30 days. One map. The AI buildout, end to end: Memory → Networking → Power → D… · @Eyescool · TwKit

@Eyescool

30 days. One map. The AI buildout, end to end: Memory → Networking → Power → Data Centers → the stocks that price it. What the month proved: – The spend is real: ~$733B of 2026 Big-4 capex guides, ~$800B by one count, $957B+ across nine CSPs. Different scopes, same direction. – The payback math is brutal: ~$300B/year to break even, ~$417B for a 10% return (Goldman). Cloud is ~$70B above trend. The gap is the trade. – The funding models split: Oracle's customer prepay, Bloom's float, Tesla's self-funding, NiSource's ratepayer model. – The constraint moved: from chips to packaging (TSMC: "limiting our customers' growth") to power (data-center demand +17% in 2025 vs +3% for the grid) to credit (CoreWeave CDS >1,100 bps). The map's lesson: follow the cash, not the press release. Revenue is the entry ticket. Cash flow decides who stays in the building. Next: the 2027 capex guides. They decide whether this map compounds or reprices. What did this month change in YOUR AI thesis? $NVDA $TSM $MSFT $GOOGL $AMZN $META $ORCL $BE $VICR #AIInfrastructure #DataCenters #Valuation

曝光 7 · 评论 0 · 点赞 0 · 书签 0