Totally agree with you, today Cloud companies: GCP, AWS and Azure native ML solutions are not a E2E platform but different teams that work independently and as a whole release a ML solution, specifically I can talk about Vertex AI. Notebooks, Training, Prediction while they provide Enterprise features (Security, Network, IAM, Encryption) do not play well with users. (Expect ML users to be Cloud Engineers) and an ML E2E workflow is hard to achieve.

I would divide the main challenges for AI startups as follows: 1. Support Enterprise Features. (Security, IAM, VPC-SC, Ecripyion) 2. If providing Compute Resources do not make that your main source of income (i.e. DeepNote, Saturn Cloud) which that may not scale. 3. Data integration. (BigQuery, S3, GCS, etc)

Databricks is one of the ones that have integrated with each of the clouds and provide this E2E workflow nicely, in addition they have seen the nascent Analytics market and invest on it. My concern with some startups: Cohere.AI similar to OpenAI GPT3, is that some are only solving some part of the ML workflow: (Seldom, OctoML, etc.) they may get some customers now, but will be hard to scale, and probably best destiny is getting acquire by major players.