If Kimi bought a ton of high-priced inference cards and memory cards right now, but by the time they arrive and are up and running, GPT, DS, or another Claude has already surpassed and left them far behind, wouldn’t those inference cards and memory cards end up being a waste or have minimal impact, missing out on the big profits during the model's leading moment? Cloud providers can perfectly address this concern for all big model companies.