Amazon Web Services (AWS): Pros, Cons, and Services

AWS is the safe answer, and that is both the recommendation and the warning. It has the widest service catalogue, the largest regional footprint, the deepest pool of engineers who already know it, and the most third-party tooling built around it. It also has the most ways to spend money by accident. If you adopt AWS, adopt the cost discipline on the same day, because the platform will not do it for you.

The footprint, and why it matters

Start with scale, because it is the one advantage no competitor has matched. Amazon’s global infrastructure page states that the AWS Cloud spans 124 availability zones within 39 geographic regions, with seven more zones and two more regions announced for the Kingdom of Saudi Arabia and Chile. AWS puts at least three availability zones in every region and describes its backbone as nearly 20 million kilometres of fibre optic cabling. In North America alone it lists 31 availability zones across nine regions, backed by 31 edge network locations.

Why should a 30-person business care about any of that? Two reasons. First, latency: if your customers are in Thailand, New Zealand, Malaysia, Israel or the United Arab Emirates, there is a region near them, and the difference between 20 ms and 220 ms is visible in a checkout funnel. Second, residency: when a contract says data must stay inside a jurisdiction, having a region in that jurisdiction turns a legal problem into a configuration setting.

Market position follows from that. Synergy Research Group, now part of TechInsights, measured second-quarter 2026 enterprise cloud infrastructure spending at $143 billion, up more than 43% year on year. AWS took 28% of it, Microsoft 20% and Google Cloud 15%. AWS’s lead is narrowing, but it is still the largest, and that shows up in the job market more than anywhere else. Hiring someone who already knows AWS is simply easier.

The services most businesses actually touch

The catalogue runs to hundreds of services. The working set for a small or mid-sized business is much smaller, and it is worth naming it so nobody feels obliged to learn the rest:

• EC2 for virtual machines, which is where most migrations land first because it needs the fewest changes to the application.
• S3 for object storage: backups, uploads, static assets, logs. S3 Standard lists at about $0.023 per GB per month in US regions, which makes it the cheapest place to put anything you are not sure you need.
• RDS for managed MySQL, PostgreSQL or SQL Server, so nobody on your team has to remember how to run a failover by hand.
• CloudFront for content delivery and Route 53 for DNS. Unglamorous, cheap, and they fix a surprising share of performance complaints.
• Lambda for scheduled jobs and webhooks that do not justify a server.
• IAM, which is not optional. Getting roles and policies right at the start is the difference between a tidy account and an audit finding.
• Local Zones and Wavelength when you need single-digit millisecond latency in a metro area or next to a mobile network. We compare that model with central cloud in our piece on edge computing against central cloud.

The discount machinery, and how to use it

AWS list prices are the price you pay only if you never look. A plain m7i.large instance, two vCPUs and 8 GiB of memory, lists at $0.1008 per hour in us-east-1, about $73.58 a month at 730 hours. AWS documentation states that Savings Plans and Standard Reserved Instances both reduce compute cost by up to 72%, which takes that same instance to roughly $21 a month at the maximum discount. Spot capacity is cheaper again: the m7i.large spot rate in us-east-1 was tracked at $0.0741 per hour in 2026, about 27% below on-demand, with the catch that AWS can reclaim the instance.

The practical rule we give clients is unglamorous. Run everything on demand for the first three months. Watch what is genuinely always on. Then buy a Compute Savings Plan covering about 60% to 70% of that steady baseline, never 100%, because you want headroom to re-architect without stranding a commitment. An unused Savings Plan is a subscription to nothing.

Data transfer: the bill nobody forecasts

This is the section that saves money. On AWS, moving data costs money in more places than people expect, and none of it appears in a compute estimate. Inbound traffic is free. Traffic inside one availability zone over private addresses is generally free. After that, a meter starts.

One number to carry away. A site serving 2 TB of images and video a month pays roughly $171 in egress alone at $0.09 per GB, before a single instance hour. Put CloudFront in front of it and most of that traffic is served from cache at a lower rate. That is usually the single highest-return change we make on a new AWS account.

Worth noting for the long term: AWS introduced a free data transfer programme for customers moving off the platform in 2024, and from 12 January 2027 the EU Data Act prohibits switching charges, including egress fees, for customers changing provider. Egress-as-lock-in is on a timer. Egress on your ordinary monthly traffic is not.

What the SLA really promises

AWS makes two separate compute commitments, and conflating them is a common planning error. The region-level SLA promises a monthly uptime percentage of at least 99.99% for services whose instances or tasks run concurrently across two or more availability zones. The instance-level SLA promises 99.5% for a single EC2 instance. Those are very different numbers: 99.99% allows about 4.4 minutes of downtime a month, while 99.5% allows about three and a half hours.

The credit bands are tiered. Fall below 99.99% but stay at or above 99.0% and you get a 10% service credit. Below 99.0% it rises to 30%, and below 95.0% to 100%. There is also a quietly useful clause: AWS does not charge for any single EC2 instance that is unavailable for more than six minutes in a clock-hour, applied automatically with no claim needed. Everything else requires you to open a case, with timestamps and logs, before the end of the second billing cycle after the incident.

That is the honest reading of every cloud SLA, not just Amazon’s. The commitment is commercially reasonable effort plus a partial refund. If four hours offline would cost you more than a second instance costs to run, the architecture decision is already made.

Where AWS is the wrong choice

AWS is the wrong choice more often than its market share suggests. If your business runs on Microsoft licences, Entra ID and SQL Server, Microsoft Azure will cost less and take less work, because you can redeploy licences you already own. If your team of three has never touched AWS and your requirement is a WordPress site and a mailbox, a managed host will serve you better for a tenth of the operational effort. If your market is primarily in Asia or Africa and price matters more than ecosystem depth, Huawei Cloud deserves a quote.

There is also a cultural cost. AWS hands you primitives and expects you to assemble them. That is excellent when you have an engineer who enjoys infrastructure and poor when you do not, because half-assembled primitives fail in interesting ways at two in the morning. Flexera’s 2026 State of the Cloud Report put estimated wasted cloud spend at 29%, the first increase in five years. Nearly a third of the average bill buys nothing, and that is not a billing bug, it is an attention problem.

Our advice is boring and it works. Pick one platform, learn it properly, tag everything, set a budget alert, and review the top ten line items once a month. Eudora does exactly that remotely for clients worldwide as part of our cloud solutions work, whether the account is in AWS, Azure or somewhere smaller.

Want a second pair of eyes on an AWS bill, or a migration plan that accounts for the transfer costs before you commit? Get in touch with Eudora Technology to talk about your project.