Kimi's computing power shortage has made everyone realize the value of cloud providers, but if you think their value is only about computing power, you're underestimating them. The true future value of large-scale CSPs lies in becoming the critical 'middleware' in multi-model architectures for enterprises. Essentially, after commoditizing the 'model layer,' CSPs shift more of their value to the infrastructure + platform service layer (hosting, optimization, routing, governance, security), not just computing power. In other words: Large model ARR is 'model layer revenue'; CSP performance is 'AI engineering revenue.' This means CSPs are helping clients reduce costs while earning a higher profit margin for themselves. Previously, this post https://(x.com)/qinbafrank/status/2074754779755295164?s=46&t=k6rimWsEbo2D2tXolYcM-A discussed the value re-evaluation of large-scale CSPs, and it was probably one of the earliest mentions of this: large-scale CSPs have computing power, can deploy open-source models, and can develop their own models. They no longer rely solely on cutting-edge closed-source large models, which leads to higher gross margins. Their massive B-end customer channels still hold significant advantages, and all of this will create new strategic advantages for large-scale CSPs. In the future, model companies will decide 'what AI can do'; CSPs will decide 'how AI can scale, reduce costs, and achieve high reliability in enterprise production systems.' For more details on AI adoption engineering, check out last week's long-form article, which dives deeper into this topic. This post is sponsored by @bitget_zh, 'Bitget Buy US Stocks: Instant entry, seamless trading.'