Production of something that complex isn't trivial or quick to ramp up, nor is there an unlimited production ceiling. And that's a problem, because there is effectively an unlimited demand for cryptocurrency mining equipment compared to other GPU markets.
I might buy the latest GPU to play PC games at max detail. But there's no increase in value in me buying 20 GPUs so there's effectively a cap on how many GPUs the gaming market will demand. Same goes for other use-cases like scientific computing. In those cases ore GPUs do allow them to solve harder problems or finish calculations faster, but buying more GPUs is more expensive so eventually scientists hit the limits on their budgets.
Cryptocurrency mining is different though since it provides a direct path to turning more computing power into more income. As long as the marginal costs to mine on a GPU are below the expected return and the payback time for the fixed cost of buying the GPU in the first place isn't too long, it nearly always makes sense to buy more GPUs. To put it another way, if it makes economic sense for me to mine on 1 GPU, it makes sense for me to mine with 10, or 100, or 1000. In fact it probably makes more sense to mine with more GPUs since there are economies of scale involved. And since miners are more concerned with their marginal costs than the fixed costs, even huge increases in GPU prices don't alter their behavior much. As long as you're making money running a card, who cares if the payoff time for it increases a few months?