While I completely agree that Nvidia's pricing strategy is highly excessive and they're doing it simply because they can, comparing a card that uses a GPU made on Samsung's old, cheap, and low-density 8N process with 6 modules of 15 Gbps GDDR6 with one that uses TSMC's 4N node, paired with 6 modules of 21 Gbps GDDR6X, isn't exactly fair.
The production costs aren't going to be particularly close. The AD104 has a transistor density that's 179% higher than the GA106 (121.4 vs 43.5 Mtrans per mm2) and the 4070 Ti uses 'full die' binned chips; the GA106 in the 3060 has two SMs disabled, so it's in the mid-upper range of the bins. AMD has pointed out that TSMC's node costs
have increased quite a bit, just across two generations.
The 4070 Ti's VRAM is only made by Micron, whereas Samsung, Micron, and SK Hynix all make 15 Gbps GDDR6 -- thus the price of the latter is going to be far more competitive than the former. There isn't much parity between the PCB components either: the old card typically has 5 VRMs for the GPU, whereas the new one comes with around 9.