1Dense vs SparseDense and sparse are two ways to describe how much of a space is filled, and the choice shapes cost, speed, accuracy, and design.Frontier 2Routing & Load BalancingHow requests find the right place to go, how traffic is shared, and what trade-offs hide behind simple words like route and balance.Frontier 3Expert ParallelismExpert parallelism spreads a sparse model’s specialist parts across machines so it can grow without asking every machine to do every calculation.Frontier 4Serving MoEMixture-of-experts models can be cheaper per token, but serving them well is a routing, memory, and operations problem.Frontier