Cloud, Data & AI
Cutting Cloud Costs Without Slowing Your Site Down
Most cloud waste has nothing to do with the servers your visitors rely on. Here is where a typical AWS, Azure or Google Cloud bill really comes from, and how to cut it while keeping the site as fast as before, or faster.

The cloud bill has crept up month after month, nobody is quite sure which part of it is driving the increase, and the obvious fix of buying smaller servers feels risky because the last time someone tried, the site slowed down. That combination keeps a lot of businesses paying for capacity they do not use. The good news is that most cloud waste has nothing to do with the servers your visitors actually depend on.
Tuning the bill is the last stage of every move we make through our AWS, Azure and Google Cloud migration work, and we also review accounts that have been running for years. Here is where the money usually goes and how we cut it while keeping the site at least as fast as before.
Find Out Where the Money Goes First
Cutting without measuring is how sites get slower. Start by making the bill readable:
- Turn on the cost tools. AWS Cost Explorer, Azure Cost Management and Google Cloud's billing reports all break spending down by service, region and day.
- Tag or label every resource with the application and environment it belongs to, then enable those tags for cost reporting. Without them, production, staging and a forgotten experiment all look the same.
- Rank the line items. In most accounts a short list dominates: compute, managed databases, data transfer, storage and snapshots, load balancers, NAT gateways and logging.
- Record a performance baseline before changing anything: response times at the 95th percentile, Core Web Vitals for key pages, and database load. Every change afterwards is checked against it.
Right-Size Compute on Real Numbers
Servers are usually sized at launch, on a guess, and never revisited. Look at CPU and memory use over several weeks, including your busiest days. An instance that never rises above a small fraction of its CPU can usually drop a size or two with no visible effect.
- Move to current instance generations. Newer families typically give more performance for the same spend, and ARM-based options such as AWS Graviton can be better value again, provided your application and its dependencies build for ARM.
- Scale with demand. If traffic follows office hours or a weekly pattern, autoscaling or scale-to-zero platforms such as Cloud Run or Azure Container Apps stop you paying for peak capacity all night.
- Switch off what nobody uses at night. Development and staging environments rarely need to run at weekends. A schedule that stops them out of hours is one of the safest savings there is.
Commit Only to What You Know You Will Use
All three providers discount steady usage in exchange for a commitment: Savings Plans and Reserved Instances on AWS, reservations and savings plans on Azure, and committed use discounts on Google Cloud, which also applies some sustained-use discounts automatically. These are worthwhile for the baseline that runs all year, but the order matters. Right-size first, then commit, otherwise you lock in a discount on servers that were too big.
Spot or preemptible capacity is much cheaper again, but the provider can reclaim it at short notice. It suits batch jobs, image processing and CI runners. It does not suit your database or the only copy of your web server.
The Cheapest Request Never Reaches Your Servers
This is where saving money and speeding the site up become the same job. Every request served from a cache costs less compute, less database time and often less data transfer than one served by the application.
- Cache at the edge. A CDN such as CloudFront or Cloudflare can serve images, scripts, stylesheets and, for many pages, the full HTML. Set proper cache headers so it is allowed to.
- Shrink what you send. Modern image formats such as WebP and AVIF, images resized to the size they are displayed at, and compressed text all cut transfer costs and load times together.
- Cache inside the application. Full-page caching for anonymous visitors and an object cache such as Redis for repeated queries reduce the size of server you need.
Our website speed optimisation service works on exactly these layers, and the hosting saving is often a pleasant side effect of a faster site.
The Line Items Nobody Remembers Creating
A surprising share of many bills is resources that serve no one:
- Disks left behind when a server was deleted, and years of snapshots with no retention rule.
- Public IPv4 addresses, which AWS now charges for whether or not they are attached to anything.
- NAT gateways processing traffic that could go through free gateway endpoints, for example from private servers to S3.
- Logs kept forever at full detail in CloudWatch or Log Analytics; most can be kept for weeks, then archived to cheaper storage or dropped.
- Load balancers, test databases and whole environments from projects that ended.
- Files in standard storage that are rarely read; lifecycle rules can move them to infrequent-access or archive tiers automatically.
- Data transfer between availability zones or regions from a chatty architecture that could keep traffic local.
Databases: Often the Biggest Single Line
Managed databases are frequently oversized because they were scaled up to cope with slow queries. Fix the queries first; a missing index or a query inside a loop can make a database look as if it needs twice the capacity. Our guide to why sites slow down as they grow covers the usual causes, and our database development team handles the tuning. After that:
- Keep high availability across zones for production, but not for staging or development.
- Check that provisioned storage performance matches what the database actually uses.
- Consider serverless or burstable database options for workloads that are idle most of the day.
- Add read replicas only when measurements show reads are the bottleneck.
Keep the Savings From Creeping Back
Costs drift back up when nobody is watching. Set budgets with alerts at sensible thresholds, enable the providers' cost anomaly detection, require tags on new resources, and manage infrastructure with Terraform so environments are created deliberately rather than by hand in the console. A short monthly look at the bill alongside the performance baseline is enough for most businesses. If your cloud move was a straight copy of the old servers, our article on lift-and-shift versus re-platforming explains why a second pass usually saves more. Ongoing reviews like this are part of our cloud hosting and DevOps support.
If you would like your bill reviewed, send us a recent cost breakdown by service and a sense of your traffic. We will point out where the waste is and which savings carry no risk to performance.
Frequently Asked Questions
Will moving to smaller servers make our website slower?
Not if the change is based on measured usage and checked against a baseline. Many servers run at a small fraction of their capacity, so dropping a size has no visible effect. We change one thing at a time, compare response times and Core Web Vitals before and after, and keep the previous size ready to restore if anything regresses.
Are reserved instances or savings plans worth it for a small business?
They are worth it for the part of your infrastructure that runs every hour of the year, such as the production database and the minimum number of web servers. They are a poor fit for capacity you might remove in a few months. Right-size first, then commit to the steady baseline only, and leave peaks to on-demand or autoscaled capacity.
How quickly can cloud cost savings be made?
Removing unused resources, scheduling non-production environments and setting log retention can usually be done within days and show on the next bill. Right-sizing, caching and database tuning take longer because each change is measured. Commitment discounts come last, once usage has settled at its new level.
Does using a CDN really reduce hosting costs?
Usually, yes. Every image, script or page served from the CDN cache is a request your servers and database do not handle, and data transfer from a CDN is often cheaper than transfer straight from the cloud provider. The site also loads faster for visitors far from your servers. The gain depends on how much of your content can be cached safely.








