Archive 27/02/2024.

“AWS Tax” & evolution


Here is a “counter-intuitive” example. Dropbox migrating off public (utility) cloud to save money.

When looking at evolution on wardley maps, shouldn’t everyone be shifting to public (utility) services? Or is there a scale (e.g. Dropbox) when this breaks down and you are better off with building your own? Or does this mean that AWS still has (too) high margins?


@simonduckwardley often repeats that the primary constraint of delivering compute as a service (cloud) is your ability to build your data centres (physical components of the service). This process is cash intensive and rather slow.

I think it should be possible to track AWS announcements and say precisely how many of those centres were build and when, and I expect the number to be growing exponentially. The fact that Amazon is still earning money from cloud means that margins are ridiculous, and prices are being used to limit the demand, and are not based on real costs of delivering the service.

So, that means that given specific scale and conditions, you might be better using your private cloud, but you have to consider future price cuts of your competitors. Without that, you are very likely to overpay (which may or may not be a big issue to you).


I wouldn’t say it’s just scale. I suspect the core elements of Dropbox’s architecture (storage, data transfers) meant they were able to find economies outside of public cloud - which I bet they still use a LOT of on top of their own.

Same for someone like Netflix.

Does Lyft have such a thing that would fit that model?


And this thread, too!


Exactly this! How do maps help with this? With financial flow charts? Example anyone?

Realistic example: Building an own datacenter is probably a stretch. However take digitalocean that now has managed databases (finally!), but the pricetag is 3x the cost of the underlying droplet/vps. If you run a somewhat larger database you will probably find a skilled devops person/team (offsite/offshore) and easily pay less for that.


Maps do not help with this. The guideline is to outsource managed components and focus on added value, and not to compete with established players. But it is a guideline, and if you have a good reason, you should not follow it.

Some people claim the Dropbox migration off the public cloud was not that much about costs but about control:

There is also a concern of Amazon competing with Dropbox, so it looks like the decision to build private infrastructure had more dimensions than just cost.

Maps advise avoiding indirect competition with players that have bigger scale and deeper pockets unless you have a very good reason. Disadvanteges of doing so fall into two categories:

  • missed alternatives - you are spending time on a feature that is not new, and you can only be marginally better (for Lyft, it would not save more than .014$ per transaction). You may have better things to do (conquer new market before competition, add differentiating services).
  • bigger inertia to change - if business condition changes (f.e. Lyft is banned from some countries) it is much easier to adapt if you do not have cost generating underutilised infrastructure.