Interactive LLMs (chat, copilots, agents) with strict latency targets Long‑context reasoning (codebases, research, video) with massive KV (key value) cache footprints Ranking and recommendation models ...