Joey on SQL Server

Take Control of Your Cloud Bill: Cost Models, Discounts and Hidden Fees Explained

As cloud costs grow more complex, understanding fixed vs. variable pricing, spotting hidden charges and using the right discount strategies are essential to keeping budgets on track.

• By Joey D'Antoni
• 07/17/2025

The basics of cloud computing seem easy: you pay money to a cloud vendor in exchange for compute, storage and networking. Still, the reality is far more complex than this.

That complexity is why we have new cloud-era disciplines, such as cloud economics and FinOps. As a long-time architect of cloud systems, one of my primary tasks is not only to understand the initial cost of a given system, but also to predict how that cost will evolve as users and data grow. I wanted to share some of that knowledge and discuss in-depth how to understand and manage your cloud computing costs.

In the early days of cloud computing, probably motivated by low initial prices on storage, the general industry perception was that the cloud was a cheaper alternative to an on-premises datacenter. Now, there is a perception that the cloud is inherently more expensive. The reality is much more nuanced than that, especially for organizations that make proactive efforts to manage their costs, understand cost models and identify where hidden fees in the cloud lie. In this article, we will examine standard cost models, methods for achieving discounts and strategies for effective cost monitoring.

To simplify things, I typically break down cloud resource costs using two basic accounting terms: fixed and variable costs. An example of a fixed cost is Ebdv5 VM or a P30 managed disk in Azure. Whether you use those resources at 100 percent of their capacity all month or don't use them at all, the line item on your bill will be the same. Variable cost items would include network egress bandwidth, logging ingestion, or a Kubernetes cluster with auto-scaling, where the costs depend on how much you use the service or scale your workload.

 However, like most things cloud, it's not that simple -- an example of a hybrid model is Microsoft Fabric, which is primarily a fixed cost. Still, it has bursting and smoothing capabilities that can (if the user pauses and restarts to resume a stalled, overused capacity) incur costs beyond the regular fixed price. Cloud vendors tend to emphasize the benefits of variable cost models; however, as an architect, my preference is for fixed costs -- I want to have high (85 percent or more) confidence in the bill. However, not everything can be a fixed cost. For example, a lot of AI services are tied to API utilization with very complex pricing models, depending on which LLM you are using.

I mentioned log ingestion and network egress in my last paragraph. These are two examples of unaccounted variable costs. They either aren't in the pricing calculator, or the architect didn't consider them when building the solution. The cost of logging and observability for busy Web sites can often exceed the cost of the compute supporting those solutions. Log solutions like Azure Log Analytics or Amazon CloudWatch are priced in two dimensions -- the underlying storage costs for storing logs, and the more expensive part, the amount of log data ingested into that storage. There are several tactics you can take to reduce the cost of logging, like limiting how much and how verbosely your application logs events, evaluating compressed data formats (this makes alerting harder), or considering building your logging stack using Elastic, Logstash and Kibana. Observability is one of the classic make-or-buy decisions that cloud projects often encounter. It's easy to use CloudWatch, but as your app scales, you might want to consider your infrastructure. In terms of network egress, it typically comes with either a successful app or a poor network design. Pay careful attention to where network egress is metered. For example, in Azure, cross-availability zone traffic is free, whereas AWS bills for it.

 Let's talk about how to save money in the cloud. While often on forums or subreddits, I'll see people ask about powering down resources when not in use. However, this is not a valid approach for enterprise-scale cost management. Don't get me wrong, I still have automation to power down resources in my subscriptions. But it simply doesn't scale. There are four main ways to save money in the cloud: spot instances, resource reservations, savings plans and spending agreements with your cloud provider (also known as Enterprise Agreements in Microsoft terminology). Let's look at each of them.

Spot instances: Spot Instances are excess compute that the cloud provider can reclaim at any time. Depending on the configuration and the provider, these may have different warning levels before termination. Using spot instances can provide significant savings, up to 90 percent from retail pricing. But you'll need to manage your workload's state externally to the VMs. Use cases for spot instances are typically batch workloads that don't need to complete within a specific time frame, such as R&D work in the life sciences.

Resource Reservations: Reservations are one- or three-year commitments you make to a cloud provider to use a given resource type. Reservations can offer up to a 72 percent discount off retail prices and can be purchased upfront or every month. Within limits, reservations can be exchanged or refunded; however, please consult with your cloud provider before purchasing to understand their specific terms. The cloud providers do not tie reservations to a specific resource; for example, if you reserved 16 CPU E-series VM cores in the US East region of Azure, and you had four 4-core VMs, they would all receive the reservation discount. Likewise, if you had one 32-core VM, you would pay the discounted price for half of the VM's cost.

Savings Plans: Savings plans are more flexible than reservations, offering slightly lower discount percentages (up to 66 percent) than reservations. Savings plans tend to be more beneficial for organizations that experience fluctuations in their workloads. Savings plans are more limited in their resource coverage than reservations.

Spending Agreements: These agreements typically provide discounted pricing to organizations that make substantial spending commitments (if you have to ask, it's probably not your org) to their cloud providers. These agreements typically provide best-in-class pricing but come with penalty clauses if the organization fails to meet its spending commitment. I've seen cases where firms increased the size of resources just to meet the spending commitment.

We've covered how the cloud costs money, and how you can save money. Now the big question: how do you avoid those terrible, runaway surprise bills? Cloud providers supply adequate cost management tools, which include budgets and alerting features. It's a good idea to set budgets based on your normal, expected monthly spend. Where budgets are especially critical is for new projects that are using a lot of variable cost services. For publicly facing projects, I want to check those numbers daily to understand if we have budgeted correctly or if we need to look at refactoring our service plans.

Cloud pricing, planning and economics have matured to the point where they are a dedicated field. Many large organizations have given considerable thought to this practice. Still, by applying some of the techniques you have learned here, you can also implement them in your organization. Understanding pricing, taking advantage of various discount approaches, and utilizing budgets to identify misconfigurations or out-of-band spending can help you keep your cloud bill under control.