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

What It Covers

Chamath, Jason, Sacks, and guest Brad Gerstner analyze the trillion-dollar IPO pipeline for Anthropic and OpenAI following SpaceX's $1.75T debut, debate whether enterprise AI token spend is generating measurable ROI, examine China's potential open-source model restrictions, and detail the launch of Trump Accounts — a new tax-advantaged investment vehicle for children backed by over $6B in philanthropic commitments.

Key Questions Answered

• •AI IPO Sequencing: Anthropic filed confidentially on June 1 and is rumored to target a 2025 IPO at a potential $3T valuation, with annualized revenue reportedly tracking above $100B by year-end. OpenAI, estimated at $70B exit-run-rate, faces additional complexity from its nonprofit-to-for-profit corporate restructuring. Altimeter's Brad Gerstner states he would buy both at scale. SpaceX's $75B raise at $1.75T provides the structural blueprint — early index inclusion, staged lockup releases, and disciplined float sizing.
• •Token ROI Reckoning: Chamath's CTO reported that internal token costs are doubling every 45 days while downstream productivity gains are capped at roughly 5%. This signals an approaching enterprise reckoning where CFOs will demand measurable EPS lift from AI spend. Uber's CTO publicly disclosed that 99% of engineers use AI tools and 70% of pull requests involve agents, yet the S&P 493's AI-attributable EPS growth appears to be 0–2% when stripping out NVIDIA and pricing-power effects.
• •Intelligent Token Routing: Leading technical teams at DoorDash and Coinbase have built middleware routing layers that send frontier-level tasks to Anthropic or OpenAI while delegating lower-complexity work to open-weight models like Kimi 2.6. Databricks found that optimizing the agent harness alone — without switching models — cuts costs by approximately 2x. The barrier is technical: most enterprises lack the engineering capacity to implement model fungibility, which is why closed frontier model revenue share continues to grow despite cheaper alternatives.
• •Open Source vs. Frontier Revenue Split: Despite 18 months of predictions that open-source models would erode frontier lab revenues, open source's share of enterprise AI spending dropped from 19% to 11% year-over-year. Frontier labs' share of wallet increased. The structural reason: open models require post-training and precise use-case definition to deploy effectively, making them suitable for mature, well-defined workflows but impractical for the exploratory, general-intelligence tasks that dominate current enterprise adoption.
• •China's Open-Source Reversal: Chinese regulators reportedly met with Alibaba, ByteDance, and Zhipu AI to discuss restricting overseas access to top Chinese models and classifying AI research leaks as national security offenses. The pattern mirrors OpenAI's own trajectory — stay open to gain developer adoption and close once you approach the frontier to capture economic value. GLM 5.2 reportedly contains watermarks indicating distillation from US frontier models, and the US government is expected to take countermeasures against distillation practices.
• •Trump Accounts Mechanics: The Invest America Act created tax-advantaged investment accounts for all US children under 18, auto-investing contributions into the S&P 500 at zero cost. Annual contribution limits are $5,000 from any combination of family, friends, or employers — with employer contributions of up to $2,500 excluded from taxable income for both parties. At age 18, balances roll into an IRA or Roth IRA. A child whose account is maxed annually from birth reaches millionaire status by approximately age 28 at historical market return rates.
• •Philanthropic Scale and Distribution Mechanics: Over $6B in philanthropic commitments launched alongside the app, including $250 per child across 25 million lower-to-middle-income kids from Michael and Susan Dell, and 350,000 SpaceX shares from Gwen Shotwell directed at low-income communities. Contributions are distributed by zip code and age. A pooled account structure allows large donors to distribute funds across the 3.7 million children born annually when per-child limits are reached. The stated goal is auto-creating accounts for all 50–70 million eligible US children within 90 days.