E-commerce Hosting Costs: AWS vs Azure vs GCP (2026)

E-commerce cloud bills have a different cost profile from SaaS or internal tooling. Six services dominate the spend, and most of them are either bursty or continuously running at a size that was chosen for peak — not for normal operating load.

1. Compute (application and web tier)

Application servers handling product pages, cart sessions, and checkout represent the largest single line item — typically 35–45% of a self-hosted e-commerce cloud bill. The problem is sizing: teams provision for Black Friday and leave that capacity running year-round. An auto-scaling group capped at 20 instances for a November sale often idles at 4 instances in January but retains the minimum instance count configured during the event. Every extra always-on instance costs $70–$200/month depending on class.

2. Database (RDS / Aurora / Cloud SQL)

Checkout requires write-consistency and low latency — so databases are sized conservatively. An RDS Aurora db.r6g.2xlarge that handles Black Friday transaction throughput costs $820/month on-demand (AWS Aurora Pricing, 2026). The same workload during off-peak months runs fine on a db.r6g.large at $194/month. That $626/month delta sits idle for 10 months of the year in most stores. Database right-sizing — using RDS read replicas for reporting and scheduled scaling for predictable peaks — is the single highest-dollar-value optimization for mid-size e-commerce.

3. CDN and egress

Product images, CSS, JavaScript, and video content move through a CDN on every page load. CloudFront charges $0.085/GB for the first 10 TB of data transfer (AWS CloudFront Pricing, 2026). A fashion retailer serving 50 TB/month pays $4,250 in CDN data transfer before accounting for cache misses that hit the origin at full EC2 egress rates. For a deep dive on how egress charges stack up across providers, see our cloud egress costs guide. Cache hit rate improvement — from a typical 70% to 90%+ — cuts CDN spend by 20–30% with no code changes, just TTL and cache-key tuning.

4. Search (OpenSearch / Elasticsearch / Algolia)

Product search is real-time and latency-sensitive. Self-hosted OpenSearch on AWS (r6g.2xlarge.search, 2 nodes) costs $580–$720/month running continuously. Teams size search clusters for peak catalog load and query concurrency — then leave them running at that size permanently. Algolia or Typesense SaaS alternatives cost $500–$1,500/month at mid-scale but eliminate the operational overhead of cluster management and automatically handle indexing spikes.

5. Image processing and storage

A mid-size catalog with 15,000 SKUs and 8 images per product at 3 resolutions means 360,000 image variants. S3 storage at $0.023/GB is inexpensive, but origin image transformation requests (via Lambda or a dedicated service like Imgix) add $80–$200/month in processing costs. Converting source images to WebP at upload time eliminates transformation overhead and reduces storage and CDN transfer by 55–65% compared to JPEG at equivalent visual quality (Google WebP Study).

6. Payment processing infrastructure

PCI DSS compliance typically requires a dedicated subnet, WAF, enhanced logging, and sometimes a separate compute tier for payment flows. AWS WAF costs $5/rule group per month plus $0.60 per million requests. A store processing 500K checkout attempts/month pays $8–$15/month for WAF alone — small, but the associated NAT gateway, VPC endpoints, and dedicated RDS instance for transaction logs can add $150–$300/month in compliance-driven infrastructure that's easy to overlook during cost reviews.

Infrastructure cost as a percentage of GMV compresses as revenue scales — but absolute spend grows faster than most founders expect between $100K and $1M GMV. Below are representative self-hosted AWS cost ranges by GMV tier, based on typical WooCommerce, Magento, and headless Next.js stacks. Use the e-commerce hosting cost calculator to model your specific stack.

$10K–$50K GMV: shared or small dedicated

At this tier, a single t3.medium or t3.large web server ($30–$65/month), a db.t3.medium RDS instance ($55/month), and CloudFront ($10–$40/month) covers most workloads. Total: $200–$300/month. Shopify Basic ($39/month) or WooCommerce on managed WordPress ($20–$60/month) will almost always be cheaper than self-hosted AWS at this GMV — the engineering overhead alone exceeds the infrastructure savings.

$100K–$500K GMV: mid-tier dedicated

At $100K–$500K GMV, typical infrastructure includes 2–4 c6g.xlarge application nodes ($120–$240/month), a db.r6g.large RDS primary plus read replica ($400–$500/month), ElastiCache r6g.large ($150/month), and CloudFront serving 5–15 TB/month ($425–$1,275/month). Total: $1,400–$1,500/month. This is the range where the Shopify vs self-hosted decision gets genuinely complex — see the section below.

$1M–$10M GMV: enterprise-grade dedicated

Above $1M GMV, stacks typically add OpenSearch or Algolia for search, SageMaker or a third-party recommendation engine, multi-AZ database configurations, WAF and Shield Standard, and dedicated DevOps capacity. Infrastructure spend in this range runs $3,500–$13,000/month depending on catalog size, media volume, and traffic geography. At this scale, the savings from Savings Plans, right-sizing, and CDN optimization are worth six figures annually — modeling them with the e-commerce cloud cost calculator takes 10 minutes and typically surfaces $500–$2,000/month in recoverable spend.

Shopify Plus starts at $2,300/month (Shopify, 2026) and includes hosting, CDN, SSL, PCI compliance, checkout, and a 0.2% transaction fee above $500K GMV/year (waived with Shopify Payments). Self-hosted stacks on AWS or Azure carry no platform fee but require infrastructure spend, engineering time, and operational overhead. The break-even point depends heavily on how you value engineering time.

The math shifts above $10M GMV where Shopify Plus transaction processing costs (even at 0%) and platform limitations in catalog complexity, checkout customization, and international market handling push large retailers toward Commercetools, SFCC, or fully custom stacks on AWS or GCP. The e-commerce cloud cost management guide covers the cost implications of headless vs monolithic architectures in more detail.

What Shopify Plus includes that self-hosted doesn't

• PCI DSS Level 1 compliance (saves $10K–$40K/year in audit and tooling)
• Global CDN with 200+ edge locations (replaces $500–$2,000/month in CDN spend)
• DDoS mitigation and WAF (replaces AWS Shield Advanced at $3,000/month)
• Shopify Payments with no transaction fee (saves 0.5–2% on payment processing)
• 99.99% uptime SLA with no engineering effort required

Below $2M GMV, Shopify Plus almost always wins on total cost of ownership when engineering time is included. Above $5M GMV, self-hosted becomes cost-competitive only if the team already has cloud infrastructure expertise and the workload justifies Savings Plans at scale.

All three major clouds support production e-commerce workloads. The differences that matter for online retail are managed database pricing, CDN integration, search service availability, and the ecosystem of third-party integrations. Here's how they compare on the services e-commerce teams actually use.

AWS: best ecosystem, highest default cost

AWS has the broadest coverage of services used in e-commerce: CloudFront, ElastiCache, OpenSearch, SageMaker for recommendations, Aurora Serverless v2 for variable checkout load, and the deepest integration with Shopify, Magento, and WooCommerce plugins. On-demand pricing is the highest of the three major clouds, but Savings Plans reduce compute costs by 30–37% and Spot instances reduce batch costs by 70–90%. At $1M+ GMV, most AWS e-commerce stacks run on Savings Plans for the always-on tier and spot for batch — effective pricing drops below Azure list rates.

Azure: lower list prices, strong for Microsoft-adjacent stacks

Azure compute list prices are 10–30% cheaper than AWS for equivalent instance classes (D-series vs c6g), and Azure SQL Hyperscale offers auto-scaling storage that eliminates database provisioning decisions. Azure is the natural choice for stores running on .NET, Dynamics 365 Commerce, or with existing Microsoft enterprise agreements that include Azure credits. Azure Hybrid Benefit extends Windows Server licenses to the cloud, saving an additional 40–50% on Windows-based application tiers.

GCP: cheapest CDN, best for data-intensive workloads

Google Cloud CDN charges $0.08/GB for the first 10 TB — slightly cheaper than CloudFront and Azure CDN. Sustained use discounts apply automatically to Compute Engine without reservation commitments, which suits e-commerce's variable load pattern. BigQuery and Looker integrations make GCP the strongest choice for stores with significant analytics and personalization workloads (product recommendation models, A/B testing infrastructure, customer lifetime value analysis). GCP's e-commerce ecosystem support (third-party plugins, Magento hosting guides, community tooling) is thinner than AWS.

Recommendation by stack

• WooCommerce / PHP stack: AWS (deepest plugin ecosystem) or Azure (lower compute cost)
• Headless Next.js + commerce API: AWS (Vercel + AWS backend) or GCP (Firebase + GCP backend)
• .NET / Dynamics Commerce: Azure (native integration, Hybrid Benefit)
• Data-heavy personalization: GCP (BigQuery, Vertex AI, Looker)
• Multi-cloud / vendor-neutral: Cloudflare Workers + any provider for edge; AWS for primary

Black Friday is the most expensive single event in the e-commerce cloud calendar — and the most common source of post-event billing surprises. Traffic spikes of 5–10x normal volume arrive over 4–6 hours, auto-scaling triggers at the compute and database layers, CDN origin pull rates spike as cache warm-up lags behind traffic, and WAF rule evaluation at 10x request rate generates 10x the WAF cost. AWS data shows Black Friday can deliver 8x normal order volume for mid-size retailers (AWS Retail, 2025).

The scale-in problem

Auto-scaling groups scale out quickly — that's the point. But scale-in is intentionally conservative: AWS default cooldown periods are 300 seconds, and teams often set scale-in policies with long evaluation windows to avoid thrashing. After a Black Friday event, it's common to find auto-scaling groups sitting at 12–16 instances for 3–4 days when normal load needs 4. At $0.17/hr per c6g.xlarge, 12 extra instances for 72 hours costs $148 — small per-instance, but multiplied across app tier, worker tier, and potentially a second region it reaches $500–$1,500 in unnecessary post-event spend.

Pre-warming strategy

The alternative to reactive auto-scaling for a known traffic event is pre-warming: scaling out manually 2–4 hours before the sale, running at full capacity for the event window, then aggressively scaling in immediately after. Pre-warming eliminates the risk of scale-out lag (instances take 3–5 minutes to be healthy in a target group) and gives you a defined end time to scale back from. For database pre-warming, enable Aurora Auto Scaling with a maximum capacity set for the event and a scheduled action to reduce it 6 hours after the sale ends.

CDN pre-warming for large catalogs

CloudFront cache cold-start during a high-traffic event means the first wave of requests hits the origin at full rate — potentially overloading application servers before auto-scaling has responded. Pre-warm by crawling your top 1,000 product pages through the CDN origin URL 30 minutes before the sale. CloudFront supports cache warming for custom origins; Fastly and Cloudflare both have explicit pre-warming APIs. A warmed cache at the start of a Black Friday event reduces origin load by 60–70% during the critical first 15 minutes.

The highest-ROI e-commerce cloud optimizations require no downtime, no architecture changes, and deliver results within the first billing cycle. Start with the items that apply to your current GMV tier — then use the e-commerce cost calculator to quantify the savings before committing engineering time. For a broader look at how e-commerce cloud cost management compares to other industries, see cloud cost management for e-commerce.

Compute Savings Plans

Every e-commerce store has always-on compute: web servers for availability, a database primary for writes, a cache layer for sessions. Cover that baseline with a 1-year Compute Savings Plan — which applies to any EC2 instance family, size, and region — and save 30–37% on covered compute with zero architectural changes. No-upfront plans require no capital commitment; savings begin in month one. At a $1,000/month compute baseline, a Savings Plan saves $300–$370/month.

Database right-sizing between seasons

The correct database instance class for January is not the correct class for November. Schedule a database resize to a smaller instance class each January and a resize back up in October. RDS supports Multi-AZ instance modification with minimal downtime (2–5 minutes for a failover). Going from db.r6g.2xlarge to db.r6g.large saves $626/month and takes 20 minutes of engineering time. Repeat this each year and capture $5,000–$7,000 in annual database savings.

CDN cache rate improvement

Most e-commerce CDN configurations have a cache hit rate of 65–80%. Tuning TTLs, removing non-essential query string parameters from cache keys, and addingVary: Accept-Encodingheaders to static assets pushes hit rate to 88–94%. Each percentage point of hit rate improvement reduces origin pull volume — and both CDN transfer costs and origin compute load — proportionally. A store serving 20 TB/month through CloudFront saves $17/month per percentage point of cache improvement. Going from 75% to 90% saves $255/month.

Spot for batch and background

Order export jobs, inventory sync, search index rebuilds, catalog feed generation for Google Shopping, and email list processing are all interruptible. Move them to AWS Spot or Azure Spot VMs with SQS retry queues and save 70–90% on that compute. AWS Batch and ECS with Spot capacity providers handle interruption retry automatically. A background worker fleet that costs $400/month on-demand costs $40–$80/month on Spot with equivalent throughput.

Egress and CDN architecture

Moving image origins to CloudFront with S3 as the origin (instead of EC2 or a separate image server) eliminates EC2 egress charges ($0.09/GB) in favor of S3-to-CloudFront transfer, which is free. CloudFront-to-internet egress at $0.085/GB is cheaper than direct EC2 egress. For cross-region replication or multi-CDN setups, the cloud egress costs guide covers how to minimize data transfer fees across AWS, Azure, and GCP.