· 供应链分析
既然有人@我，我不同意你作为TAM（总可寻址市场）基础的根本假设，但你的观点有合理之处。 AI算力增长实际上是一个结构性市场，目前正因前沿大语言模型(LLM)的使用而呈指数级增长，并将随着应用AI（机器人等）继续增长。暗示MSFT/Google的5-10年合同确认的使用量“溢出”（临时流动性）会消失是误导性的。 如果你想改述为Neoclouds目前捕获的溢出在超大规模云厂商完成建设后可能不存在，我同意。5-10年后我会更担心（在GCP 10年合同和Azure 5年合同结束后）。特别是当$AVGO与超大规模云厂商的定制ASIC、TPU/Tranium等+建设完成后。（作为交易者，我在2年周期内看涨，5年以上则不那么看多）。 第二点是试图对AI建设这样投机性的事物进行TAM的定量建模是错误的研究方法。 此外，10-15%的数据过于悲观，因为它忽略了如果AI算力是增长的结构性市场，超大规模云厂商的重复溢出。但你基于电网容量会跟上的假设，我不同意，这没问题。
Since someone pinged me, I disagree on your fundamental assumption that you're basing TAM off of, but valid points. AI compute growth is actually a structural market that's growing exponentially right now fro new frontier model LLM usage, and will continue to grow from applied AI applications (robotics, etc). Implying MSFT/Google's 5-10 year contracts confirmed usage contracts "overflow" that will vanish (temporary liquidity) is misleading. If you wanted to reword it in saying the overflow that Neoclouds captured now might not exist after hyperscalers complete their buildout, sure. I'd agree with you, next 5Y-10Y out I'd be more worried (after GCP 10y contracts end of Azure 5Y contracts end). Especially when $AVGO x Hyperscaler custom asics, TPUs/Tranium, etc + buildout gets complete. (As a trader I'm bullish on a 2 year timeframe, not so much 5Y+ plus). Second thing is trying to quantitatively model TAM of something so speculative such as AI buildout is the wrong way to approach it. On top of that, the 10-15% figure is overly pessimistic because it ignores repeat-overflow from hyperscalers if AI compute is a growing structural market. But you're going off the assumption grid capacity will catch up, which I disagree, and that's fine.