Thinking about AI infrastructure returns with primary data. Does it look like Charlie Munger's example of investing in a new textile loom?

Some interesting data points that poke some holes in the AI trade thesis. Microsoft and Amazon had stellar quarters, but a lot of that was based on cloud compute - is that really consumers and businesses buying their end-use AI tools, or is it still part of the infrastructure buildout? Does end-use AI really have enough volume (and enough of a moat) to justify the trillions going into the build?

First: the BIS economist estimate that roughly a third of current AI CapEx represents zero-sum competitive spending rather than market-expanding investment. If correct this is the most damning number in the entire bear case because it means the aggregate return on AI infrastructure investment is structurally impaired regardless of technology maturation.

Second: the complementary investment gap. Historical pattern shows $1 of hardware investment requires $5-10 of complementary intangible investment in process and organizational redesign to capture the productivity gain. US organizational capital investment is declining as a share of GDP and job-switching sits near all-time lows. If the complementary investment wave isn't coming, the productivity thesis doesn't materialize.

Third: the Google advertising exception. I argue that digital ad revenue won't be new money in the system, it'll be search revenue redirected to AI. If I'm wrong about that and AI creates genuinely new advertising inventory that expands the total market, the revenue gap closes faster than my analysis suggests. Anyone with better visibility into digital advertising market dynamics would be useful here.

Full piece: [https://cavemanscreener.substack.com/p/bridges-to-nowhere-part-iv-a-lesson](https://cavemanscreener.substack.com/p/bridges-to-nowhere-part-iv-a-lesson)