Roundtable, Interview, Fireside Chat
Meta's Muse Hits #1 | Menlo Sounds the AI Bubble Alarm | Keith Rabois vs Airwallex: Who is Right?
- Anthropic is rescheduling its $2 trillion IPO from October to November to present a cleaner financial narrative by including October results in the November filing, with bankers anticipating a significant valuation uplift in Q4 contingent on market stability.
- The company intends to self-insure product liabilities due to its valuation, while its S-1 filing will explicitly disclose that at least half of its workforce believes the technology could destroy the world to satisfy securities law requirements.
- OpenAI projects a net burn of $278 billion leading to a cash exhaustion point in 2028 unless burn rates decrease, requiring $700 billion in CapEx over three to four years to generate $350 billion in revenue, with significant backstop support from Oracle and NVIDIA.
- OpenAI expects revenue to grow 10x from $35 billion to $350 billion in three to four years, though actual burn is forecast to be 30% to 50% higher than current models, potentially necessitating $400 billion in required burn and a funding round valued over $1.5 trillion.
- Jensen Huang maintains a stance that there is a zero probability of AI destroying humanity, a position expected to hold until 2030 when open-source models catch up.
- Meta's Muse app is projected to dominate the App Store and utilize autonomous agents as a competitive shield against ChatGPT, though unchecked agentic commerce threatens to reduce basket sizes and erode advertising revenue by allowing users to bypass ad-upsell models.
- Amazon faces a projected loss of 5% to 10% in advertising revenue as agents route around its systems, whereas Shopify is expected to benefit from agentic commerce by serving as the payment rail for these transactions.
- Traditional systems of record such as Rezzy are anticipated to lose business as agents exploit broken APIs and open surfaces to bypass them, while the $100 billion LLM market is expected to see a 20% shift toward cheaper "system one" classifiers, reducing spend in that segment from $20 billion to under $4 billion.
- OpenAI and Anthropic plan to avoid competing in the low-margin "system one" classification market to focus on their AGI mission, continuing to offer simplified but lower-quality model variants for simple tasks.
- Venture capital entry points are set for $8 million to $10 million minimums for standard seed rounds, with Andreessen Horowitz targeting "pre-inception" deals with checks of $20 million to $40 million, while LPs are expected to abandon managers focusing on steady compounding with low IRRs in favor of high-risk, high-return strategies.
- Specific investment decisions include voting against proposals with near-zero or negative gross margins, passing on Crusoe at a $3.9 billion valuation due to margin target concerns, and requiring circular, guaranteed financing from Tier 1 partners like Microsoft for that deal to proceed.
- The data center investment sector is characterized by high leverage, making it highly vulnerable to any slowdown in AI usage, while Airwallex CEO Jack lacks a sufficient advocate bench to counter allegations of Chinese ownership and must recruit influencers to handle PR crises.