alt Hacker NewsMaybe, but that's not a great analogy. The standardized, user-accessible sockets mean many different CPUs can be paired with many different motherboards. There's also a wide variety of sizes and features in motherboards, plus they have buses for connecting various kinds of peripherals. GPUs have none of this flexibility or extensibility.
Yeah, but you’re missing the specialization angle.
NVIDIA and Intel as companies are specialized in the design (and in the latter case, manufacturing) of chips. Board OEMs are specialized in making a consumer-ready product, maintaining worldwide sales and distribution channels, and consumer relations.
Of course, it wouldn’t be impossible for NVIDIA to start doing these things on their own (see Apple, who designs chips, designs computers around those chips, and operates retail stores where those computers are sold), but presumably NVIDIA prefers the current arrangement, where they can just focus on the chips and leave the rest to OEMs.
See also Intel under Gelsinger, who sold off the NUC and server lines (finished products) to focus on the core business (x86 chips).