How cloud providers overcharge you (and how to fix it)

Cloud bills balloon quietly through defaults, usage-based pricing, and “invisible” resources—so the fastest way to cut costs is to audit specific traps and change a handful of settings. The core message is that cloud spending rarely comes from one obvious purchase; it accumulates through log retention that never expires, bandwidth charges triggered by seemingly small assets, per-operation database pricing, and leftover infrastructure that keeps running after a project is “deleted.”

The first and most persistent cost leak on AWS comes from log storage defaults. When new resources are created—whether a Lambda function, an RDS instance, or an ECS cluster—AWS automatically sets up CloudWatch log groups. The retention policy for those log groups defaults to “never expire,” meaning stored log data accrues indefinitely. AWS then charges twice: once to ingest logs and again to store them monthly by gigabyte. Over time, that turns into ongoing “rent” for old text logs from years-old activity. The practical fix is to set retention to a short window such as 5–7 days (or even 1 day). For teams that dislike CloudWatch’s experience, the transcript also suggests moving logs off AWS by pulling them into a self-hosted Grafana setup (e.g., on a VPS) and then reducing CloudWatch retention to near-term storage.

Next comes the bandwidth trap, driven by egress fees. A small 10 MB looping hero video can become 10 GB of transferred data for 1,000 visitors and 1 TB for 100,000 visitors. On platforms like Vercel, crossing included bandwidth thresholds can trigger incremental charges (the transcript cites $0.15 per extra gigabyte), turning a marketing asset into a recurring margin killer. The recommended workaround is to host the video on YouTube or Vimeo and embed it, effectively shifting bandwidth costs to those platforms. For a more “native” HTML video experience, the transcript recommends hosting heavy assets on Cloudflare R2, which charges for storage but $0 for egress bandwidth.

A third trap targets image optimization costs. Framework image components (the transcript mentions Next.js/“XJS”) can trigger paid transformation limits on platforms like Vercel, with quotas described as low enough to be exhausted quickly on image-heavy sites. The suggested fix is to stop relying on the platform’s on-the-fly image component and instead serve pre-generated WEBP images using the standard HTML <img> tag. The argument is that modern network speeds make micro-optimizing load time less critical than avoiding recurring vendor processing fees; batch-converting media to WEBP during build time is presented as a way to capture most benefits without ongoing charges.

Other traps include choosing regions with local pricing premiums—where servers in places like São Paulo or Cape Town can cost up to 50% more than the same setup in the US—and the “NoSQL hype” pattern of using serverless databases like DynamoDB, which charge per operation and can become expensive when product requirements shift. The final, widely common issue is the zombie resource trap: deleting a server doesn’t necessarily delete dependent services like load balancers. The transcript recommends using AWS resource group search (tag editor) across all regions and resource types to find and terminate leftovers, and it suggests infrastructure-as-code (AWS CDK) to keep cleanup reliable.

The takeaway is blunt: audit logs, check regions, eliminate zombie resources, and replace “default” managed conveniences with cost-aware alternatives—before the bill quietly compounds into thousands of dollars a year.