Where the limits are set by hardware for agentic execution (compute/network/storage) && inference speed