· 方法论
涉及标的： $AMZN$CRWV$MSFT$NBIS$UPWK
我发现像 @AskPerplexity、Grok 等 AI 工具在研究 Neoclouds（新型云服务商）和每日新闻频发的板块时，结果往往极不准确。它们会遗漏大量细节，例如 $NBIS 的白皮书中有关于与 $CRWV 等对比的利用率信息，这些数据对 AI 不可见（暗示了 GPU 利用率带来的毛利率增长）。当 Meta 与 $CRWV 达成 140 亿美元交易，且上周 $AMZN、$MSFT 等发布财报显示 AI 资本支出增加时，AI 可能会忽略那些未直接提及 $NBIS 但暗示超大规模云厂商交易增加的顺风因素。或者几天前的额外降息提振了涉及债务的 Neoclouds 的前瞻性盈利。亦或是关于 Clickhouse 等子公司或投资组合公司的新信息，AI 可能无法将其串联起来。FinX 散户在将这些信息拼凑成前瞻性增长方面做得更好，鉴于 AI/Neocloud 的总可寻址市场(TAM)具有高度投机性和快速增长，很难建模。这也是我采用估值方法的原因（在我之前关于 $NBIS 如何达到 1000 亿美元市值的帖子中），但如果看 $UPWK 等更标准的业务，我的估值建模方式则不同。
I've actually found AI tools like @AskPerplexity, Grok and others to be extremely off when looking at Neoclouds and sectors with new news every day. They miss a lot of details, eg. with $NBIS you have utilization information from Whitepapers comparing Nebius to $CRWV and others that aren't publicly viewable to AI (which insinuated growing gross margins from GPU utilization). When Meta struck a $14B deal with $CRWV and you have earnings report last week from $AMZN, $MSFT, and others you have increased capex spend into AI, and they might miss tailwinds that don't mention $NBIS directly with increased changes of more hyperscaler deals. Or when there's an additional rate cut a few days ago that boosts forward earnings especially with Neoclouds that involve debt. Or when there's new information about subsidiaries or portfolio companies like Clickhouse that AI might put 1+1 together with. FinX retail does a lot better job with piecing all this information together into forward growth and AI/Neocloud TAM is really hard to model given how speculative and rapidly growing it is. That's kinda why I go with valuation approaches (in my previous post on how you get to $100B MC for $NBIS), but if you look at $UPWK and more standard businesses. I do valuation modelling differently