ענןCloudAWSStorageעלויות ענןFinOpsDevOpsתשתיותIOPSVPCAuto Scalingניהול עלויותבועז זינימןBoaz Ziniman

The Forgotten Layer: Why Cloud Storage and Network Costs Keep Growing Quietly

Boaz Ziniman, former AWS Head of Developer Relations for EMEA, reminds us that while everyone focuses on AI, the infrastructure layer underneath keeps growing and costing money. Disks provisioned just in case, IOPS nobody reviews, and cross-zone traffic become a fixed monthly bill, and vendor pricing changes even when your configuration does not.

Boaz Ziniman

מומחה לטכנולוגיה, בינה מלאכותית וחדשנות, שהופך רעיונות מורכבים לסיפורים ברורים, שימושיים ומניעים לפעולה.

September 26, 2026

The Forgotten Layer: Why Cloud Storage and Network Costs Keep Growing Quietly

Why look at infrastructure now, not just at AI?

Boaz Ziniman, with 30 years in the industry and a long tenure at AWS, explains that preparing for an upcoming event pulled him away from the model layer and back to fundamentals: machines, disks and networking. His observation is direct: most of us focus on what is shiny, while the infrastructure holding everything up is treated as "it works, don't touch it."

In his view, "don't touch it" is also a decision. Like any decision, it carries a price, and that price is paid every month whether anyone notices or not.

How does managing storage and networking differ from managing compute?

With compute, the industry has matured. Monitoring and Auto Scaling tools, accumulated know-how and vendor support make it possible to add and remove capacity with almost no manual effort. Whatever is unused simply disappears.

Networking and storage tell a different story. Systems are spread across multiple availability zones and even regions for resilience, and every byte moving between them is billed. Network and VPC rules only accumulate, because nobody wants to delete a rule that something might still depend on.

How does silent waste build up in storage?

Cloud disks almost always grow and rarely shrink. Reducing a disk means provisioning a new one, copying data, accepting downtime and getting sign-off. It is technically possible, yet the disk allocated just in case several months ago is still being billed every month.

Size is only part of the problem. Performance is also bought as insurance: IOPS set to avoid a bottleneck, or a larger disk chosen purely for throughput because that is how it used to work. Nobody checks whether those assumptions still hold, and nobody volunteers to lower performance in production.

Why does the bill change when the configuration does not?

According to Boaz, cloud providers adjust prices and pricing models frequently, and a recent update is set to change invoices for a large number of customers. Your settings stay the same, but what they cost does not.

While attention is fixed on AI, this infrastructure layer keeps expanding quietly, and every system in the organization, intelligent or not, runs on top of it. The question Boaz leaves readers with: when did someone on your team last reduce anything in storage, and what did it take?

Key Insights

•In the cloud we learned to add and remove servers almost automatically, yet storage and networking are still governed by the unwritten rule of "it works, don't touch it."
•A disk provisioned just in case a few months ago keeps costing money every month, and performance such as IOPS is often bought out of habit that nobody revisits.
•Spreading a system across multiple availability zones and regions improves resilience, but every byte that moves between them costs money.
•Cloud providers change pricing frequently, so your configuration can stay identical while your bill changes.

Frequently Asked Questions

Why do cloud storage costs tend to only go up?

Because enlarging a disk is easy, while shrinking one is a project: a new disk, data migration, downtime and approvals. In practice nobody volunteers, so disks provisioned just in case stay in place and are paid for every month.

What is wrong with over-provisioning IOPS?

Performance bought to avoid a bottleneck, or a larger disk chosen only for its throughput, often rests on assumptions that were true in the past. Nobody checks whether they still hold, and reducing performance in production feels risky, so the cost remains.

Why is cross-zone network traffic in the cloud so expensive?

To improve resilience, organizations spread systems across several availability zones and even regions. Every byte moving between them is billed, and network and VPC rules keep piling up because nobody wants to delete a rule something might depend on.

How can an organization start controlling infrastructure costs?

Boaz suggests asking one simple question: when did someone last reduce anything in storage, whether size or performance, and what did it take? The answer reveals whether there is a real process for revisiting allocations or whether this layer is simply growing quietly.

Based on a post by Boaz Ziniman

View original post →

About the speaker

Boaz Ziniman

מומחה לטכנולוגיה, בינה מלאכותית וחדשנות, שהופך רעיונות מורכבים לסיפורים ברורים, שימושיים ומניעים לפעולה.

View profile →