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

Summary

Theses (episode spine)

• Sax: Banning Chinese open-source AI models would backfire — U.S. developers need access to public-domain contributions, and America would end up on “an island of overly expensive AI” if forced into a closed-source duopoly.
• Sax: Anthropic is engaged in regulatory capture — pushing a distillation-as-IP-theft narrative to protect its valuation rather than solving the problem itself (e.g. via KYC on its own platform, which it has declined to implement because it would slow growth).
• Chamath: The distillation panic is really a valuation-preservation game by closed frontier labs, which he says are mispriced 25-50x versus open alternatives as commoditization accelerates; he argues frontier-model economics could evaporate within a few years, with real value migrating to the application and infrastructure layers.
• Chamath: Third-party trackers show a stall in Anthropic’s revenue growth and a rising share (over 50%) of untracked, free “dark token” open-source usage; he believes open source could derail Anthropic’s and OpenAI’s IPOs via margin compression as enterprise customers (e.g. Lovable, 11 Labs) shift to open models like GLM 5.2.
• Sax (counter to Chamath): Anthropic and OpenAI show no real slowdown — Anthropic’s ARR went from about $10B to over $70B this year and should approach $100B; OpenAI’s tracked ARR rose from $33B in May to $41.3B in July with an internal forecast near $75B exit ARR — he calls Chamath’s slowdown thesis a “flop” drawn to bait government intervention.
• Sax: Anthropic’s $1.5B book-piracy settlement exposes hypocrisy — the company argues it can train on any public output under fair use but calls Chinese distillation of its own outputs “IP theft”; he thinks this framing could backfire, undermining Anthropic’s own fair-use defenses and possibly its IPO.
• Chamath and Freeberg: Alphabet (Google) is a standout AI-exposure buy — a roughly 32% 25-year average return on invested capital, an aggressive but disciplined capex ramp, and diversified exposure via GCP, YouTube and custom silicon, plus stakes in SpaceX and Anthropic (a reported $100B one-quarter markup), make it what Freeberg calls “the best public market stock to own” for AI.
• Freeberg: China’s long game is to commoditize the global knowledge/services economy via open-source AI while retaining dominance in the “molecule economy” (manufacturing and energy), where it has roughly 20x U.S. manufacturing capacity and is building toward 8x U.S. electricity production.
• Freeberg and Sax: NYC Mayor Mamdani’s “rental ripoff” tenant-protection package (banning certain application-fee practices, recognizing tenant unions, continuing a rent freeze) will backfire — reducing landlords’ incentive to maintain or rent out units (citing roughly 50,000 vacant “ghost apartments” in NYC) and ultimately hurting the working-class tenants it’s meant to protect.
• Chamath: Proposes running NYC’s rent-control policy as a natural experiment against deregulated cities like Austin; predicts landlords will respond by raising asking rents 3-4x and requiring large prepayments to price in the inability to evict or vet tenants, so rents will rise rather than fall.

$GOOGL (Alphabet / Google)

Convergence / divergence: Both hosts converge fully bullish on Google, with no dissent from Sax or Jason on the stock itself — the debate was about the market’s negative reaction to the earnings print, not about Google’s long-term positioning.

Speaker calls:

• Chamath (bullish, long-term): Google’s 25-year average return on invested capital is roughly 32%, so the market should give management the benefit of the doubt on the raised capex; he calls it a compounding machine that wins whichever AI models proliferate because it monetizes via cloud, silicon and ads.
• Freeberg (bullish, long-term): GCP’s access to enterprise data (email, drive) and model-agnostic infrastructure make it uniquely positioned; even in the worst case where Google’s own models and apps fail, it can still profit as the lowest-cost infrastructure host for other companies’ models, making it the best public stock to own for AI exposure.

Cross-check:

• Price: $319.74 (as of 2026-07-24 close). P/E 15.9 trailing / 24.1 forward. Market cap $3.99T. Next earnings: 2026-10-27.
• Recent headlines worth knowing: Q2 2026 earnings (reported 2026-07-22) beat on revenue but the stock sank on a raised capex forecast ($195-205B for the year) and the company’s first-ever quarter of negative free cash flow; Google Cloud reportedly on a $100B run-rate.
• ⚠️ Inconsistencies: none flagged — the transcript’s account of the ~7% post-earnings drop and capex raise matches contemporaneous reporting.

Topics discussed

Chinese Open-Source AI (Kimi K3) and the Distillation Ban Debate

Summary: The White House is weighing whether to ban Chinese open-source AI models after Moonshot AI’s Kimi K3 matched frontier-model performance at roughly half the cost, reviving “DeepSeek moment” panic; Polymarket odds of a 2026 U.S. ban on an open-source model jumped from 22% to 45% in days. Anthropic has alleged Kimi K3 was distilled from its models via what it calls “industrial-scale distillation attacks,” though hosts note Anthropic’s own blog post never used the phrase “IP theft.”

Speaker views:

• Sax: No White House decision has been made; he opposes any ban, arguing it would hurt American developers and that Anthropic should fix its own terms-of-service enforcement (e.g. KYC) instead of seeking a government-enforced ban on competitors.
• Chamath: Distillation could be stopped easily via KYC on Anthropic’s own platform; the “industrial-scale distillation attack” framing was coined by Anthropic to justify seeking government protection as part of a valuation-preservation game.
• Freeberg: Distillation — learning from a competitor’s output rather than copying its code or weights — is a common, legal cross-industry technique (he compares it to Google benchmarking against Yahoo/Microsoft search results in its early days); the real legal question is model weights versus outputs, not distillation itself.

Potential impact: Chamath argues a government ban on U.S. companies using open-source models would “tank the stock market” by artificially inflating enterprise AI costs, and would ultimately crater Anthropic’s and OpenAI’s valuations once markets recognized the revenue was propped up by regulation rather than market demand.

Anthropic’s $1.5B Copyright Settlement

Summary: Anthropic settled the largest AI copyright lawsuit in U.S. history for $1.5B after downloading roughly 7 million pirated books from sites like LibGen to train Claude; authors receive about $3,000 per book across 500,000 covered books (91% already claimed), and lawyers are receiving $101M. Hosts frame it as one data point in a broader wave of AI-training copyright litigation, including New York Times v. OpenAI and Thomson Reuters v. Ross.

Speaker views:

• Sax: The settlement happened because Anthropic pirated the books outright rather than buying single copies; the underlying fair-use question for AI training remains unresolved in court, and Anthropic’s position that it can freely train on others’ output while calling Chinese distillation of its own output “theft” is hypocritical and legally risky.
• Chamath: Content owners will keep pushing for licensing settlements; suggests AI companies pool a share of revenue to pay creators for ongoing access, though he doubts rights-holders will settle for as little as 10%.
• Jason: Proposes AI companies collectively contribute 10% of revenue to a pool that compensates content owners for continued access to new material.

Potential impact: Sax argues Anthropic’s public “IP theft” framing of Chinese distillation may be a legal misstep that undermines its own fair-use defenses and invites content owners to claim a larger share of its revenue, potentially threatening both Anthropic’s and OpenAI’s IPO prospects.

Google and Tesla Q2 Earnings: AI Capex Surge

Summary: Alphabet raised its capex forecast to $195-205B for the year (with Google Cloud now on a $100B run-rate) while posting negative free cash flow for the first time ever, sending shares down about 7%; Tesla’s capex surged 140% year-over-year toward a forecast $25B, with shares down about 14%. Hosts debate whether the spending signals confidence or overreach.

Speaker views:

• Chamath: Google’s ~32% 25-year average ROIC justifies aggressive AI infrastructure investment; contrasts it with Apple, which he says returned about $900B to shareholders via buybacks and dividends instead of investing more ambitiously in new products.
• Freeberg: Google’s enterprise data advantage and model-agnostic cloud make GCP uniquely positioned versus SaaS competitors tied to a single model.
• Jason: Notes a tweet comparing Google’s capex to roughly 20% of the U.S. military budget as a likely explanation for the market’s surprised reaction.

Potential impact: Hosts frame the heavy AI capex as long-term value creation rather than waste, arguing it positions Google to profit across model outcomes via its cloud, silicon and advertising businesses.

SpaceX Stock Under Post-IPO Pressure

Summary: SpaceX shares are down about 30% from their first-day closing price and trade around a $1.5T valuation, down from the roughly $2T IPO-day pop, with staggered lockup expirations adding further supply pressure.

Speaker views:

• Jason: Notes the decline and the staggered lockup schedule as a factor in continued pressure on the stock.
• Chamath: Highlights Alphabet’s 10% stake in SpaceX as part of Google’s broader portfolio value, alongside a reported $100B one-quarter markup on its Anthropic stake.

NYC Rent Control / “Rental Ripoff” Policy (Socialism Corner)

Summary: NYC Mayor Mamdani proposed a “rental ripoff report” that would bar landlords from charging credit-check application fees, prevent requiring both a credit check and a 40x-income standard, formally recognize tenant unions, and continue an existing rent freeze; hosts also cite an activist calling evictions “violence” and NYC’s roughly 50,000 vacant “ghost apartments” alongside its existing Airbnb ban.

Speaker views:

• Freeberg: Delivers an extended argument, invoking John Quincy Adams, that private property rights are foundational to American liberty, and that framing landlords or wealthy owners as inherently unjust is the first step toward a tyrannical erosion of property rights.
• Sax: Argues restricting evictions harms other tenants, not just landlords, by letting disruptive or delinquent tenants remain, degrading building conditions for law-abiding, often lower-income residents; also blames NYC’s renovation code requirements for creating “ghost apartments.”
• Chamath: Proposes running NYC’s policy as a natural experiment against deregulated markets like Austin; predicts landlords will respond by raising asking rents 3-4x and requiring large prepayments to price in the inability to evict or vet tenants, so rents rise rather than fall.

Potential impact: Sax and Chamath both predict the policies will raise effective rents and shrink usable housing stock (more vacant “ghost apartments”) rather than improve affordability.