Cloud Cost Optimization
Right-sizing, reserved instances, spot instances, and waste elimination. Stop burning money on unused resources.
“The cloud is just someone else's computer. And someone else's computer has a very aggressive billing department.”
What you'll learn
The bill is a production signal
Cloud spend is not a finance-only number. It is what your architecture costs to exist. If you cannot map dollars to a service, a team, and an environment, you cannot cut without folklore.
Tag everything that can be tagged. Enforce tags at create time if the platform allows it. 'We'll tag later' is how later becomes a spreadsheet with 40% unallocated.
Right-sizing beats a clever discount
Most first-pass architectures are 2–4× too large because someone copied a production-shaped instance into a prototype and never came back. Compute Optimizer, Azure Advisor, and GCP Recommender are rude and usually correct.
Look at CPU, memory, and disk for a boring week, not a launch day. If p95 CPU is 8%, you are paying for a feeling.
Reservations and spots are different bets
Reserved instances and savings plans pay you for committing to a baseline. Use them on the load that is always on and not leaving. One-year is easier to defend than three if the product still changes shape every quarter.
Spot (preemptible) is 70–90% off because the machine can vanish. Batch, CI runners, and stateless workers belong there. Your primary database does not, unless you enjoy defining 'primary' as 'the one that survived'.
Storage and the zombie disk
Idle load balancers, unattached volumes, old snapshots, forgotten IPs, and gp3 volumes sized for a migration that ended in 2024: this is where the quiet money goes. Lifecycle policies on object storage are not glamorous. They are how logs stop costing more than the app.
Move cold data down a tier on purpose. Infrequent access is not a personality. It is a price list.
Waste is a meeting you can schedule
A weekly 20-minute cost review with the people who can actually turn things off beats a quarterly panic. Budget alerts should fire before the CFO's dashboard does.
FinOps is a name for 'engineering and finance look at the same numbers'. You do not need a new department to stop running a staging cluster 24/7 for a team that works 40 hours.