Comment by messh

This matches my experience. I run a pay-per-use VM service (shellbox.dev) entirely on Hetzner auction servers with Firecracker microVMs. Sub-second boot, full Linux environment, SSH-only interface. The entire "cloud" layer is Firecracker + Btrfs reflinks for instant copy-on-write cloning. No managed Kubernetes, no proprietary orchestrators.

The total cost of that stack is remarkably low — cheap enough to offer VMs at $0.02/hr running and $0.50/mo stopped, which undercuts most hyperscalers for bursty workloads. The "billions in investment" framing is exactly the problem. Most of what hyperscalers sell is convenience wrappers around commodity compute, and the lock-in is the product.

Wrote up the economics here if anyone's curious: https://shellbox.dev/blog/race-to-the-bottom.html