More Trillion Dollar IPOs, Anthropic $3T, Zuck's Price War, China Ends Open Source?, Trump Accounts

Summary

Brad Gerstner fills in and the hosts spend much of the episode on the coming wave of mega-cap AI IPOs, arguing that SpaceX’s IPO set a precedent for how to bring trillion-dollar businesses public. They debate whether Anthropic or OpenAI should go first, with concerns about OpenAI’s restructuring complexity and a consensus that these companies may still have substantial upside if revenue keeps compounding. The discussion then turns to the economics of AI usage: token prices have fallen sharply, but enterprise spending is still rising, especially for long-running agentic tasks where reliability matters more than a few dollars of inference cost. The hosts also dig into open source, model routing, harnesses, sovereign AI, and China’s possible move toward tighter control of its top models. The episode closes with a detailed policy segment on Trump Accounts, framed as a child-investment, tax-advantaged, and philanthropic vehicle meant to broaden equity ownership and capital accumulation across American families.

Notes

• Brad Gerstner fills in as guest host and the conversation quickly shifts from banter to AI, IPOs, and capital formation.
• Chamath says he joined a UN AI commission in Geneva with Mark Benioff, Jensen Huang, Brad Smith, Anthony Tan, and others.
• The commission is framed as focused on open source and sovereign AI.
• Jason tees up a “trillion dollar IPO rush” and says SpaceX is already trading around its IPO price.
• Anthropic’s confidential filing is positioned as the next major public-market event after SpaceX.
• Jason cites a 65% Polymarket probability that Anthropic goes public this year.
• Gavin Baker is cited as saying Anthropic could end 2026 above $100 billion of revenue and be very profitable.
• Chamath says Anthropic and OpenAI likely face different public-market paths because of burn, customer mix, and corporate structure.
• Chamath says token spend is compounding faster than productivity gains in his own organization.
• Brad says the SpaceX IPO was “textbook,” raising $75 billion at a $1.75 trillion valuation.
• SpaceX is described as trading near $2 trillion on about $35 billion of forward revenue.
• Brad says SpaceX pioneered mechanics around total raise, pricing, liquidity, index inclusion, and lockups.
• Brad flags index inclusion as risky if a post-IPO company has not been vetted and is likely to swing sharply.
• Brad says he would be surprised if OpenAI IPOs before Anthropic because OpenAI has more restructuring complexity.
• Brad argues trillion-dollar IPOs can still compound, but the easy upside pop is probably gone.
• Chamath says enterprise AI eventually has to prove ROI, while consumer AI is safer because it has many buyers.
• Jason says Uber’s CTO reports 99% engineer adoption of AI tools, 70%+ of pull requests attributed to agents, and 2,500 agentic skills built.
• Brad says the enterprise TAM is effectively every company on the planet, which is why revenue can scale so fast.
• Brad says if AI companies end the year above $100 billion, they could still 3-5x revenue next year.
• Jason argues the bottom-up product and $20 entry point explain the speed of AI adoption.
• Jason says he cut token costs by 95% using OpenRouter, GLM, Bittensor/Tao, and agent routing.
• Brad says token prices have fallen 90% over 2.5 years, yet frontier labs are still gaining share of wallet.
• Chamath says sovereign AI is inevitable because countries do not want to depend on closed American models.
• Jason says open source is hard to operationalize compared with just using Claude.
• Chamath warns that if AI adoption scales quickly without clear ROI, companies may cut other costs before they cut headcount.
• Brad rejects an either/or framing and says open source and frontier intelligence will coexist.
• Chamath says Meta has moved from open source to a price war with a one-hundredth-cost pitch.
• Jason says Meta is releasing a low-price agentic model through its API and Meta AI.
• Brad says premium agentic workloads can require millions of tokens, so the price delta versus a cheap model may be irrelevant.
• Brad says replacing a $200/hour consultant can justify the more expensive frontier model if it is reliable.
• David Sacks says enterprises want off frontier models, but most lack the technical ability to route effectively.
• Sacks says open-source share of enterprise spending is declining, from 19% to 11%.
• Sacks says open models work best for mature, well-defined use cases after post-training and data collection.
• Databricks is cited as finding the same model can cost about 2x less with a different harness.
• The harness layer is described as multiplexing both models and task-specific skills.
• Brad says hyperscalers and inference clouds will offer tools for model fungibility.
• Brad also raises the non-consensus possibility that intelligence is still not converging.
• Lovable and ElevenLabs are described as major frontier-model customers building their own models.
• Brad pushes back that the best voice agent may still need frontier models to win.
• Decagon is cited as sending about 90% of usage to open models after post-training and customization.
• Sacks says mature workflows can move to post-trained open models once the use case is well defined.
• David Sacks says frontier labs form an effective two-company revenue duopoly.
• Jason notes that open-source usage is invisible in revenue metrics even when it is used heavily in production.
• Jason says Chinese regulators may restrict overseas access to top Chinese models and treat AI research leaks as a national-security offense.
• Sacks says China’s tactic is to stay open until it catches up, then close to capture value.
• Brad says the U.S. government is united around staying ahead of China.
• Jason and Brad frame energy as the likely bottleneck for AI scale, not chips or software.
• Chamath says U.S. load growth could leave the country about three Californias short of energy by 2050.
• Jason says Taiwan’s LNG reserves are only a few weeks, making it vulnerable to blockade.
• Brad says the Trump Accounts launch was a four-year mission that went live on July 4.
• The account model is described as $1,000 at birth invested in the S&P 500, free for life.
• Brad says the accounts invest in the S&P 500 with no fees and lifetime free administration.
• The launch reportedly created over 1.5 million accounts and over $1 billion of deposits in 24 hours.
• Brad says contributors can use QR codes and Apple Pay, and the platform can target by zip code and age.
• Philanthropists including Michael and Susan Dell, Gwynne Shotwell, and Micron are cited as funding children’s accounts.
• Brad says the platform could become the largest direct philanthropic platform in U.S. history and raise $100 billion in a year.
• Brad frames the program as a way to make kids financially independent rather than dependent on government.
• David Sacks says the program is also a major estate-planning and tax-advantaged wealth-management tool.
• Sacks outlines the mechanics: up to $5,000 per year from family and friends, plus $2,500 employer contributions, with rollover into IRA or Roth IRA.
• Brad says the platform could scale from tens of millions of accounts to over 100 million within a decade.
• Brad estimates $2 trillion to $4 trillion could be added to accounts over 15 years for families that would otherwise have zero.
• Sacks says a maxed Trump account could make a kid a millionaire by age 28 under historical market returns.
• Brad says the quoted $50,000 and $200,000 values assume only $50 per month contributions.
• Brad says the program is intended to redirect employer contributions into kids’ accounts and lower taxable income.
• Jason argues the policy should be judged separately from Trump and treated as a broad capitalism-expanding program.
• Brad says the program should be as big as Social Security.
• Brad says children can withdraw only up to 25% for home purchase, business startup, or college, with the rest rolling into an IRA.
• Brad says 37 states require financial literacy and 25 states may add money into children’s accounts.
• Jason praises the government software and credits Joe Gebbia.
• Brad says the team aimed to build one of the best consumer products ever launched.
• Brad says the initiative is a bipartisan shift in the social contract, comparable to the creation of Social Security.
• Brad says AI company gains and other policy wins could be funneled directly into citizens’ accounts.
• Sacks says kids cannot open IRAs until they have earned income, so Trump Accounts capture early compounding years.
• Brad says nobody saves until about age 25, so the account captures the most powerful early compounding years.