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

SpaceX raised $75 billion at a $1.75 trillion valuation in its IPO and now trades at roughly $2 trillion on approximately $35 billion of forward revenue, making it the seventh largest company in the world. Brad Gerstner noted the offering pioneered staged lockup releases tied to milestones and early index inclusion, but SpaceX shares fell from $200 to $150 post-IPO and the peak-to-trough drawdown in the six months following was approximately 50 percent.

Anthropic confidentially filed for IPO on June 1st, with Polymarket giving a 65 percent chance it goes public this year. Gavin Baker believes Anthropic could end 2026 with over $100 billion in revenue and trade at $3 trillion if it went public now. Gerstner said the chances of both Anthropic and OpenAI going public in the next six to nine months are very high barring a black swan event, and that Altimeter would be a buyer at scale in both offerings. OpenAI is rumored to be running at roughly $70 billion in annual revenue versus Anthropic at over $100 billion, though OpenAI carries more corporate restructuring complexity that makes it less likely to IPO first. Both are expected to debut above $1 trillion, though companies valued above that threshold at IPO are unlikely to produce 50 to 100 percent durable post-offering bounces.

Chamath Palihapitiya argued Anthropic may be accidentally profitable while OpenAI has high cash burn due to greater reliance on consumer revenue. He warned that his company's token costs are doubling every 45 days while downstream productivity gains are at most 5 percent, and that his CTO told him more tokens are needed for each next iteration of improvement because gains have asymptoted. He tested Anthropic's model on AI lift to S&P 500 EPS growth and found actual enterprise AI ROI based on publicly available data appears to be between 0 and 2 percent, with most apparent gains coming from pricing power and buybacks rather than AI productivity. His prediction is that companies able to IPO now should do so before diminishing AI productivity returns seep into the market. Gerstner pushed back, arguing that enterprise adoption is still so early that experimental spending lacking direct ROI today does not change the trajectory of frontier labs, and that $200 billion of incremental annual revenue would be incomprehensible in the history of the world.

Meta released Llama-based Muse Spark 1.1 as a strong agentic coding model at roughly one one-hundredth the price of frontier models, with Zuckerberg signaling a price war. Gerstner argued the cost gap between a cheap model at roughly $3 and a frontier model at roughly $15 becomes irrelevant for premium agentic workloads like replacing a $200-per-hour software engineer if the frontier model is more reliable. Despite skyrocketing overall AI usage, open source share of enterprise AI spending fell from 19 percent last year to 11 percent this year, though that metric undercounts open source utilization because users pay only compute costs. A Decagon founder argued open models work best for mature, well-defined use cases while frontier models are necessary during the discovery phase, and Databricks found that choosing the right harness alone can reduce costs by approximately two times. Measured by token revenue, only Anthropic and OpenAI make meaningful revenue, suggesting the AI lab landscape is consolidating toward a duopoly.

Reuters reported that CCP officials are considering restricting overseas access to China's top AI models, with Chinese regulators meeting with Alibaba, ByteDance, and Z.AI, and China making theft or leaks of AI research a national security offense. Sax observed that ByteDance's top model has always been closed source, Alibaba's Qwen is moving from open to closed, and Zepu's GLM 5.2 is going closed after catching up to American frontier models, describing the pattern as staying open until reaching the frontier then closing to capture value. Gerstner noted GLM 5.2 carries watermarks indicating Chinese labs were distilling American models and said the US government is expected to act against that distillation, though Sax warned that ham-fisted decisions at lower bureaucratic levels or from Congress could undermine the US position.

The Invest America Act passed as part of the reconciliation bill and Trump accounts launched on July 4th, generating over 1.5 million account creations and over $1 billion in deposits within the first 24 hours. The core structure is a $1,000 investment at birth placed into a no-fee S&P 500 account, with individuals able to contribute up to $5,000 per year per child, employers up to $2,500 tax-free, and philanthropists also able to contribute. Money compounds tax-free until age 18, at which point the balance can roll into a Roth IRA. Major early contributors include Michael and Susan Dell contributing over $6 billion covering 25 million lower and middle income children, Gwynne Shotwell contributing $350 million in SpaceX shares, Micron contributing $250 million, and Gerstner personally contributing $100 million covering all children under age five in Indiana. Gerstner said if an account had been maxed out over the past 30 years the child would be a millionaire by age 28, and Calacanis estimated the program could add $2 to $4 trillion to families who would otherwise have had zero over the next 15 years. Approximately 25 states are planning to add money directly into children's accounts using redirected existing program funds, and the plan is to auto-create 50 to 70 million accounts within 90 days using government data.