All In Podcast Episode 278 Recap and Notes June 26 2026

June 26 2026

• Friedberg out, Travis Kalanick/Gavin Baker in
• Socialists Sweep NY Democratic primaries

  • 10th district - Brad Lander

  • 13th district - Darializa Avila Chavalier

  • 7th district - Claire Valdez

  • Clear response by voters to their message - its working

  • DSA platform is quite “out there” - mainstream Democrats seem to have lost their party

  • DSA co-chair is quoted as saying they are using the Democratic Party as a ballot access vehicle

    

  • How related is this to importing mass migrants in the previous administration?

  • Did South American right wing election winners win because USAID got defunded?

  • Truth and Justice are an immune system for society

    • When the immune system is compromised, social ills go on the rise

  • Communism is in all of us

    • have you ever been lazy and wanted something for free
      • If this is the way you want to live your life - you get Communism

  • Voting base of DSA is downwardly mobile wealthy educated white liberals

    • They are losing votes with people of color and poor people

  • DSA is winning because of Zohran - a singularly talented politician

  • Dems have no one else on his level

  • Social media ban on < 16 year olds in Canada, UK, Australia might cause a change in those countries’ attitudes towards Communism

    • There is a derivative of this ban that is about the adults, not the kids - its about restricting anonymous accounts on X which is more about controlling free speech

  • When you elect a Republican DA, all cause mortality for young black men lowers by 7%

  • DSA are winning in LA area also

  • Lander beat Goldman mainly on the Israel issue

  • Israel has lower approval among younger voters

  • Seems they have a branding problem right now?

  • A lot of Democrats who voted for Trump were radicalized during Covid when they heard what their kids were being taught in schools

  • Need to make sure kids know that America is not evil and to not hate their country

• Chinese open source models catching up to American frontier models
  • GLM 5.2 is very good
    • Apparently some distillation has been done?
    • Best open weight model for coding
    • Similar to previously strongest frontier models (like Opus 4.8, etc)
    • But now latest frontier models are being held back (Fable/GPT5.6, etc) so that will aid China’s catching up in the race
  • On some level it seems Anthropic has got what it wanted - getting regulatory capture?
  • Clearly we’re on shot clock vs the China companies
  • Chinese companies are also being ordered to train on Huawei chips - if that proves the Chinese stack out - US will lose the opportunity to own the international market
  • Clearly no one will have monopoly on the best AI - so you will likely have a council-of-models
  • These open source models are good for the infrastructure providers because they shift some of the value down to them (from the frontier models taking their cut)
  • Could we have composable models?
  • Nvidia might be the US open source champion (might put them in competition with their customers - similar to the model companies making their own chips)
• Micron earnings to the moon
  • Entire 2026 supply is sold out
  • Revenue up 4x YoY ($9B → $42B)
  • Big jump in guidance
  • Stock is up 10x
    • Gavin Baker called this in the predictions episode
  • Crazy Apple price increases (memory related likely)
  • DRAM is the most important bottleneck
  • Floor pricing in new contracts is ahead of Peaks in previous contracts
  • These stocks are still cheap relative to AI
  • Memory/DRAM will be 30-40% of AI hyperscaler CapEx
  • So now AI is hurting people by increasing prices of consumer electronics also
  • TeraFab can’t come soon enough - can Elon deliver?
  • 1GW data center takes $35B silicon and $25B power and cooling
    • This is where the Space DataCenter case might make sense if its less than $25B for power and cooling
    • Could be $5B for orbital compute - this is deflationary likely and the $25B terrestrial cost is likely inflationary
  • MegaPod trademark filed (seems like a data center in a box)
  • Data centers have man traps (to illustrate how high the security requirements are)
  • Everything that cuts against training (putting so many MegaPods far away from each other - even if you have your own network) works for inference (distributed works for it)
  • Prefill (how much can you remember) and Decode (how fast can you speak) are being disaggregated (Groq, Cerebras, etc)
  • What if Elon packs distributed compute into Powerwalls - so you get free battery in exchange for contributing compute to him
• IPOs
  • SpaceX liquidity event (employees - like half of them bought at IPO) - so hard to say if shorting will work or not
  • Cerebras - lot of selling is due to stock hitting deal price for a lot of holders
    • Their December OpenAI deal might not generate returns for them until Q3
    • Investor(s) watching how fast they can bring power on - apparently its ~600MW - which is very hard to do (only SpaceX AI, CoreWeave, Crusoe might be non-hyperscalers who have brought online ~1GW of power)