Good question. I guess depends on how many users we get. Each users gets their own dedicated GPU. Most video generations systems (and AI systems in general) can share GPUs during generation. Since we are real time, we don't do that. So, each user minute is a GPU minute. This is the biggest driver of the cost.