Rumor is that an interposer alone is $20, and HBM is also rather expensive. Basically, you'd easily double the cost of these $99 chips if you put HBM2 onto them... maybe triple the costs if you consider that AMD should charge a margin and actually make a profit from these designs.
After that, the GPU would be too-strong on memory and too weak on compute. The minimum HBM speed 1024-bits / 250GB/s from one stack. Yeah, the same memory-bandwidth as a RX 580 with 2304 shaders, while the APU 2400G only has 704 shaders.
The Intel + AMD chip manages to solve the problem: the CPU, GPU, and HBM2 are all on the interposer. So the expensive interposer has a shared design. Second, the GPU is a fully dedicated Vega design that's roughly the size of an Rx 580. That way, the dedicated-GPU can "keep up" with the super-fast HBM2 RAM.