How Much Does Cloud Computing Cost?

Published on October 16, 2025 | Prices Last Reviewed for Freshness: November 2025
Written by Alec Pow - Economic & Pricing Investigator | Content Reviewed by CFA Alexander Popinker

Educational content; not financial advice. Prices are estimates; confirm current rates, fees, taxes, and terms with providers or official sources.

Cloud computing turns infrastructure into a utility that you rent by the unit, then release when finished, and the bill reflects exactly what you consumed across compute, storage, and data movement. Prices vary by service model and region, and your effective rate depends on commitments, credits, and optimization discipline. NIST’s well-known definition anchors the basics across IaaS, PaaS, and FaaS (NIST SP 800-145).

Three levers drive nearly every cloud bill, compute instances and serverless runtimes, storage classes for hot or cold data, and network charges for data egress and cross-region hops. Billing units show up as per hour or per second for vCPU and RAM, per GB-month for object and block storage, and per GB for transfer to the internet, while managed databases, load balancers, and Kubernetes control planes add their own line items. FinOps practices exist to tame this variability by tying usage to business value with tagging, budget alerts, and accountability.

Article Insights

• Expect object storage around $0.018–$0.023 per GB-month, egress around $0.085–$0.12 per GB, and Lambda requests $0.20 per million before compute time (see Cloudflare’s egress discussions).
• Commitments cut rates—up to 65–72% for steady compute—and spot saves 60–90% when interruptions are acceptable (AWS Savings/Reserved pricing).
• Most surprise comes from bytes leaving—design around egress and inter-region paths to protect your budget (GCP network pricing).
• Observability and support uplift the bill, with examples like $15–$40 per host per month for monitoring suites and $100 support minimums (Datadog pricing).
• Credits and private offers can offset five to six figures, especially for startups in formal programs (AWS startup credits).

How Much Does Cloud Computing Cost?

Cloud computing costs vary by the level you choose—from ~$0.018 per GB-month for object storage to $0.20 per million for Lambda requests—before compute time.

List prices are public and predictable, though your effective price often ends up lower when you combine savings plans, reserved capacity, spot markets, startup credits, and marketplace discounts. In the United States, common list rates as of October 2025 include object storage around $0.02–$0.023 per GB-month, data egress around $0.085–$0.12 per GB for first-tier volumes, and serverless requests at $0.20 per million plus compute time, with regional surcharges outside North America and Europe (see this overview of hidden costs).

On-demand gives flexibility at the highest rate, reserved and savings plans trade commitment for discounts up to 72%, and spot or preemptible capacity can deliver 60–90% savings for tolerant workloads. Multi-region high availability usually increases spend, both for duplicated storage classes and for inter-region data transfer (AWS EC2 pricing).

According to Orient Software, pricing components include storage, computing power, network usage, as well as indirect costs such as exit fees and data overage charges. Different service models like Infrastructure as a Service (IaaS), Platform as a Service (PaaS), and Software as a Service (SaaS) each have distinct pricing structures, impacting overall expenses. Companies should carefully evaluate base rates against hidden fees and usage fluctuations to optimize cloud spending (cloud cost breakdown).

According to research highlighted by CloudZero, enterprises with over 1,000 employees typically spend between $2.4 million and $6 million annually on cloud services, accounting for about 19% of total IT budgets. Gartner forecasts public cloud spending to reach $723 billion globally in 2025, up from $595 billion in 2024, with significant growth projected in IaaS and PaaS segments. Despite this, companies often struggle with underutilized infrastructure and waste around 21%, underscoring the importance of cost optimization (CloudZero guide; US Cloud).

Cloud pricing varies among major providers like AWS, Microsoft Azure, Google Cloud, and others. Pricing depends on compute hours, storage capacity, data transfer, and other service-specific factors. For example, Azure users may experience rapid cost growth without optimization. As cloud adoption increases, hybrid strategies are becoming prevalent, further complicating cost structures (Sedai; CloudZero snapshot).

Real-Life Cost Examples

Example A, small web app on serverless. A static front end on a CDN, an API on Lambda, data in S3 and a lightweight Cloudflare Worker for an edge route. At 2 million monthly requests and 200 GB of egress in US regions, the ballpark monthly total is $40–$70, combining $0.20 per million Lambda requests plus GB-seconds, $0.023 per GB-month for S3 storage on 100 GB, and CDN egress around $0.085 per GB after the free first terabyte on CloudFront, while Cloudflare Workers paid plan starts at $5 with no egress fees.

Example B, SaaS on autoscaled VMs with managed Postgres. A three-tier app on AWS with two t3.medium app instances averaging 60% utilization, one RDS db.t3.medium, an Application Load Balancer, 500 GB in S3, 1 TB egress, and daily snapshots. A realistic monthly total lands around $350–$550, driven by $0.0416 per hour per t3.medium, RDS rates for similar instance classes, S3 storage at $0.023 per GB-month, and $0.09 per GB for the first 1 TB of egress, before any savings plans or credits.

Example C, analytics stack with warehouse and BI. Teams that stream to object storage, transform on Databricks or a similar engine, and query a warehouse like Snowflake tend to feel egress. With 20 TB scanned monthly, 5 TB egress to BI, and elastic clusters scheduled for 200 DBU or credits per day, monthly totals often range $6,000–$12,000, depending on negotiated DBU or credit rates and cache hit ratios, plus $0.085–$0.12 per GB of egress from CDN or origin (RDS pricing context).

Also read our articles on the cost of Azure, AWS servers, or Adobe Marketing Cloud.

Example D, video platform with multi-region DR. Object storage of 50 TB hot, 2 PB archive, 100 TB monthly egress across North America and Europe, CloudFront delivery, transcode functions, and warm secondary in a second region. Expect $12,000–$20,000 monthly at list rates, skewed by egress at $0.085 per GB for first-tier CDN traffic, then storage at $0.023 per GB-month hot and much lower for archive classes, plus inter-region replication fees (CloudFront pricing noted above).

Cost Breakdown

Compute is usually the first lever. VM families price by vCPU and RAM per unit time, GPUs carry large premiums, and sustained-use or committed-use programs cut rates materially. Serverless compute charges per request and GB-second and shifts you to a spend pattern that rises linearly with traffic. Containers under managed Kubernetes add cluster control-plane fees in some clouds (e.g., EKS/GKE), while Azure AKS does not charge for the control plane (AWS networking/pricing).

Storage splits into object, block, and file. Object storage is cheapest for most workloads, with hot classes around $0.018–$0.023 per GB-month in major US regions and lower tiers for archive that add retrieval costs. Block storage rides with VMs, priced by provisioned IOPS and GB-month. File services land in between, often premium for performance. Lifecycle rules reduce the bill when you demote stale data automatically.

Data movement is the silent cost center. Internet egress commonly starts near $0.085–$0.12 per GB for entry tiers by region, inter-region transfer is also billed with matrices published by each provider, and in some clouds cross-AZ traffic is $0.01 per GB inside a region. Architect for proximity, use CDNs, and measure before it stings.

Managed services add comfort and line items. Databases charge hourly for instance size, storage, and backup; message queues and observability have per-request or per-GB models; and support plans can add percentage of spend or monthly minimums such as $100 for AWS Business Support.

Factors Influencing the Cost

Architecture choices change the slope of your spend. Stateless services scale horizontally with autoscaling and spot markets; stateful systems lean on larger instances and tuned storage; cache layers can shrink both compute and egress. Kubernetes helps bin-pack but introduces cluster fees and observability overhead that should be budgeted explicitly (AWS VPC pricing).

Utilization patterns matter a lot. Peaky workloads love serverless or autoscaling VMs; steady batch jobs often win with commitments; idle resources burn money without delivering value. Governance maturity is the multiplier—tagging, showback/chargeback, and anomaly detection make cost a team sport rather than a monthly fire drill (FinOps basics).

Alternative Products or Services

Hyperscalers are the default for breadth and regions, yet many web apps get better unit economics on managed platforms. Cloudflare Workers paid plan starts at $5/month with no egress fees; Vercel Pro includes generous bundled data transfer and edge requests; Netlify has credit-based plans that bundle bandwidth and functions.

Data platforms and DIY paths trade control and price dynamics. Snowflake and Databricks meter by credits/DBUs and can be cost-effective with disciplined scheduling and caching, while a DIY Kubernetes + OSS stack often shifts money into operations and observability. Edge/CDN offload reduces origin egress and tail latency for global audiences. On-prem or colo can still win for steady, high baseline workloads when amortized over years.

Ways to Spend Less

Commit where it makes sense. AWS Savings Plans and Azure reservations can shave up to 65–72% for steady usage, and GCP committed use discounts reduce vCPU/RAM unit prices for one- or three-year terms. For interrupt-tolerant queues, workers, and analytics bursts, spot instances can cut 60–90% off list (EC2 pricing).

Engineer for efficiency. Right-size instances, enable autoscaling with sensible floors/ceilings, push bytes to CDNs to trim egress, compress payloads, collocate compute with data to avoid paid cross-region paths, and tier storage with lifecycle policies. Delete stale snapshots/logs, and watch log ingestion because per-GB rates scale fast (CDN pricing).

Expert Insights & Tips

Practice FinOps like a product—cost allocation tags on every resource, budget alerts per team, anomaly detection, and weekly reviews that connect spend to value delivered (FinOps Framework). Predictability improves with demand forecasting and capacity modeling; run price simulations using vendor calculators and your negotiated rates. Observability isn’t free—budget for it thoughtfully (FinOps updates).

Total Costs

Think in multi-year terms with both opex and the capex you avoided. One-year projections for steady services should include cloud spend, support tiers, observability, and ops labor, while three- and five-year views show the benefit of commitments, refactors, and reserved capacity (Forrester TEI (AWS CloudOps)). Security and observability tooling often sit outside the main cloud bill but influence TCO (Forrester on Cloud Run).

Hidden & Unexpected Costs

Many teams first feel surprise in networking. NAT gateways charge hourly and per GB; inter-region and cross-AZ transfers are billed; and public IPs or global accelerators may appear when architects move fast. CDN cache fill and invalidation, warehouse query metering, and log retention/rehydration can also grow quickly without caps (SLA considerations).

Add smaller incidentals to every estimate. Support plan uplifts like $100/month minimums, TLS certificate or DNS features from third parties, and marketplace fees sit outside pure compute, storage, and network.

Warranty, Support & Insurance Costs

Support plans buy response times backed by SLAs and credits for downtime. AWS, Azure, and Google publish SLA credit tables; while credits reduce future invoices, they don’t cover consequential loss, so some enterprises buy separate downtime insurance. Disaster-recovery drills incur testing hours and transient egress for failover (Committed-use docs). Compliance adds logging, key management, and audit artifacts, which shows up as more storage and higher support tiers (CloudWatch pricing).

Financing & Payment Options

Commitments and credits change cash flow and unit economics. Startups can secure large credit packages if eligible; enterprises often prepay credits under agreements that include private offers and negotiated rate cards. Marketplace private offers can bundle software and infrastructure with discounts, and multi-year price protections lock in rates for predictable growth (AWS Marketplace study).

Opportunity Cost & ROI

Speed to market and flexibility carry value. Analyses from Forrester/IDC (often vendor-commissioned) set expectations for where benefits show up—reduced downtime, faster deployments, developer productivity. Treat them as frameworks to build your own models.

Seasonal & Market-Timing Factors

Discount windows exist. Fiscal year ends, new region launches, and startup-program refreshes can coincide with better credit availability or private-offer terms—as seen in expanded credit pools for AI startups (Reuters). Time large migrations to align with these moments.

Answers to Common Questions

How much should a small startup budget monthly?
Many early products that lean on serverless and a CDN see $30–$150 at light traffic, then scale with requests and egress, with credits covering much of the first year if eligible (Lambda pricing).

Why is data egress expensive and how can I reduce it?
Egress fees reflect network delivery costs and vary by destination and region; first-tier US rates often start near $0.085–$0.12 per GB. Reduce by caching at the edge, compressing assets, minimizing cross-region chatter, and keeping compute near data (CloudFront pricing).

On-demand vs. reserved vs. spot—what’s the trade-off?
On-demand is flexible but highest price; reserved/savings plans exchange commitment for up to 72% off; spot instances deliver the lowest rates with termination risk—best for batch, stateless workers, and CI (EC2 pricing).

Is serverless cheaper than VMs for APIs?
At low-to-moderate, spiky traffic, serverless often wins because you pay only for requests and GB-seconds; VMs become more efficient once steady utilization passes a threshold. Always compare using your traffic profile (Lambda pricing).

How do I project next year’s annual budget accurately?
Adopt FinOps practices, tag every resource, export billing data, run price simulations with calculators, and track forecast variance monthly; adjust commitments incrementally rather than annually (FinOps Framework).