Similarly, I'm not clear _why_ Matrox decided not to compete in 3d performance or why all the rest of them (S3 etc) failed to become players there while ATI and nvidia did.
Matrox misfired with the Parhelia. The card was *supposed* to compete with nVidia and ATi on performance (it was a very ambitious design, large chip, first card to support VS2.0, and a 256-bit memory bus, which was considered very wide at the time), but it couldn't deliver on its promises.
As a result, Matrox left the gaming market, and concentrated on specialized hardware only.
PowerVR has a similar story... they tried to go for gaming consoles (Sega DreamCast) and competitive discrete cards for PC, but could not compete well enough, so they concentrated on the emerging mobile market instead, where they are now a big player (even Intel's Atom chipsets use a PowerVR-designed GPU, the GMA500).
Most companies were just blindsided by 3dfx from the get-go, and never managed to deliver a competitive 3d card at all (legendary names such as Tseng Labs, S3, Cirrus Logic, Trident disappeared overnight with the 3d acceleration revolution).
ATi and Matrox were the only ones who could adapt and compete in the 3d acceleration world... and nVidia has been the only successful videocard startup since 3d acceleration.
Technically ATi the company is dead ofcourse, since they were acquired by AMD... so nVidia is the only remaining video card company for mainstream discrete cards. It's been a real rollercoaster ride.
In fact, if we look beyond the PC market... some other important companies also dropped out of the 3d race. How about Silicon Graphics for example? They were one of the first to develop 3d acceleration hardware, and OpenGL is their API. These days they just use off-the-shelf hardware, rather than developing their own, and OpenGL is now maintained by the Khronos group.