Harness surveyed 1,200 FinOps practitioners for their 2025 State of FinOps report. The headline: 21% of enterprise cloud spend is waste. Flexera's State of the Cloud 2025 puts the number at 28%. Either way, we are talking about tens of billions of dollars globally — resources provisioned, running, and billed for work that is not happening.
55% of developers surveyed said their cloud commitments (reserved instances, savings plans) are based on guesswork. Not forecasting. Not usage data. Guesswork.
I run an engineering studio that deploys on AWS, GCP, and Hetzner. We see cloud waste on almost every client engagement we join. Not because the teams are careless. Because cloud billing is designed to be opaque, and nobody is incentivized to watch it.
Where the Waste Hides
Dev/staging environments that never shut down. Every project spins up dev and staging servers. They run 24/7. The development team works 8-10 hours per day, 5 days per week. That is 50 hours of use and 118 hours of idle time. The bill does not differentiate.
Over-provisioned production. The team provisioned for peak traffic 18 months ago. Traffic is 40% below peak now. Nobody resized the instances because "what if traffic spikes again?" The cloud provider is happy to keep billing for capacity you do not use.
Orphaned resources. A developer spun up an RDS instance for testing. The feature was cancelled. The instance is still running. Nobody remembers it exists. The bill goes to the infrastructure line item and nobody questions it because it is a rounding error. Multiply by 50 orphaned resources across 3 years and the rounding errors become real money.
Unused reserved instances. The team bought 1-year reserved instances for a workload that was decommissioned 4 months in. 8 months of paid-for capacity with no workload running on it.
What We Do Differently
Right-size from the start
We deploy production services with monitoring from day one. Not just uptime monitoring. Resource utilization monitoring: CPU, memory, network, storage IOPS. After 2 weeks of real traffic data, we right-size the instances to match actual usage with a 30% headroom buffer. Not the 200% buffer that "just in case" provisioning creates.
Auto-scaling as default
For web applications with variable traffic, auto-scaling is not a nice-to-have. It is the default. Scale up when traffic increases. Scale down when it drops. Pay for what you use. Nautical Commerce processes 200K+ monthly transactions with traffic patterns that vary significantly by time of day and day of week. Fixed provisioning for peak would waste 60% of capacity during off-peak hours.
Monthly cost reviews
We include cloud cost review in the monthly engineering report for every client. Not a finance report. An engineering report: which services cost what, which costs are growing, which costs can be reduced. Engineers who see the bill make different provisioning decisions than engineers who never see it.
Hetzner for the right workloads
Not everything needs AWS. Our own website runs on a Hetzner VPS ($8.60/month) with Cloudflare CDN in front. The equivalent on AWS (EC2 + CloudFront + Route 53) would cost $50-$100/month. For a static site serving HTML, the cheaper option is the right option.
We deploy client applications on AWS or GCP when the workload requires it: managed databases, auto-scaling, global distribution, compliance requirements (HIPAA, SOC2). We deploy on Hetzner when it does not. The decision is based on requirements, not on which cloud provider has the best enterprise sales team.
The Fix
Cloud waste is an engineering discipline problem, not a tools problem. The teams that waste the least are teams that: monitor resource utilization from day one, right-size after real traffic data (not estimates), auto-scale variable workloads, shut down dev/staging environments outside business hours, audit for orphaned resources monthly, and review cloud costs as an engineering metric.
These practices cost nothing to implement. They save 15-25% of cloud spend. On a $10,000/month cloud bill, that is $1,500-$2,500/month. On a $100,000/month bill, that is $15,000-$25,000/month. The savings compound every month the practices are in place.
Our DevOps services include cloud cost optimization as a standard part of infrastructure setup. Not as a separate engagement. As a part of doing the infrastructure work correctly from the start.
Last updated September 28, 2025