It seems like there is a very health space for an MOE targeted GPU where it has essentially an 5070ti with 16gb ish GDDR7 but then also has 128 GB LPDDR5x (or even just DDR5 as expansion dimms on it?). Putting this into the same card would likely reduce the transfer hit when a cache miss happened and the gpu had to load from the slower LPDDR5x. No need to have PCIe 5x16 limiting memory transfer if it is on the card.…
Sounds like Bolt's GPU
their main mem is LPDDR5x with DDR5 SODIMMs for expansion. LPDDR5x is 256GB/s? on the machines you see it implemented on which is a lot faster than the DDR5 expansion. For comparison, gddr7 on a 5070ti pumps ~900GB/s. 4x the tokens/sec if you are memory bound.