Serverless

Every computing model before serverless made you think about capacity. IaaS: how many VMs? PaaS: how many dynos? CaaS: how many container replicas? You would estimate traffic, provision capacity, and either overpay for idle resources or under-provision and crash during spikes.

Serverless flips this model. There are no servers to provision. No capacity to plan. No idle costs. You write a function. Upload it. When a request comes in, the cloud spins up an execution environment, runs your function, returns the result, and charges you for the milliseconds it ran. If nobody calls your function for a week, you pay nothing. If 10 million requests hit simultaneously, it scales to handle all of them, and you still do not think about servers.

This sounds too good to be true, and in some ways it is. There are real trade-offs: cold starts, execution time limits, debugging difficulty, and vendor lock-in. But for the right workloads, serverless is genuinely transformative. It is why Vercel built their entire platform around serverless functions, and why this course platform you are using right now runs on them.

Each compute model makes a different trade-off between control and simplicity. Here is where serverless sits: