Ask HN: Memory limitation with cursor parallel agents

I spoke to a friend of a company that you all have probably heard of 10b+ val. They currently have a problem where the engineers are not able to run more than 5 instances of their server locally. They want to run more instances (10) so that AI agents can work on these and everyone is 10x more productive.

My answer to this was to remotely host instances and let cursor work on it. He also said the another team might (?) be working to build something to help internally. Why wouldn't they just use a company like fly.io or modal for this? Or just give employees GCP VMs that have more compute and they just ssh into it?