UII BRIEFING REPORT 178 | JULY 2025
Serverless cloud container platforms offer rapid application scalability, which can be charged for by the second, making them ideal for AI inference. Unfortunately, the metrics used to charge customers for their usage vary between providers and often do not reflect application behavior. For this report, Uptime Intelligence trained a machine learning model to simplify these complex pricing models in the form of decision trees. The objective is to aid business leaders and FinOps practitioners in the cost-effective implementation of these new services.
KEY POINTS
- Serverless container platforms are charged for based on the number of seconds of infrastructure capacity consumed. This approach does not easily relate to application behavior, and pricing model nuances between providers make comparisons difficult.
- AWS was found to be the cheapest provider in most scenarios. However, AWS’ requirement for a minimum container instance can make it more expensive for small workloads with short-term demand — in these cases, Google Cloud is usually the cheapest.
- Google Cloud, Microsoft Azure, and IBM Cloud all offer free tiers for workloads processing up to 200,000 requests per month without provisioned capacity.
- Instance-based pricing can be more cost-efficient than request-based pricing models for steady, predictable workloads. If a container running on Google Cloud’s platform is active for at least 70% of the time and handling 40 million requests, instance-based pricing is usually more cost-effective than request-based pricing.