SemiAnalysis 正在批评一份关于生成式人工智能投资回报率 (ROIC) 的摩根士丹利报告，并建议调整资本支出计算。虽然摩根士丹利的报告总体上看好 GenAI ROIC，但 SemiAnalysis 认为该报告对 IT 资本支出和数据中心资本支出的假设不准确。SemiAnalysis 提出了这些支出的替代数字，这将略微降低预测的 ROIC，但仍表明回报强劲。 AI
影响 为人工智能基础设施投资提供了更准确的财务前景。
排序理由 该集群由对财务报告的批评组成，而不是主要公告或发布。
AI 生成摘要 · Google Gemini · 来自 4 个来源。 我们如何撰写摘要 →
Instead – we suggest that analysts go with $38/W for IT Capex but rather than use the full $14/W for Datacenter capex, instead only factoring cost for 5 out of the 15 years of useful life – so about $4.7/W for a grand total of ~$43/W. This will only drop ROIC slightly from what
However – they seem to have backed into the correct answer anyway thanks to another error in figuring the invested capital! A $12.1/W NOPAT and a 31% ROIC works out to an invested capital of $40.3/W. However – they got there by assuming $23/W of IT Capex and about ~$16/W of
While this is a good starting point and a solid framework for attacking this problem – there are a couple of areas that we think our friends at MS could fine tune. For starters – the analysis assumes an IT capex/W of $23B/GW – this is quite far off the mark! In the InferenceX
Satya Nadella recently cited Morgan Stanley’s ROIC report as demonstrating the returns are there for Token As a Service (TaaS) businesses – a 31% ROIC for Hyperscaler IaaS business and another 46% ROIC at the TaaS layer! (1/4) https://t.co/LOr7ZMYXpW