Skip to content

Meta Tokenmaxxing, Intel Joins Terafab, Frontier AI vs. China | Diet TBPN

TBPN

Hosted by John Coogan and Jordi Hays

33m
Transcript

Original podcast

Where to listen

Summary

Written by a model from the transcript; not yet reviewed.

Meta employees are competing on an internal leaderboard called Claudinomics for token usage, with 60.2 trillion tokens consumed over a 30-day period. Tyler runs back-of-envelope math showing the actual cost is roughly $136 million per month at OpenRouter's input-to-output ratio, or as low as $55 million per month if Meta's usage skews toward cached tokens like Claude Code does—far below the one-third-of-Anthropic's-ARR figure that circulated. John Coogan builds a bull case that this internal spend justifies Meta's Superintelligence Lab as vertical integration: if they can amortize training costs over their own multi-billion-dollar token bill, they don't need a standalone consumer product.

The episode also covers Intel joining TeraFab alongside SpaceX, XAI, and Tesla to design, fabricate, and package chips in Austin, Texas, targeting one terawatt per year of compute. Anthropic is making a preview of its cybersecurity model Mythos available to about 50 critical infrastructure organizations including Amazon, Microsoft, and Apple. Finally, OpenAI, Anthropic, and Google are collaborating through the Frontier Model Forum to detect adversarial distillation attempts by Chinese competitors, a rare truce amid ongoing inter-lab tensions.

  • The 60.2 trillion tokens at Meta likely cost around $136 million per month rather than the one-third-of-Anthropic's-ARR figure, because 98.9% of tokens in coding agents are input tokens at a much lower rate. 5:17
  • Meta's internal token spending could justify their Superintelligence Lab as vertical integration, since they'd amortize training costs over their own multi-billion-dollar internal token bill rather than needing a standalone product. 9:19
  • Intel's TeraFab partnership with SpaceX, XAI, and Tesla targets one terawatt per year of compute from a single facility in Austin, Texas. 12:34
  • OpenAI, Anthropic, and Google are sharing information through the Frontier Model Forum to detect adversarial distillation attempts that violate their terms of service. 30:53
  • Anthropic is making a preview of its cybersecurity model Mythos available to about 50 companies maintaining critical infrastructure, including Amazon, Microsoft, and Apple. 28:13

Highlights

  • 0:02

    Claudinomics Token Leaderboard Revealeddata

    Meta employees are competing on an internal leaderboard called Claudinomics, with 60.2 trillion tokens used over 30 days, sparking debate over how much Meta is actually spending with Anthropic.

  • 5:17

    98.9 Percent of Tokens Are Inputdata

    Tyler cites OpenRouter data showing 98.9% of all tokens are input (including cached), meaning the vast majority of coding-agent token usage is cheap input rather than expensive output.

  • 8:46

    Meta Vertical Integration Bull Caseinsight

    John argues that at a $250K AI budget per engineer, Meta's internal token spend is large enough to amortize the training cost of a frontier model without needing a standalone consumer product.

  • 10:21

    Token Maxing as Distillation Coverinsight

    Jordi proposes that the token-maxing leaderboard is a smokescreen while Meta is actually generating trillions of tokens of frontier model output to build a training corpus for their next model.

  • 12:34

    Intel Joins TeraFab Projectstory

    Intel announced it will work with SpaceX, XAI, and Tesla to design, fabricate, and package ultra-high-performance chips at scale, targeting one terawatt per year of compute from a single Austin facility.

  • 20:22

    Plex Honduras Retreat Disasterstory

    Tech company Plex took 120 employees to Honduras for a Survivor-themed corporate retreat that devolved into a CEO with E. coli, a former Navy SEAL hazing unfit office workers, a porcupine through the ceiling, and a fire ant hill on the golf course.

  • 28:13

    Anthropic Mythos Cybersecurity Modeldata

    Anthropic is making a preview of its model Mythos available to about 50 critical infrastructure organizations including Amazon, Microsoft, Apple, and the Linux Foundation to find and patch hardware and software bugs.

  • 30:53

    US Labs Unite Against Chinese Distillationinsight

    OpenAI, Anthropic, and Google are sharing information through the Frontier Model Forum to detect adversarial distillation attempts by Chinese competitors, a rare collaboration amid ongoing inter-lab tensions.

About this episode

Diet TBPN delivers the best of today’s TBPN episode in 30 minutes. TBPN is a live tech talk show hosted by John Coogan and Jordi Hays, streaming weekdays 11–2 PT on X and YouTube, with each episode posted to podcast platforms right after. Described by The New York Times as “Silicon Valley’s newest obsession,” the show has recently featured Mark Zuckerberg, Sam Altman, Mark Cuban, and Satya Nadella. Follow TBPN: https://TBPN.com https://x.com/tbpn https://open.spotify.com/show/2L6WMqY3GUPCGBD0dX6p00?si=674252d53acf4231 https://podcasts.apple.com/us/podcast/technology-brothers/id1772360235 https://www.youtube.com/@TBPNLive

Transcript

0:00
33:02

6,317 words · 3 speakers

Speaker names marked are inferred from speaking time and the show’s known hosts — they are not voice-verified.

Topics

  • AI Token Economics
  • Meta Superintelligence Lab
  • Chip Manufacturing
  • AI Safety
  • Model Distillation
  • Corporate Culture
  • Anthropic Revenue
  • TeraFab

Page updated · transcript and summary are machine-generated