· 供应链分析
涉及标的： $CIFR$CRWV$IREN$MSFT$NBIS$WULF
我觉得你有点混淆了这些术语，让我澄清一下。 全栈服务(Full stack offering) = AI云 $NBIS $IREN $CRWV 裸金属(Bare Metal) = $CIFR, $WULF 类型的服务（AWS将自家GPU接入 $CIFR） $IREN 在决定为 $MSFT 交易购买数十亿美元GPU时，转向提供全栈AI云服务。这导致了因执行风险而产生的特别股息(DG)。 $IREN 在财报电话会上明确表示，他们之所以做全栈（包括GPU采购+软件编排），是因为“更高的收入潜力和更强的回报”。 关于你提到的软件编排，它主要用于改善内部运营支出(opex)和利润率，但在面向客户的差异化方面略有优势。 与电力相比，GPU利用率和编排对最终利润率的影响巨大，我在之前的帖子中做过定量分析。 对于为 $NBIS 等支付裸金属服务的客户，他们购买的不仅是原始硬件，而是经过优化的软硬件堆栈。
So think you're confusing the terms a bit, let me clarify. Full stack offering = AI Cloud $NBIS $IREN $CRWV Bare Metal = $CIFR, $WULF type offerings (AWS plugs in their own GPUs into $CIFR) $IREN is pivoted to doing a full stack offering AI cloud when they decided to buy billions in GPU for the $MSFT deal. This is what led to the DG due to execution risk. $IREN on their earnings call said it themselves that they did full-stack (including GPU purchases + software orchestration) due to "higher revenue potential and stronger returns" For software orchestration that you mentioned, it's mainly for improving internal opex and margins, but is a slight advantage for customer facing differentiators. GPU utilization and orchestration plays an incredible difference in terms of final margins, when comparing to power and I did a quantitative breakdown in an earlier post. For customers that pay for bare metal for $NBIS and others, they are paying for an optimized hardware and software stack, not just raw hardware.