3dfx built the first real consumer 3D accelerator in 1996. Voodoo Graphics ran dedicated 3D rendering while your 2D card handled the desktop via pass-through cable. Glide API was faster than anything else. They owned 80-85% of the consumer 3D market by 1998.
Then they made the fatal move: bought STB Systems for $141M to manufacture their own boards and cut out partners like Diamond and Creative. Those partners immediately switched to NVIDIA's RIVA TNT and GeForce 256.
GeForce 256 introduced hardware T&L, proper 32-bit color, and tight Direct3D integration. Glide became irrelevant. Voodoo3 shipped in 1999 still doing 16-bit while the market had moved on.
NVIDIA bought 3dfx's assets for ~$70M in December 2000. 3dfx filed bankruptcy in 2002.
The lesson: 3dfx invented the category but tried to own the entire stack. NVIDIA stayed focused on silicon, let partners handle distribution, and used that leverage to build CUDA and the architecture that runs modern AI.
Some AI companies today are repeating the same vertical integration mistake. Owning too much of the present kills your ability to shape the future.
Then they made the fatal move: bought STB Systems for $141M to manufacture their own boards and cut out partners like Diamond and Creative. Those partners immediately switched to NVIDIA's RIVA TNT and GeForce 256.
GeForce 256 introduced hardware T&L, proper 32-bit color, and tight Direct3D integration. Glide became irrelevant. Voodoo3 shipped in 1999 still doing 16-bit while the market had moved on.
NVIDIA bought 3dfx's assets for ~$70M in December 2000. 3dfx filed bankruptcy in 2002.
The lesson: 3dfx invented the category but tried to own the entire stack. NVIDIA stayed focused on silicon, let partners handle distribution, and used that leverage to build CUDA and the architecture that runs modern AI.
Some AI companies today are repeating the same vertical integration mistake. Owning too much of the present kills your ability to shape the future.