When I was at DAZN, we had 𝗼𝘃𝗲𝗿 2,000,000 𝗰𝗼𝗻𝗰𝘂𝗿𝗿𝗲𝗻𝘁 𝘃𝗶𝗲𝘄𝗲𝗿𝘀 for big events like El Classico. Almost everyone logs in just before kick-off, creating huge (100x) traffic spikes. Back then, Lambda couldn't handle spikes of this magnitude. So we had a rule that any APIs along the critical path of a user logging in and watching a match must run in containers unless the API response is cacheable. Fast forward to 2024, after the scaling change announced at re:Invent 2023, I suspect Lambda can handle even spikes like this. Before you say "That's going to be very expensive!" yes, but the spikes are short-lived, so it won't be expensive 24/7. And do you know what else was even more expensive? The 100s of idle EC2 instances in the non-production environments. They were easily costing more than $10k/month. Plus all the developer hours that went into patching container images, troubleshooting networking issues, etc. Things that we wouldn't have to deal with if we could use Lambda for these APIs. When you have 250+ engineers, this amounts to a huge opportunity and staffing cost. So, remember, when thinking about costs, you need to think in terms of Total Cost of Ownership (TCO). Especially in the age of AI because an AI-assisted engineer can give you so much more if his/her time is used on the right problems. ----- If you liked this, then you might also enjoy the tips and tutorials I publish every week about AWS and serverless. Subscribe to my free newsletter and level up your AWS game: theburningmonk.com/subscribe… #aws #serverless #awslambda