Exponential Learning

Hosking Partners' Q3 2026 letter, via MastersInvest: a London value manager beat global markets by 6.3% YTD owning almost none of the AI winners — and its diagnosis of the AI buildout is the sharpest we've read. Cap ex doubled to T (a railway-boom share of GDP); 00B+ of IG debt issued in 2026 alone, .1T more by 2030; credit investors now run the industry through SPVs and off-balance-sheet leases while token prices fall. The core claim: the data centre is a factory, compute is a commodity — load-the-plant economics will push token pricing toward marginal cost, the way solar was commoditized (and China-dominated). The .5T OpenAI/Anthropic flotations are the exam. Plus: how they won without AI (Greek refiners, Norwegian tankers +440% since 2017, Saga's 25% ROIC, 80 Japanese small caps), the honest 260bp cost of missing mega-cap tech, a full steelman of the other side, and why the factory thesis actually CONFIRMS the energized-land view — factories commoditize, tollbooths don't.

What is Exponential Learning?

Humans can absorb multiples more audio content than written. To that end, this podcast is for absorbing as much intelligence as possible as the catch-all for interesting articles, research reports, or podcast summaries. I hope you enjoy the knowledge as much as we do.