· 供应链分析
涉及标的： $IREN$MSFT$NBIS$NVDA$ORCL$WLAC
好问题！超大规模云厂商已签署确切的利用率+支出协议，例如 $NBIS 与 $MSFT 的交易，微软将在5年内支出170-200亿美元。这并非完全是在承担高资本支出(capex)的风险，因为他们是在为产能付费以解决瓶颈。他们仍在支出数百亿美元，部分原因是无法自建，部分原因是前期高资本支出存在更大风险。我在核心论点中写道，Neoclouds（新云）+ 有意将GPU分配给这些Neoclouds是 $NVDA 维持主导地位、实现多元化并避免未来利润率压缩的方式。我们也看到超大规模云厂商在自建方面遇到困难（如 $ORCL 亏损1亿美元），但这恰恰表明Neoclouds在软件编排处理GPU利用率以及通过其建设方式维持高毛利率方面拥有巨大护城河。$NBIS、$WLAC 等毛利率已达70%左右，现在只是执行问题。这不仅是芯片稀缺，这些小型公司+转型HPC的比特币矿工（如 $IREN）实际上拥有竞争优势，即护城河。
Great question! Hyperscalers signed confirmed utilization + spend, for example with $NBIS deal with $MSFT, Microsoft is spending 17-20B over 5 years. It's not exactly risking high-capex, since they're spending to pay for capacity since there's a bottleneck. They're still spending tens of billions of dollars, partly because they can't do it themselves and partly because there's a larger risk in high capex upfront. I wrote in my central thesis that Neoclouds + intentional GPU allocation to these Neoclouds were $NVDA's way of maintaining dominance, diversification, and not risking margin compression down the road. We've also seen hyperscalers have trouble with buildout on their own, (eg. $ORCL losing $100m), but this just goes to show that there is a huge moat with Neoclouds in terms of software orchestration handling GPU utilization and maintaining high gross margins with how they do their buildout. $NBIS, $WLAC, and others already had 70% or so gross margins and it's just execution now. It's not just scarcity in chips, there's actually a competitive advantage that these little guys + Bitcoin miners turned HPC ( $IREN), that is the moat.