Tokenmaxxing, SF Street Name Auction, Corporate Retreat Gone Wrong | Riley Walz, Aditya Bandi, Zach Shore, Hongwei Liu, Zak Kukoff, Thomas Laffont

The episode delivers a wide-ranging tech and finance update centered on AI infrastructure spending, hardware supply-chain moves, corporate missteps, and several funding announcements. Discussions highlight Meta’s internal push for AI-native operations alongside rapid revenue growth at frontier labs, while guests provide on-the-ground detail from defense aviation, indoor mapping, and late-stage venture investing.

Meta’s Token Usage and Vertical Integration Push

Meta employees have been competing on an internal leaderboard tracking Claude usage, with the company burning roughly 60 trillion tokens over a 30-day period. Back-of-the-envelope calculations place the associated spend in the range of several hundred million dollars annually at current API rates, far below earlier headline estimates that assumed maximum output pricing. The activity is framed as part of a broader directive to make the organization AI-native, with internal tools such as “My Claude” and acquired assets like Manus running on frontier models. One line of analysis suggests the spend could be amortized across training runs for future Meta models, creating a vertically integrated inference and data-generation loop even if enterprise contracts limit direct training on third-party outputs. Observers note the incentive risks of pure token-volume metrics, echoing classic Goodhart’s-law concerns.