Moving a monolith to serverless can reduce infrastructure costs when workloads spend meaningful time idle and can scale independently—but serverless is not automatically cheaper. Manohar Halappa reports a 30–35% infrastructure-cost reduction for workloads migrated in two modernization programs; his 2026 account does not provide the baseline bill, comparison period, workload volumes, or calculation method needed to reproduce that figure. Treat it as a result from those workloads, not a general savings rate.
What the reported 30–35% reduction does—and does not—show
In his September 20, 2026 DEV Community article, Manohar Halappa says workloads migrated in two modernization programs saw infrastructure costs fall by 30–35%. The available account does not disclose the starting costs, measurement window, workload volumes, or calculation details. It therefore offers a reported outcome, not an independently audited comparison or a forecast for other systems. Read Halappa’s article on DEV Community.
As an Amazon Associate I earn from qualifying purchases.
The useful takeaway is the decision method behind the claim: look for capacity paid for while idle, select capabilities that can scale independently, and move them in stages. Halappa’s phrase, “Serverless is a tool, not an ideology,” captures the right caution. The question is whether a workload fits the model and whether the full cost of operating it falls—not whether serverless is fashionable.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problemsWhich workloads are worth considering?
Start with workload shape and boundaries, not with a service name. Serverless is often a better candidate for stateless, short-lived, request-driven or event-driven tasks with variable demand and idle periods. A continuously busy service may be more economical in containers, as AWS Prescriptive Guidance notes in its discussion of re-architecting a monolith as microservices without containers.
#1 Best Overall
Before choosing a capability to extract, ask:
- Where are we paying for capacity that is not being used, and what is current utilization?
- Does the capability have a clear business boundary and a workload that benefits from independent scaling?
- How often does it run, how long does each execution take, and what CPU and memory does it need?
- What databases, services, and other capabilities does it depend on?
- How does it fail, recover, scale, and get deployed today?
- How business-critical is it, and what latency and reliability does it require?
These questions help distinguish a genuinely independent workload from a tightly coupled part of the monolith. Extracting a service without a useful boundary can add network calls and failure points without reducing the capacity the system needs.
How to estimate the full cost
Compare the cost of doing useful work, not just the compute line item. For AWS Lambda, the standard pricing model includes requests and execution duration, with duration charges affected by allocated memory. Other AWS services and data transfer can add costs. The actual price depends on service configuration and region, so use the current AWS Lambda pricing page for an estimate rather than relying on a generic per-request figure.
AWS cost guidance also identifies execution count, memory allocation, and duration as inputs, and recommends right-sizing. Its guidance on considering serverless for .NET is relevant to workload fit and cost factors; it does not validate Halappa’s reported savings.
Build the comparison around the same workload and service expectations. Include:
- Current utilization and the cost of idle capacity.
- Request or event frequency, execution duration, memory needs, and concurrency.
- Databases, storage, data transfer, logging, monitoring, and other adjacent services.
- Cost per transaction, request, or completed unit of processing.
- Latency, including cold starts where relevant, plus reliability and recovery performance.
- Operational effort: deployments, service ownership, observability, and incident response.
There is no universal percentage to insert into this calculation. Measure a representative workload before and after migration, with comparable traffic and service levels, and account for all services required to deliver it. A lower compute bill is not a saving if total cost per completed transaction rises or reliability deteriorates.
How to migrate incrementally
Move one bounded capability at a time, keeping an explicit interface between it and the monolith. AWS-oriented examples in Halappa’s account include API Gateway with Lambda for suitable request-driven work, SQS-triggered processing instead of idle polling, Step Functions for multi-stage workflows, and S3-backed event processing for large data operations. These are patterns to assess, not a required architecture.
- Establish a baseline. Record workload volume, utilization, cost, latency, reliability, and operational effort for the capability you may extract.
- Define the boundary. Specify the capability’s interface, dependencies, data access, and expected behavior when its services are unavailable.
- Choose a fitting execution model. Use event-driven functions for suitable intermittent tasks; consider containers for services that must run continuously.
- Prepare safe failure handling. Make event processing idempotent, set timeouts, use bounded retries with backoff, and route messages that cannot be processed to a dead-letter queue.
- Deploy with rollback in mind. Keep a clear path to restore the previous behavior, and use infrastructure as code so environments and changes are reproducible.
- Observe the new boundary. Track errors, retries, workflow state, latency, and service-level costs across the monolith and extracted component.
- Compare outcomes before expanding. Evaluate cost per useful unit alongside latency, reliability, and delivery measures; adjust or roll back if the change makes the service worse.
What can make serverless more expensive?
A distributed system has costs that a single deployment can hide. Extracted services create more deployments, network calls, failure boundaries, IAM policies, and observability requirements. Data transfer, databases, storage, logs, and retries also contribute to the bill. A poorly designed microservices architecture can cost more than the monolith it replaces.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →That is why extraction should follow a real scaling or ownership boundary rather than a desire to maximize the number of functions. If a service is constantly active, has tightly coupled database access, or needs frequent interaction with other services, include those factors in the cost and operational comparison before migrating it.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
