Drivers, drivers, drivers.... you know who escapes all the blame in this? Microsoft. They were the problem for everyone except 3DFX who managed to get their own independent API well established in the market... (Although, nobody buys a voodoo for DX compatibility either) Apparently what happened to Rage was that the hardware was tuned to implement every feature of DirectX in harmony with the API... then Microsoft changed it. This then meant ATI and others in development who had also relied on MS's right hand knowing what it's left hand was doing, were left with having to make an ugly software shim of a driver to kludge old API hardware to new API... and since drivers were going to be "cake" if the hardware lined up feature for feature with the software, probably didn't have the talent on hand to do this perfectly or quickly. This I believe also soured relations between Microsoft and many 3D companies, because by the time they brandished lawyers at each other, MS was probably saying "Fine! No more advance info, the product is the product when it releases!"
I think this then added some months to hardware still in development, which is why the "flood" looked too late and too little when other 3D chips launched. Nvidia on the other hand, had launched NV1 early with a different design philosophy. This put them out of phase with DX development anyway, so they were not thrown off by MS being assclowns and maybe had not soured relations by threatening lawyers, thus were able to swerve NV2 into synch with DX while everyone else was still thrashing around.
OpenGL on the other hand was a game of chicken...
An API designed by committee... yay 😒 ... it is/was huge, many features, covered most 3D graphics use cases, so was entirely too cumbersome for gaming in it's entirety. The game of chicken then, was picking the features that would be used in future games, and focusing on making those fast, while not implementing anything that was going to cost you silicon and heat and maybe not be used anyway. So gaming 3D boards were in wait and see mode, afraid to make too bold a move and end up with an unused feature that slowed the board down relative to competitors. Meanwhile professional 3D chips could do it all, or at least most of it, and were stuck with ponderous performance due to having to get the whole herd of elephants to dance the tango. 3DFX was established enough and didn't care enough due to having glide to fall back on, which was very GLish anyway, to worry about making a wrong move, so they could just go "well here's our miniGL subset, have fun"... so that ended up being the core subset of the standard everyone else stuck close to... but not necessarily how other cards would be best used.
Anyway, I think the success of glide, the duplicitousness of Microsoft and the scary behemoth of OpenGL caused more resources than ultimately looked wise were devoted by other 3D companies to internal own API maintenance and development, as a hedge, which might have been redirected to improvement of windows DX drivers. But it probably looked at the time, that that also could be a fruitless path, where success was not guaranteed. Looking back it's easy to think that everything happened "on rails" bringing us to present day, "obviously DirectX was the way", but nothing ever happens like that. There were hardcore DOS gamers in 1998 still who thought DX was a steaming pile that would never get anywhere (It's was kinda 3DFX's core market)
edit: Typo above, NV2 was kinda "more of the same" for different markets than NV1, the NV3 is where they swung back into PC 3D with the Riva 128
Unicorn herding operations are proceeding, but all the totes of hens teeth and barrels of rocking horse poop give them plenty of hiding spots.