It depends on the specific use case. Some of the use cases I envision have sharp spikes in demand, and serverless can provide better service and price/performance. Part of faast.js is a cost analyzer that can tell you in real time how much your workload costs. What I found is that most people are probably using the wrong memory sizes for their lambda functions to optimize for price/performance. More on that when I write my next blog post... If you want a preview, check out this chart from the documentation: https://faastjs.org/docs/cost-estimates