GPUs in every home would be wildly inefficient. Hyperscalers have massive economies of scale by consolidating the inference. That's why it's cheaper to pay per api rates for deepseek than it is to run it locally on your home server and pay for electricity, even if your server is FREE.