plus backups or snapshots, plus whatever managed services (a database, a load balancer, monitoring) the application needs. On AWS, Azure, and Google Cloud, these are unbundled by default — everything is billed separately, which gives fine-grained control but also fine-grained ways for the bill to grow unnoticed. On DigitalOcean, Linode, Vultr, and Hetzner, storage and a bandwidth allowance are usually bundled into the base plan price, which makes total cost closer to compute price for basic use cases, though extras like backups, load balancers, and managed databases are still billed on top.

A Realistic Small-Business Workload

To make this concrete, picture a small business running a WordPress-style website with moderate traffic — a few thousand visitors a month, some images and PDFs served to visitors, and a MySQL-type database. A server with 2 vCPUs and 4 GB of RAM is a reasonable fit for this workload; a growing application with more concurrent users or background processing might need 4 vCPUs and 8–16 GB. The examples below use these two configurations because they map onto real small-business needs rather than to whatever configuration happens to look cheapest on a pricing page.

AWS Pricing Considerations

On AWS, the closest equivalent to a small-business server is an EC2 instance from the T3 burstable family. As of pricing checked in August 2026 for the us-east-1 (N. Virginia) region, a t3.medium (2 vCPU, 4 GB RAM) runs about $0.0416 per hour, or roughly $30 a month if left running continuously. A t3.xlarge (4 vCPU, 16 GB RAM) runs close to $120 a month on the same basis.

That compute price is the smallest part of what most small businesses will actually pay. Storage on AWS is billed separately through EBS (Elastic Block Store); the standard general-purpose volume type, gp3, costs about $0.08 per GB per month, so a 50 GB system disk adds roughly $4 a month — modest, but not zero, and snapshots for backup purposes are billed on top of that at a similar per-GB rate. Since February 2024, AWS has also charged for every public IPv4 address attached to a running resource — about $0.005 an hour, or roughly $3.65 a month per address — which used to be free for an address attached to a running instance. Outbound data transfer to the internet costs $0.09 per GB for the first 10 TB each month after a 100 GB free allowance, which is trivial for a low-traffic site but adds up quickly for anything serving video, large downloads, or a busy API. AWS also removed its classic always-free tier for new accounts opened after July 15, 2025, replacing it with a $200 credit that expires after six months; accounts opened before that date keep 750 free hours of a small instance for their first year.

Put together, a small WordPress-style site on a t3.medium with 50 GB of storage, one public IP, light backup usage, and modest traffic realistically lands somewhere in the $45–$65 a month range once storage, the IP charge, and backup snapshots are included — noticeably more than the $30 compute number alone suggests.

Azure Pricing Considerations

Microsoft Azure's rough equivalent to AWS's T3 family is the B-series burstable virtual machine. Checked in September 2026 for a US region, a B2s (2 vCPU, 4 GB RAM) runs about $0.0416 per hour — close to $30 a month continuously — while a B2ms (2 vCPU, 8 GB RAM) runs about $0.0832 per hour, and a B4ms (4 vCPU, 16 GB RAM) runs about $0.166 per hour, or roughly $121 a month.

Azure bills storage through Managed Disks, and the price depends heavily on which tier is selected — Standard HDD, Standard SSD, Premium SSD, or the newer Premium SSD v2 — with a typical Standard SSD landing somewhere around $0.05–$0.10 per GB per month depending on size tier and redundancy option (locally redundant storage is cheaper than zone-redundant). Outbound bandwidth on Azure runs close to $0.087 per GB in the first tier for most regions, slightly under AWS's headline rate. A notable Azure-specific cost is that a public IP address is billed separately and, unlike AWS's flat per-address model, its price depends on whether it's a Basic or Standard SKU and whether it's static or dynamic.

For the same general workload — a 2–4 vCPU server, moderate storage, one public IP, standard traffic — Azure tends to land in a similar range to AWS, generally within 10–15% either way depending on the exact configuration, disk tier, and region chosen. Azure's B-series also supports "credit banking" for CPU bursts, which can make cost slightly more predictable for spiky but generally light workloads.

Google Cloud Pricing Considerations

Google Cloud's general-purpose E2 machine family is the natural comparison point. Checked in August 2026 for the us-central1 (Iowa) region, an e2-medium (2 vCPU, 4 GB RAM) costs about $0.055 per hour, or roughly $40 a month continuously — somewhat higher than the AWS and Azure equivalents at this tier. An e2-standard-4 (4 vCPU, 16 GB RAM) runs about $0.134 per hour, or close to $98 a month.

Google Cloud applies automatic Sustained Use Discounts to E2 and similar instances that run for a large share of the month, without requiring any upfront commitment — a feature AWS and Azure only offer through separate reserved-pricing programs. Persistent disk storage on GCP typically runs from around $0.04 per GB per month for standard disks up to roughly $0.17 for higher-performance SSD-backed persistent disks. Data egress to the internet is priced from $0.12 per GB in the standard tier for most regions — the highest headline egress rate of the three major providers — though Google's Premium Tier network and CDN options can change that calculation for traffic-heavy applications. Google Cloud does not charge separately for a basic ephemeral public IP the way AWS now does, though static external IPs carry a small charge when not attached to a running instance.

For a low-traffic small-business server, GCP's slightly higher compute price is offset somewhat by not paying a separate IP address fee; for a traffic-heavy application, GCP's higher egress rate can outweigh that saving.

DigitalOcean and Other Simpler Providers

DigitalOcean, Linode (rebranded as part of Akamai Cloud after Akamai's 2022 acquisition), Vultr, and Hetzner take a different approach: rather than pricing compute, storage, and bandwidth as three separate line items, they bundle a fixed amount of SSD storage and a monthly bandwidth allowance into a single flat plan price. This makes budgeting considerably more predictable for a small business, even if the underlying hardware isn't always identical to what AWS, Azure, or Google Cloud offer.

DigitalOcean's Basic Droplets start at $4 a month for the smallest configuration; a more realistic small-business tier, the CPU-Optimized 2 vCPU / 4 GB Droplet, runs $42 a month with 25 GB of NVMe storage and 4 TB of outbound transfer included, checked September 2026. A General Purpose 2 vCPU / 8 GB Droplet runs $63 a month with the same transfer allowance. DigitalOcean moved to per-second billing on January 1, 2026, which helps for short-lived workloads. Backups are optional and billed as a percentage of the Droplet's base price — 20% for weekly backups, 30% for daily — and outbound transfer beyond the pooled allowance costs $0.01 per GiB.

Linode's Shared CPU plans, checked September 2026, start at $5 a month for a 1 GB instance and run $24 a month for a 4 GB / 2 vCPU plan with 80 GB of storage and 4 TB of transfer, or $48 a month for an 8 GB / 4 vCPU plan. Dedicated-CPU plans, which guarantee CPU cores rather than sharing them, start around $36–$43 a month for a comparable 4 GB configuration. Object storage, load balancers (NodeBalancers, at $10 a month each), and DNS management are priced or bundled separately, but inbound transfer and DDoS protection are free across all plans.

Vultr's Regular Performance line, checked September 2026, offers a 2 vCPU / 4 GB plan at $20 a month with 80 GB of SSD storage and 3 TB of bandwidth, and a 4 vCPU / 8 GB plan at $40 a month with 160 GB of storage and 4 TB of bandwidth — among the more affordable options for these specs.

Hetzner Cloud, a German provider, has historically undercut every competitor on this list, largely because its European data centers include a strikingly generous 20 TB monthly transfer allowance on most plans. A 2 vCPU / 4 GB CX-series instance in an EU region was priced around €4.49–4.59 a month after an April 2026 price adjustment. Hetzner raised prices again in June 2026, specifically affecting its CPX and CCX lines and its US and Singapore regions more than its European ones, with some US-region CPX instances seeing monthly prices roughly double to triple. Because Hetzner has adjusted pricing twice within 2026 alone, readers should check Hetzner's current pricing page directly rather than relying on any figure printed here, especially for US or Singapore deployments, where prices run meaningfully higher than the same specs in Germany or Finland.

What Changes the Final Monthly Bill

The gap between a provider's advertised compute price and what actually shows up on the invoice comes from a predictable set of categories. None of them is hidden in the sense of being concealed — they're all documented on public pricing pages — but they're easy to underestimate because they don't appear next to the instance price when a business is comparing options.

Storage and Backup Costs

The disk a server's operating system and application live on is billed separately from compute on AWS, Azure, and Google Cloud, typically in the range of $0.04 to $0.15 per GB per month depending on performance tier. Snapshots and backups, which most businesses should be running, add a further charge — either a flat percentage of the server cost (common on DigitalOcean and Linode) or a per-GB rate for the data actually stored (common on the hyperscalers). A business that skips backups to save a few dollars a month is making a cost decision with disproportionate downside risk.

Data Transfer and Egress

Outbound data transfer — anything sent from the server to a visitor's browser, an API client, or another data center — is priced per gigabyte on AWS, Azure, and Google Cloud, generally between $0.09 and $0.12 per GB after a modest free allowance. For a low-traffic brochure site this is nearly irrelevant. For a business serving video, large file downloads, or a busy API, it can become one of the largest line items on the bill. Simpler providers bundle several terabytes of transfer into the base plan price, which is one reason their total cost of ownership often looks more favorable for bandwidth-heavy small-business workloads even when their raw compute specs are less flexible.

Networking and IP Costs

AWS now charges roughly $3.65 a month for every public IPv4 address attached to a running resource, a change introduced in February 2024 that surprises businesses used to the old free model. Azure's public IP pricing depends on SKU and allocation method. Google Cloud doesn't charge for a basic ephemeral IP but does for static ones left unattached. NAT gateways, load balancers, and VPN endpoints — all common in slightly more complex setups — carry their own hourly and per-GB charges on the major clouds.

Operating System and Licensing Costs

Linux instances generally carry no OS licensing fee beyond the compute price. Windows Server instances do, and the license cost is baked into the higher per-hour rate rather than billed separately — a Windows instance typically costs noticeably more per hour than the same hardware running Linux. Some enterprise Linux distributions, including Red Hat Enterprise Linux, also carry a per-vCPU licensing charge on top of the base compute price as of a mid-2024 pricing change on AWS.

Managed Services

A managed database (Amazon RDS, Azure Database, Google Cloud SQL) typically costs more than running the same database software on a general-purpose compute instance, because the provider handles backups, patching, and failover automatically. For a small business without dedicated database administration staff, this premium is often worth paying, but it should be budgeted as a separate line item rather than assumed to be included in server pricing.

Regional Pricing Differences

Prices vary by region on every provider, sometimes by 10–20% or more. US and Western European regions tend to sit at the lower end of a provider's price range; parts of South America, Africa, and some Asia-Pacific regions run noticeably higher. A European customer pricing an AWS or Azure server in Frankfurt or Paris will typically see a slightly different figure than a US customer pricing the same instance type in Virginia — differences driven by regional infrastructure cost and, in the EU, by VAT which is often shown separately from the headline price. Businesses budgeting in euros or pounds should check the specific regional price rather than converting the US dollar figure at the current exchange rate, since providers set regional prices independently.

On-Demand Versus Committed Pricing

Every major provider offers a discount for committing to usage in advance — AWS Reserved Instances and Savings Plans, Azure Reserved VM Instances, Google Cloud Committed Use Discounts — typically cutting the compute price by 30% to 60% for a one- or three-year commitment. Google Cloud additionally applies automatic Sustained Use Discounts without any upfront commitment. For a small business confident its infrastructure needs will stay roughly stable, a one-year commitment on a core production server can meaningfully lower the compute portion of the bill; for a business still finding product-market fit, the flexibility of on-demand pricing is usually worth the premium.

When Cloud Servers Are Cheaper Than Traditional Hosting

Cloud servers tend to make the most financial sense for workloads with variable or unpredictable demand, for businesses that don't want to manage physical hardware, or for anyone who needs to scale up or down quickly. There's no upfront hardware purchase, no need to provision for peak capacity that sits idle most of the time, and no hardware replacement cycle to plan around. A small business testing a new product, running a seasonal spike in traffic, or simply not wanting to think about server maintenance generally comes out ahead with cloud infrastructure, even once storage, bandwidth, and backup costs are added in.

When Cloud Servers May Become Unnecessarily Expensive

The calculation flips for workloads that are large, steady, and predictable over a long period. A business running the same fixed compute load 24/7, year-round, for several years, will often find that a traditional dedicated server or a self-managed rack — or simply a long-term reserved cloud instance — costs less than paying on-demand cloud rates indefinitely. Cloud pricing also tends to become expensive relative to alternatives once a workload requires heavy, sustained data transfer; a business serving large volumes of video or file downloads directly from cloud storage or compute can accumulate an egress bill that dwarfs the compute cost, and may be better served by a CDN-focused or bandwidth-included provider from the outset.

How to Estimate a Realistic Monthly Cloud Budget

A workable estimate starts with the compute instance: multiply the hourly rate by 730 for a continuously running server, or by the actual expected running hours for anything intermittent. To that, add storage at the provider's per-GB rate for the disk size actually needed, not the smallest available size. Add an estimate for monthly outbound data transfer based on expected traffic — providers publish free allowances and per-GB overage rates, and most offer a calculator to model this. Add any public IP address charges. Add a reasonable allowance for backups or snapshots, since skipping this to save money is a false economy for most businesses. Finally, add the cost of any managed service — a database, a load balancer, monitoring — the application actually needs, rather than assuming it's bundled with the compute price. Summing these five categories gives a far more realistic monthly figure than the headline instance price alone.

How Small Businesses Can Reduce Cloud Costs

The most reliable savings come from right-sizing the instance to the actual workload rather than over-provisioning "just in case," and from shutting down non-production servers — staging environments, test instances — outside business hours rather than leaving them running continuously. Choosing a lower-cost storage tier where performance requirements allow, cleaning up old snapshots and unattached disks that continue billing after a server is deleted, and consolidating public IP addresses where multiple aren't actually needed all trim recurring costs with little risk. For workloads with predictable, steady usage, committing to a one-year reserved or committed-use rate on the core production server — while keeping variable or experimental workloads on-demand — captures a meaningful discount without sacrificing flexibility where it matters. Finally, comparing simpler providers such as DigitalOcean, Linode, Vultr, or Hetzner against the major clouds is worth doing for straightforward workloads that don't need the breadth of the AWS, Azure, or Google Cloud ecosystems, since bundled storage and bandwidth on these platforms often produces a lower and more predictable total bill for the same underlying specs.

How to Choose the Right Provider

The choice depends less on which provider is "cheapest" in the abstract and more on what the business actually needs. A team already using other AWS, Microsoft, or Google business tools often benefits from staying within that ecosystem for integration and billing simplicity, even if a competitor's raw compute price is marginally lower. A business whose workload is simple — one server, moderate traffic, no complex networking — is often best served by DigitalOcean, Linode, Vultr, or Hetzner, where the bundled pricing model produces a bill that's both lower and easier to predict. A business anticipating rapid, unpredictable growth, or one that needs a wide range of managed services (advanced databases, machine learning tools, enterprise compliance certifications), generally has more to gain from AWS, Azure, or Google Cloud's breadth, even at a higher baseline cost. European businesses with data residency requirements may find Hetzner's EU-only infrastructure, or the EU regions of the major clouds, relevant to compliance decisions independent of price.

My Comment

The most useful thing a small business can do before choosing a cloud provider is stop treating the hourly instance price as a quote and start treating it as one line in a five-line estimate. Compute, storage, data transfer, IP addresses, and backups are priced independently on every major platform for a structural reason — providers can't predict in advance how much storage or bandwidth a given customer will actually use — but that same structure means the advertised price systematically understates the real cost, and it understates it by different amounts on different platforms depending on how traffic-heavy or storage-heavy the workload is. A bandwidth-light internal tool will often look cheapest on AWS or Azure; a bandwidth-heavy public website will often look cheapest on a provider like Hetzner or DigitalOcean that bundles generous transfer into the base price, even if the raw per-vCPU rate looks less impressive on paper. There is no universal winner, and any comparison that declares one without specifying the workload, the region, and the date should be treated with some skepticism — including comparisons, this one included, that will themselves go stale as providers adjust pricing throughout the year.

Sources

• Amazon EC2 On-Demand Pricing
• Amazon EBS Pricing
• Amazon VPC Pricing (Public IPv4 and data transfer)
• AWS News Blog: New Public IPv4 Address Charge
• Azure Virtual Machines Pricing
• Azure Managed Disks Pricing
• Azure Disk Storage Billing — Microsoft Learn
• Google Cloud Compute Engine Pricing
• DigitalOcean Droplet Pricing
• DigitalOcean Pricing Overview
• Akamai (Linode) Cloud Computing Plans Documentation
• Vultr Regular Performance Compute
• Vultr Pricing
• Hetzner Cloud — Regular Performance VPS
• Hetzner Price Adjustment Notice, 15 June 2026