I think a big reason that AI/ML/Robotics is closely associated with FAANG companies is that these are very expensive fields to get into compared to web developm

I think a big reason that AI/ML/Robotics is closely associated with FAANG companies is that these are very expensive fields to get into compared to web development or CLI utilities. FAANG companies can also move much faster than FOSS efforts when it comes to putting together great teams. There was a REALLY good comment on FOSS vs. closed source commercial software that was brought up a couple weeks ago, which I wish I could quote verbatim but unfortunately you'll have to deal with my paraphrased version: > "In early UNIX days, we got things done on computers using lots of small tools that did one thing well because compute resources and developer talent were both in scarce supply. Nowadays, software companies can put together teams that can manage huge monoliths of software. In order for FOSS to thrive, we need to put less time into trying to beat these companies at their own game (e.g. code behemoths like the ones you see on Linux desktop projects) and focus more on making great tools that do one thing really well (e.g. cURL, GNU Readline)" ML/AI is also really hard because it requires people to establish much stronger interfacing standards between the important steps of the data analysis process AND the substeps within those steps. Keras[2] was an enormous step forward in this regard, but the creator now works at Google and I assume he isn't very interested in continuing to make Keras the standard frontend for Deep learning model specification. HuggingFace[1] Is an awesome counterexample to this trend of AI/ML being lead by FAANG. I'm fiending for an opportunity to check out Thinc, but I'm quite occupied with security coursework these days. 1: https://angel.co/company/hugging-face https://angel.co/company/hugging-face 2: https://keras.io/ https://keras.io/