Counter-example: I worked in ML research at one of the largest companies in the field (in the billions of predictions per day) and 16GB was more than enough to do anything I would ever do on a laptop/tablet/etc..
Furthermore, I found that solutions which couldn't be successfully worked on with Basically, when the difference between a dev sample and a prod sample is 3+ orders of magnitude, your ability to scale the process up/down, fail gracefully, even battery life, etc. are far more important than having large swathes of memory to fail on. Especially on the go.
If you do deal with production (big) data, you're ssh'ed in to some always-on, process-maintained hardware if only for the sake of latency and interruptions. And you better have the toolset to do it. How much memory does a terminal take?
For graphics, I understand - but then you're definitely not working on a surface pro or on the go at all really.