The Fight Over Open Source AI, Anthropic's $1.5B Payout, NYC Socialists: Evictions = Violence?

What It Covers

The All-In hosts debate whether banning Chinese open source AI models like Kimi K3 would harm American developers, analyze Anthropic's $1.5B copyright settlement, examine Google and Tesla's massive CapEx investments driving negative free cash flow, and critique New York City Mayor Mamdani's rent control policies that prohibit landlord credit checks and effectively restrict evictions.

Key Questions Answered

• •Open Source Ban Economics: Banning Chinese open source models would impose a 50-100x token cost premium on American enterprises versus global competitors. Using Coca-Cola as a framework: forcing US companies to pay $50 for what competitors buy for $0.50 structurally disadvantages every American business using AI, ultimately compressing margins across the entire S&P 500 and triggering broad market repricing. The policy would harm the very companies it claims to protect.
• •Distillation Hypocrisy Framework: Anthropic and OpenAI simultaneously argue they can train on all public content under fair use while claiming Chinese labs committed IP theft by training on their outputs — an identical process. If stopping distillation is the actual goal, the solution is implementing KYC verification on model accounts, not banning American developers from using public domain models. Anthropic's failure to KYC customers prioritizes revenue growth over security.
• •Regulatory Capture Playbook: Anthropic grew from $10B ARR in January 2025 to over $70B ARR mid-year — the fastest scaling tech company ever recorded — yet simultaneously lobbies for government protection against competitors. The strategic tell: if distillation were the real concern, Anthropic would push to ban Chinese access to American models at the source, not ban American developers from accessing Chinese outputs already in the public domain.
• •AI Value Chain Compression: Frontier model margins face structural compression as open source models now match closed models on 95% of tasks. The value is migrating to two layers: infrastructure (cloud providers, chips) and applications. Google Cloud, already on a $100B annual run rate, exemplifies infrastructure capture. Companies like Cursor built competitive coding tools by fine-tuning Kimi K2.5 on proprietary data — demonstrating the open source fork-and-specialize model works at production scale.
• •Google CapEx Signal: Google's 32% average return on invested capital over 20 years since IPO provides the baseline for evaluating its current negative free cash flow — the first time since going public. When a capital allocator with a 32% ROIC track record accelerates infrastructure spending, the historical pattern suggests this is a buy signal rather than a warning. Google Cloud's model-agnostic architecture positions it to profit regardless of which AI models win.
• •Open Source Internet Analogy: The open source AI trajectory mirrors the 1990s browser wars: Netscape's proprietary browser lost to open source Firefox and Apache HTTP server, and all accrued value shifted to application-layer companies like Google, Amazon, and eBay. Applying this framework to AI suggests frontier model providers are in the Netscape position, while cloud infrastructure providers and application-layer companies capture durable long-term value as model costs approach zero.
• •Housing Supply Economics: Austin's permitting deregulation provides a measurable data point: each new unit added to housing supply directly reduces rents. New York City's simultaneous policy stack — rent freezes, eviction restrictions, bans on credit checks, and Airbnb prohibition — creates perverse landlord incentives to leave units vacant rather than renovate. Reports cite 50,000 ghost apartments in NYC, reducing available supply and pushing rents higher, the opposite of stated policy goals.