Two factors:
- size doesn't matter (that much): For CPUs, you are creating large (relatively speaking) rectangular
objects on a circle. Thus, the closer your circle outline approaches a straight line (i.e., the larger the circle is),
the less waste silicon you have to cut away. For tiny LED dies, you can basically use the full circle even on
smaller wafers.
- For CPUs, the price is in the processing; for LEDs, it's the materials: LED structures are gigantic compared
to those found on a CPU. They are built using a handful of low precision (again, relatively speaking) production
steps. CPU production is as close to magic as we can nowadays get and uses dozens to hundreds of processing
steps (a wafer for a modern CPU spends about a month inside the fab), where a machine for a single step costs
millions. The cost for the raw wafer itself really doesn't matter.
So:
LEDs are cheap and fast to manufacture, and basically fit on arbitrary wafer shapes:
Raw material prices matter, wafer waste is minimal and doesn't. --> cheap, small wafers.
CPUs are damn hard and slow to manufacture, really only fit on rectangles:
Who cares about wafer prices, but we want optimal yield per processed wafer, so as to have
as little waste as possible --> wafers as large as possible, damn the cost of the wafer.
And lastly: There is no structure size progress in LEDs as there is with CPUs: Once you have a
factory for LEDs, you can use it for a while without upgrading to the newest and smallest process
all the time. So why buy new equipment if the old is perfectly competitive?