Covers 39:02–42:53. What the cloud actually made easier about scale — and what it didn't.

High scale is still hard

  • Object storage gives an elastic storage model — no more disk capacity planning — but sharding still leaks into application code; it can't be made fully transparent. At scale where one machine can't process the workload, real engineering thinking is still required, cloud or not. 39:31 39:59

Where the cloud really helped: the low end

  • The big win is scaling down: serverless can spin instances up/down very quickly and cheaply, enabling very low-scale services that would be much harder without the cloud (static VM allocation of memory/CPU). 39:59 40:21 40:39
  • Host Gergely's example: a small serverless website costing ~13 cents/month under minimal load. 40:39

Is sharding disappearing as a concern?

  • Less an effect of cloud than of hardware getting more powerful: a big modern machine handles workloads that used to require sharding, so more workloads run on a single machine at significant scale. 41:43
  • Parallelism doesn't go away — efficiently using hundreds of CPU cores on one machine is still a concern, and sharding is one way to achieve parallelism. 42:03
  • Cross-machine sharding is less pressing but not gone: some workloads still genuinely require it. And replication stays relevant at every scale, because it exists for fault tolerance, not scalability. 42:22