Organizations under pressure to cut cloud costs often reach for discount programs first as the simplest lever. It feels like a win to commit to reserved capacity or enterprise agreements because it lowers the rate paid per unit of compute or storage.

However, this approach conceals a dangerous trap. When discounts are secured before workloads are optimized, or if the wrong discounting mechanisms are used, organizations can inadvertently lock in waste. Instead of solving the problem, they multiply it by cementing oversized, underutilized infrastructure into long-term financial commitments that are hard to unwind.

Why This Pitfall Is So Damaging

Cloud discounts appeal to organizations because they’re easy. Finance teams can negotiate them without needing input from engineering or application owners. Yet if workloads are oversized, discounts can lock in the wrong configuration. If Kubernetes containers over-request CPU and memory, clusters run hot on paper but cold in practice, driving up node counts and costs. For example, if AI jobs reserve expensive GPUs but rarely use them, those GPUs sit idle while still being paid for.

The result is the same every time: a lower price per unit, but far too many units consumed.

Kubernetes: The Proof Point

This very scenario plays out in Kubernetes environments across the industry. According to a recent survey, 98% of senior IT leaders in the U.S. say Kubernetes is becoming a major driver of cloud spend, but 91% admit they cannot effectively optimize their clusters.

That gap explains why discounts are so dangerous here. If nearly every organization is struggling to right-size Kubernetes, then locking in today’s usage patterns also embeds today’s waste. The problem is compounded by the way Kubernetes allocates resources. Developers tend to inflate their requests for CPU and memory to avoid risk. Kubernetes then honors those requests, even if the workload uses only a fraction of them. Over time, this leads to clusters that appear full but are in fact underutilized, prompting the autoscaler to add even more servers.

The result is a cycle of over-provisioning that drives up bills without delivering matching business value.

Why Organizations Struggle to Avoid It

The core problem is that discounts are simple and optimization is complex. Securing a contract with a cloud provider is a straightforward financial transaction. Right-sizing workloads, on the other hand, requires cooperation between finance, operations and application teams. It means examining evidence, understanding workload behavior and adjusting configurations accordingly. That collaboration is not always easy to achieve in organizations where teams operate in silos.

This is why the FinOps Foundation’s State of FinOps 2025 report named workload optimization and waste reduction as the top challenge for practitioners today. Discounts may deliver quick wins, but they do nothing to solve the underlying inefficiency.

How to Break the Cycle

The good news is that this pitfall is avoidable. Cloud commitments don’t have to put inefficiency in stone. The key is to change the order of operations: Understand the optimal resource configurations first, then negotiate discounts. Then, every dollar committed connects to real demand instead of inflated assumptions.

To make that shift, organizations should:

• Track actual usage by monitoring real CPU, memory and GPU consumption over time, rather than relying on requested or provisioned numbers.
• Analyze these numbers to determine the true resourcing requirements, which are usually much lower than what is configured. In Kubernetes, that means aligning requests and limits with real utilization, and understanding the resulting node capacity requirements.
• At this point, it is safe to secure discounts by making commitments to these numbers, not the original configurations. If there is a large difference, then it may be wise to favor flexible discounting mechanisms, such as savings plans, so there is freedom to change the underlying cloud resources in use.
• Proceed to optimize the Kubernetes resources, as well as the underlying node resources, to move toward the intended state. Automation is key in container environments, as the manual effort to optimize thousands of pods can be significant.
• Collaborate across teams so finance and engineering can work together to achieve these goals. Engineers understand workloads, finance understands commitments, and only with shared evidence can the right decisions be made.
• Adjust the commitments if necessary to reflect the actual resources in use. Some instruments allow for conversions, and others will automatically apply the discounts to whatever is in use. Having based the discounting on the optimized state will prevent you from becoming “under water,” where actual utilization is below the commitment level and any optimization is pointless.

Why Order Matters

The temptation to rush to discounts is understandable. Cloud providers make them easy to buy, and they deliver an immediate sense of savings. But having the foresight to avoid undesirable lock-in and pave the way for resource optimization yields more than a quick win. It may be harder than simply signing a contract, but it creates the kind of durable efficiency that compounds over time.

Cloud economics is not a procurement problem; it is an operational discipline. Discounts can reduce rates, but only culture can reduce waste. The real opportunity is to build a practice where finance and engineering share ownership of efficiency and treat optimization as continuous, not occasional. When that shift happens, discounts become a bonus on top of efficiency rather than a crutch to excuse inefficiency.

The stakes are high. Discounts before optimization may keep today’s budget reviews quiet, but they set tomorrow’s leaders up for years of waste they cannot escape. With Kubernetes already driving the majority of cloud spend, the cost of inaction compounds every day. The choice now is clear: lock in inefficiency or build a foundation for sustainable growth.

KubeCon + CloudNativeCon North America 2025 is taking place in Atlanta, Georgia, from November 10 to 13. Register now.