What a weak ass recover $SOXX with 10yrs treasury dragging. But it's ok. I want to share a little bit history and thought about this AI Infra sell-off. In 2012, three of ASML’s most important customers helped finance the technology they would eventually depend on: EUV. Intel acquired 15% of ASML, TSMC 5%, and Samsung 3%. Together, they committed approximately €1.38 billion toward EUV and 450mm lithography development over 2013–2017. The customers shared development risk, while ASML retained control over development and its intellectual property. That arrangement helped align the equipment supplier’s technology roadmap with its customers’ Future Manufacturing NEEDs. It was primarily an R&D program; the equity subscription proceeds were returned to existing shareholders through a special buyback. If you're interested, check That history provides context for the current AI infrastructure cycle: investment across this supply chain depends on companies making commitments years before the final computing capacity becomes available. Demand commitments travel in this direction: Hyperscalers and AI companies → NVIDIA, AMD and Hyperscalers → foundries and memory manufacturers → equipment suppliers such as ASML. ASML’s order growth requires customers to turn their forecasts into equipment commitments, then this is ASML contract liabilities at Sept 27: €9.30 billion, versus €3.57 billion in June Meanwhile, all foundries expansion in high speed: - TSMC’s Fab 20 in Hsinchu and Fab 22 in Kaohsiung are locations for its multiphase 2nm expansion - TSMC’s Phoenix campus, at 5088 West Innovation Circle, is expanding beyond its operating N4 fab into additional facilities for more advanced processes. - Samsung’s Taylor campus in Williamson County, Texas, near Samsung Highway and FM 973, represents an initial minimum investment of $17 billion - Intel’s Fab 52 at its Ocotillo campus in Chandler, Arizona ASML’s warning that planned equipment supply falls short of customer demand through 2028. Here is my thought, we have power constraints in states where could persist in some regions no access to electronicty, while chip constraints tighten after 2028, maybe even early. If powered capacity accelerates faster than chip supply, late 2028 could mark another swing — from securing electricity to securing the silicon needed to use it. The last argument I’m thinking about is that $CRWV, or whichever sensibly leveraged neoclouds, may be a drop of bad blood. Hyperscalers would be happier if they went bankrupt. Moreover, it’s kinda good for the AI infra trade if $CRWV gets abandoned by hyperscalers.