If this is just designed for AI economics. The end output of these datacenters are the tokens which are mostly relatively really small data and is not sensitive to latency.