Amir Salmaniplatform and reliability engineer

work / leaving the hyperscaler

We want off the hyperscaler. Where do we start, and what does it cost?

Start with your bill and your last 3 incidents, not with hardware. I build and run platforms that have no hyperscaler under them, so I can tell you what the same system costs and what it asks of you when the machines are yours. I haven't migrated a team off AWS; what I've built is the destination.

What this usually turns out to be

Rarely “we hate the cloud”. More often one of these:

  • The bill is one line, it grows, and nobody can say which part is which.
  • A customer or a regulator wants to know exactly where the data sits.
  • You need GPUs, and renting them costs more each year than buying them.

Each has a different answer. The first can sometimes be fixed without moving at all.

How I’d start

  1. Read the bill with you until every large line has an owner.
  2. Price the same platform on machines you’d own or rent by the month, including the people it needs.
  3. Write down what you’d give up. Managed databases and someone else’s pager are worth real money.
  4. Only then decide. Sometimes the answer is to stay.

What it takes

Leaving isn’t the hard part. Running it afterwards is: upgrades, storage, recovery, and somebody who answers at 03:00. That’s the part I’ve done, and it’s where I’d spend most of the scope.

What I won’t do

Tell you to leave because leaving is what I sell. If the numbers say stay, the review says stay.

where I've done it

Tell me what you need