AWS vs Azure vs GCP: Cost-optimization playbook for mid-size e-commerce platforms

TL;DR: AWS, Azure, and Google Cloud rarely have a single cost winner. Each is cheaper on a different line item. A mid-size e-commerce platform's bill is usually shaped by four cost categories: compute, storage, database, and data transfer, all of which can be temporarily inflated by traffic spikes during sales events. Egress rates can vary by roughly 30% between the cheapest and most expensive providers at the same volume, and rightsizing overprovisioned instances can reduce compute spend by 20 to 40% on its own, before any provider-switching decision comes into play. The real question isn't which cloud is cheapest. It's which cost levers apply to your specific traffic pattern. 

Key terms 

• Rightsizing: matching a cloud instance's size to its actual usage rather than its provisioned capacity, usually the fastest way to cut compute spend. 

• FinOps: the cloud financial management discipline that brings engineering, finance, and business teams together to manage cloud spend as a shared, ongoing practice rather than a one-time audit. 

• Committed-use discount: a discount earned by committing to a set level of cloud spend or resource usage for a fixed term, one or three years, in exchange for a lower rate than on-demand pricing. 

• Sustained-use discount: an automatic discount, unique to Google Cloud, applied to compute usage that runs for a large share of a billing month, with no upfront commitment required. 

• Egress (data transfer): the cost of moving data out of a cloud provider's network to the internet or to another region, often the most overlooked line item on an e-commerce cloud services bill. 

AWS vs Azure vs Google Cloud is the wrong comparison if the question is "which one is cheapest." Each provider wins on a different axis: AWS on ecosystem maturity and third-party tooling, Azure on the lowest headline egress rate, and Google Cloud on automatic discounts that need no upfront commitment. A mid-size e-commerce platform's costs depend on more than one of these at once, which is why the right question is which provider's cost structure fits your traffic pattern.  

In this piece, we cover what drives cost for an e-commerce workload, how the three providers price and discount compute differently, and where teams leave money on the table before they've even considered switching providers. 

What drives cloud costs for a mid-size e-commerce platform? 

Teams usually monitor compute closely, but there are four categories that have an impact on an e-commerce cloud resources bill:  

1. compute for the application and checkout logic 

2. storage for product images and media 

3. the database layer 

4. data transfer out to customers and content delivery networks (CDNs) 

Traffic spikes, the kind that come with flash sales, Black Friday, or a marketing push, temporarily inflate these four categories at once. This is why a platform that is sized for its average month gets a painful surprise during its busiest week.  

Architecture is also important: a monolith must be overprovisioned to survive a spike, while a platform built on independent services can scale up only the components under load and scale back down once the spike passes. 

How do AWS, Azure, and Google Cloud price compute differently? 

All three providers charge on-demand rates by default and offer steep discounts for predictable cloud usage, but the mechanics differ enough to change which one fits a given workload. 

Sources: AWS Cloud Financial Management, Azure Reserved VM Instances, Google Cloud Committed Use Discount Recommender, current as of publication. 

Google Cloud's Sustained Use Discounts apply automatically once a workload runs long enough in a month with no purchase, no term commitment, and no risk of overcommitting. AWS and Azure only discount steady usage if a team proactively buys a commitment, which saves more at the deepest tiers but carries real risk if usage drops mid-term. 

Spot Instances and Spot VMs sell spare provider capacity at deep discounts, which the provider can later reclaim with little notice. They are a good fit for stateless, fault-tolerant workloads like batch image processing or CI pipelines, and a poor fit for anything customer-facing, like checkout, that can't tolerate unplanned interruptions. 

Which provider offers the cheapest e-commerce data transfer and bandwidth? 

Based on  current published pricing models, cross-verified across multiple independent sources, on default settings, Azure is the cheapest of the three for egress. It charges $0.087 per GB for the first tier of internet traffic, just ahead of AWS at $0.09 per GB. Google Cloud's default Premium Tier routing is as high as $0.12 per GB, but that comparison overlooks an option most cost analyses skip: Google Cloud's Standard Tier routing drops the rate to roughly $0.085 per GB, undercutting both AWS and Azure. 

Azure has one further advantage regardless of tier: it removed cross-availability-zone transfer fees in 2024, a meaningful saving for any multi-zone, high-availability architecture, while AWS and Google Cloud both still charge for that traffic. 

There's a bigger shift underway that changes how much any of this should weigh in a provider decision. AWS, Azure, and Google Cloud all began waiving egress fees for customers fully exiting their platform starting in 2024, though the waivers are exit-only, require case-by-case approval, and exclude some services. The EU Data Act goes further: starting January 12, 2027, following a transition period that began in September 2025, switching and exit egress charges will be banned outright for providers serving EU customers. None of this touches the day-to-day egress rates compared above, those remain in effect for ongoing cloud operations, but it does mean the traditional argument for tolerating a worse-fit provider because leaving is expensive is getting weaker by regulation, not just by choice. 

What is rightsizing, and how much can it save? 

Rightsizing means matching an instance's provisioned size to its usage, measured by real CPU, memory, and network metrics rather than the size a team guessed at during initial deployment. It's consistently the highest-impact, lowest-effort cost lever available, ahead of any committed-discount purchase or provider switch. 

Most teams badly overprovision their instances, and cost management analyses commonly put righsizing’s savings in the 20 to 40% range. AWS, Azure, and Google Cloud all include free native tools like AWS Compute Optimizer, Azure Advisor, and GCP Recommender that analyze usage and give you actionable downsizing recommendations. You don't need paid software to start, but you do need to make rightsizing an ongoing habit: as your store grows and traffic shifts, your server needs will change, too. 

How do storage costs and database costs behave for e-commerce? 

Storage and database costs behave differently from cloud computing costs because they accumulate continuously rather than being tied to active server instances.  

Storage costs tend to be predictable because media assets like product photos, videos, and catalog imagery rarely get deleted. As the catalog grows, most teams leave years of it sitting in the default, most expensive storage tier. Cloud service providers like AWS S3, Azure Blob, and Google Cloud Storage offer multi-tiered storage options ranging from high-performance Standard tiers to low-cost Archive tiers. All three support lifecycle policies that move objects to a cheaper tier automatically once they cross an age or access threshold. A product photo for a discontinued item doesn't need the same tier as this week's homepage banner, but without a lifecycle policy, it usually gets billed as if it does. 

Database costs behave differently, mostly driven by workload predictability and read-to-write ratios. AWS Aurora, Azure Database, and Google Cloud SQL offer provisioned instances eligible for reserved or committed discounts for steady traffic, especially when combined with multi-year commitments.  Serverless or autoscaling tiers also bill per second and scale to zero when idle, making them ideal for unpredictable or highly seasonal traffic patterns. Additionally, offloading catalog browsing and reporting queries to read replicas prevents traffic spikes from straining the primary database, providing a cost-effective alternative to scaling up the entire instance. 

What cost-optimization tools should e-commerce teams use? 

Start with each provider's native tool before adding anything else. They're free, already included, and cover the highest-impact levers, rightsizing, idle-resource detection, and commitment recommendations, without a new vendor relationship. AWS Cost Explorer and Trusted Advisor, Azure Cost Management and Advisor, and Google Cloud's Billing reports and Recommender all do this out of the box. 

Native tools run out of depth at business-level cost attribution: tying spend to a specific product line, team, or customer, not just a service or resource. That's when third-party platforms and a formal FinOps practice start to earn their cost, especially for a platform running workloads across more than one provider. 

How does each provider discount for predictable usage? 

Once a workload's baseline is well understood (roughly six months of stable production traffic is usually enough), the next real savings come from committing to a discount program.  

AWS Savings Plans apply across instance families and services, so you can shift usage freely. Reserved Instances lock in a specific instance type and region for a deeper discount, giving up that freedom in return. Azure Reservations work similarly, and stack with Azure Hybrid Benefit if a team already owns Windows Server or SQL Server licenses, pushing combined savings toward 80% in the right scenario. Google Cloud's Committed Use Discounts require the same kind of term commitment, but its automatic Sustained Use Discounts mean a workload earns some savings even before that commitment gets made. 

All three providers punish the same mistake: locking into a term based on today's traffic without planning for what happens if usage drops. A commitment may look efficient at the beginning, but can become a liability if workloads decrease. None of the three providers lets a team walk away from an active term without a real cost. 

How should a mid-size e-commerce platform run a cost audit? 

Here is a straightforward way to approach it: 

1. Get a baseline of 30 days of usage before buying any discounts. Look at compute, storage, and data transfer separately. A full month of real data will show you idle resources and obvious rightsizing candidates right away. If you buy multi-year commitments on inflated or temporary usage, you end up locked into overpaying for years. 

2. Separate your baseline traffic from traffic spikes. Normal daily traffic and Black Friday peaks can look completely different. Size your committed-use discounts for your day-to-day baseline, and rely on autoscaling or spot capacity to handle temporary traffic surges. 

3. Look closely at where your data transfer goes, not just how much you use. Traffic going to a CDN, moving between availability zones, or heading straight out to the public internet all carry different costs and need different fixes. Lumping it all under one "data transfer" line makes it hard to see what to optimize. 

4. Schedule regular reviews instead of treating cost optimization as a one-off task. As your product catalog grows and traffic patterns shift, your resource needs change too. If you only look at costs during initial setup or migration, those initial cost savings will fade before you know it. 

Why cost alone won’t pick your cloud provider 

The honest answer is that no single provider wins outright. AWS offers the deepest third-party tooling ecosystem, Azure the lowest headline egress rate and no cross-zone fees, and Google Cloud the only discount that requires no commitment at all. A mid-size e-commerce platform that chooses a provider based solely on sticker price, without accounting for rightsizing discipline, egress patterns, and commitment timing, will overspend regardless of which logo appears on the invoice. 

That's a heavier analysis than most teams have time to run alongside shipping products. Svitla works with e-commerce platforms on this kind of cloud cost assessment, from architecture review through provider migration and ongoing FinOps practice.