Is serverless computing truly cheaper than managing dedicated virtual machines?

All answers to this question.

There is definitely a "tipping point." Serverless is incredibly cost-effective for bursty traffic or background tasks that run intermittently. However, for a high-volume, steady-state API, the "invocation cost" and the cost per millisecond can become significantly more expensive than a reserved VM instance. A good rule of thumb: if your CPU utilization is consistently above 40%, you're probably better off with a dedicated server or a containerized environment like Fargate or Kubernetes. Serverless is for agility and low-maintenance; it's not always the cheapest option for heavy, sustained workloads.

   Answered 2025-01-26 by Sandra Jenkins

How are you handling "cold starts" in your current serverless setup, and is that affecting your end-user experience?

   Answered 2025-01-29 by Gregory Hall

• Gregory, cold starts were our biggest headache with Java-based functions. We had to move to Go to keep latency down. To answer your broader question about cost: if you're paying for "provisioned concurrency" to solve cold starts, you're essentially paying for a server that's always on anyway. At that point, the cost benefit of serverless disappears. We found that for our core API, a small auto-scaling group of VMs was 20% cheaper and much more predictable in terms of performance.

     Commented 2025-02-02 by Matthew Clark

Serverless is best for "Glued Code"—connecting different services together—rather than hosting the entire application logic.

   Answered 2025-02-05 by Amanda White

• Spot on, Amanda. It’s an architectural tool, not a universal replacement for servers. Using it for the right tasks is the key to actual savings.

     Commented 2025-02-06 by Ryan Henderson