Open-Source AI Fight, Anthropic's $1.5B Settlement, NYC Eviction Politics

Chapters

1. 0:00 Policy and regulation

   White House weighs Chinese open-source AI ban

   China's Moonshot AI released Kimi K3, an open-source model on par with Opus 4.8 and GPT-5.6 at roughly 50% lower cost, triggering a DeepSeek-style panic. Axios reported the White House is considering banning Chinese open-source models, while Wired reported Howard Lutnick opposes such a ban; Polymarket odds of a 2026 ban jumped from 22% to 45%. Sacks says on good authority no decision has been made, and he warns that banning open source would backfire and punish American developers. He accuses Anthropic of regulatory capture, arguing it could stop distillation with KYC but instead wants government protection.

2. 7:54 AI and tech

   Distillation, commoditization, and market stakes

   Chamath explains distillation as taking another model's outputs at massive scale to train your own model, notes KYC and card controls would stop it, and says everyone from Anthropic to OpenAI to Chinese labs has distilled. He argues AI models are commoditizing faster than any sector in Silicon Valley history, and a US open-source ban would force firms like Coca-Cola to use government-selected models costing 50 to 100 times foreign alternatives, rerating the whole market. Friedberg says banning open source in America would not stop the world from using Chinese open models, so the US would still lose.

3. 14:39 AI and tech

   The free-speech and innovation case for open source

   Friedberg makes the free-speech and innovation case for open source: distillation looks at outputs, not stolen code, and OpenAI and Anthropic already train on the world's content. He compares open-source AI to Apache and Firefox, arguing open infrastructure let Google, Amazon, and millions of sites capture value, while closed internet gatekeepers would have throttled growth. Open AI, he says, diffuses value across the economy and blunts the socialist critique of billionaire control. He says the government should pursue real copyright violations and trade sanctions but not ban open-source models.

4. 23:53 AI and tech

   Frontier-lab economics and IPO threat

   Chamath says foundation-model pricing power is evaporating in months, not years, citing open-router data showing most tokens are now dark or self-hosted; he argues real value is shifting to cloud and silicon. Sacks counters with revenue estimates showing Anthropic up from $10B to over $70B ARR and OpenAI from $33B to $41.3B by July, and calls the regulatory-capture campaign a LeBron-style flop ahead of IPO roadshows. Calacanis pushes back that startups like Lovable and 11 Labs are leaving frontier APIs and that open source will create margin compression that derails Anthropic and OpenAI IPOs.

5. 41:34 Policy and regulation

   China's long game in the molecule economy

   Friedberg zooms out to China's long game: the US built trillions of dollars of GDP on the knowledge and services economy, while China built roughly 20 times the manufacturing capacity and 8 times the electricity generation. If open-source AI flattens the value of bits, the remaining value is molecule conversion — physical manufacturing and energy — where China has an overwhelming structural advantage. The hosts treat open-source commoditization not just as a startup issue but as a global power shift, while noting enterprises still must weigh data sovereignty against the implementation work of open source.

6. 45:46 Deals and companies

   Anthropic settles copyright suit for $1.5B

   Anthropic agreed to pay $1.5 billion to settle claims that it trained Claude on 7 million pirated books, the largest copyright settlement in US history; authors receive roughly $3,000 per book and lawyers $101 million. Sacks says Anthropic lost leverage because it did not buy even one copy, yet still claims fair use allows training on creators' work. Chamath argues Anthropic invented the industrial-scale distillation attack narrative to protect its IPO, and notes its own blog never calls distillation IP theft. He calls the company's position completely hypocritical.

7. 55:43 Policy and regulation

   Fair use, books, and training-data licensing

   Friedberg argues knowledge is diffuse: if an author opts out of AI training, an engine can still learn from reviews and metadata about the book without ingesting it, which he sees as copyright-compliant. Calacanis responds that direct competition cases such as Thomson Reuters v Ross and the music industry create real exposure, and advises AI companies to pool 10% of revenue for licensing. Sacks warns Anthropic and OpenAI have already told the government that training on creators' output without consent is IP theft, effectively confessing their own product is stolen; a better case would focus on deceptive Chinese account creation.

8. 1:06:59 Markets

   Google, Tesla, and SpaceX capital spending

   Google and Tesla posted strong quarters but fell sharply on capital spending. Google Cloud reached a $100 billion run rate, capex guidance rose to $195-205 billion, and free cash flow turned negative for the first time as a public company; Tesla's capex surged 140% year over year to $25 billion. SpaceX trades at $1.5 trillion, down 30% from its day-one close. Chamath remains bullish on Google, citing 32% average ROIC and model proliferation feeding its cloud and silicon businesses. Calacanis contrasts Apple's roughly $900 billion in buybacks and dividends.

9. 1:17:25 Policy and regulation

   NYC evictions, rent control, and private property

   At a New York City rental hearing, an activist called evictions violence, and Mayor Zohran Mamdani's plan would bar landlords from charging application fees and from combining credit checks with income requirements. Friedberg quotes John Quincy Adams on property and liberty, arguing socialism starts by branding owners as evil and ends in tyranny. Sacks says eviction bans hurt long-standing tenants by letting problem tenants degrade buildings and public spaces. Chamath says the policy will raise rents because landlords will demand prepayment and higher starting rents, while Austin-style permitting reform is the actual cure.