It's easy to look back with nostalgia, but perhaps a reality check is worthwhile.

Firstly, at the time (mid to late 90s) most computers were desktop. There were relatively few servers, and the internet was small.

During this period Linux wasn't terribly good. Windows was a better desktop that "just worked". It was a loooong way from perfect, and rebooted regularly, but hardware support worked, you could print to any printer, network and so on. Macs had better design, and a better interface. Linux on the desktop required endless fiddling, very specific hardware. And couldn't play games.

Sun, Unix, VMS et al made better servers. They supplied software and hardware. They came with support departments etc. And were very (very) expensive.

As the volume of servers grew, as commodity hardware got better and cheaper, Linux was able to compete simply on price. As adoption grew, so did hardware support etc. Linux is especially well suited to servers (very limited software library needed, very little need for UI and do in.)

The desktop market was a lot more competitive. Windows was at most $100, and choosing an OS based on $100 saving is silly. And in pretty much all measures (perhaps excepting reboots) Windows or Mac was better. Linux has never gained much of the desktop market mostly because it was objectively a worse desktop. And has no business model to get onto the desktop so loses completely on the marketing front. (And marketing matters. Without software the OS is meaningless.)

Fast forward to today, and your question. All the items you mentioned are server related. All are as good or better than commercial offerings.

Open-source AI is certainly a server based activity. But there are no trillion $ AI data centre's. So the obvious niche is on-prem AI. No tracking etc. The barrier to entry there is the hardware cost. To provide a similar experience to say Anthropic you need to spend a LOT. To spend that kind of money you need to be reasonably sure of results.

As hosted AI gets more expensive it will become easier to argue. But with the pace of model evolution (and tools surrounding those models) buying local hardware is (at best) a purchase best deferred till later.

Eventually OSS will catch up. But that may be a decade or two away.