Roundtable, Interview, Fireside Chat
Meta's Muse Hits #1 | Menlo Sounds the AI Bubble Alarm | Keith Rabois vs Airwallex: Who is Right?
- Anthropic IPO Shift: Anthropic postponed its planned October IPO to November to incorporate Q3 financial data, aiming to present a "cleaner story" with audited numbers rather than the noisy, unverified data available in October.
- The delay suggests management confidence in securing a valuation premium (potentially 20x–30x oversubscription) rather than a "hell or high water" rush to raise capital immediately.
- The $2 trillion market cap valuation implies the company can self-insure against product liability risks, negating the need for external reinsurance despite potential existential risks.
- Risk disclosures in the S-1 will explicitly address concerns about AI safety, with management (specifically Dario Amodei) acknowledging internal risks, thereby satisfying securities laws regarding risk transparency.
- OpenAI Financial Outlook: OpenAI projects a net cash burn of $278 billion by 2030 and an infrastructure capital expenditure (CapEx) requirement of approximately $700 billion to reach $350 billion in annual revenue.
- Current cash on hand stands at $122 billion, necessitating massive external capital raises (rumored at $1.5 trillion valuation) to cover the gap.
- A significant portion of the required CapEx is off-balance sheet, facilitated by partnerships with infrastructure providers like NVIDIA and Oracle who lease compute capacity.
- Jensen Huang (NVIDIA) stated a "zero probability" that AI will destroy humanity by 2030, a stance the speakers attribute to the technical limitations of current models.
- Meta "Muse" Launch: Meta released "Muse," an AI assistant that has secured the number one spot on the App Store, displacing competitors like Instinct.
- Muse functions as a "Trojan horse" strategy, utilizing Meta's massive distribution network to offer autonomous agents and a superior consumer-facing LLM.
- The launch generated approximately $100 billion in market capitalization for Meta, resulting in a 7-8% stock spike in a single week.
- The product allows users to build complex custom workflows, such as creating a personal CRM to track 150+ sponsors in real-time, demonstrating "composable software" capabilities.
- Retail & E-Commerce Friction: Amazon blocked access to Muse's agentic commerce features, while Shopify partnered with them.
- Amazon's opposition is driven by the risk of lost ad revenue (which exceeds its e-commerce profit) and reduced basket sizes caused by agents bypassing shopping cart friction.
- Shopify supports the integration as it drives additional traffic to small merchants without cannibalizing a significant ad business.
- Speakers predict that agentic commerce will "maim" legacy aggregation systems like Resy (OpenTable) and potentially Amazon by routing users around traditional intermediaries.
- Vercel "Jeff" Launch: Vercel introduced "Jeff," a specialized model designed for rapid classification and routing tasks rather than generative text.
- Jeff is described as a "System 1" thinker (fast, low-cost) compared to traditional LLMs (System 2, slow, expensive), capable of answering simple queries in milliseconds for roughly 1/100th the cost.
- The tool targets the "unbundling" of developer stacks, handling ~20% of LLM calls that currently waste expensive tokens on simple routing or classification.
- This shift risks reducing revenue for major foundation model providers (OpenAI, Anthropic) by diverting low-complexity traffic to cheaper, specialized alternatives.
- OpenAI and Anthropic are expected to avoid competing in this low-margin space immediately, offering only "crippled" low-end models (OpenAI Mini, Claude Haiku) while focusing their resources on AGI.
- Venture Capital Trends: Traditional seed round check sizes are increasing, with minimums rising to $8–10 million and some deals requiring $20–25 million.
- Andreessen Horowitz launched a $40 million "University" to support pre-inception investing, mimicking the Paul Teal Fellowship but at scale.
- Nominal GDP growth (2.5x from 2010–2026) combined with venture exit inflation explains the necessity of larger check sizes to maintain equivalent ownership and IRR targets.
- There is a strategic divergence between managers chasing high-IRR "triple-double" deals (e.g., Sarah Guo, Sequoia) and those focusing on slower, capital-efficient compounders, with the former currently favored by LPs.
- Investment Committee Decisions:
- Factory (Approve): The firm approved investing in Factory, an enterprise coding agent provider at a $5 billion valuation, driven by the thesis that "coding is the motherlode" and the growing demand for data sovereignty and on-prem/private cloud solutions.
- Legora/Harvey (Pass/Defer): Leadership passed on leading Legora's next round due to reports of negative 50% gross margins and a lack of confidence in a quick turnaround, despite the strong category potential.
- Crusoe (Pass): The firm declined to participate in Crusoe's $3.9 billion Series B at a $30.9 billion valuation, citing that the data center infrastructure business is too levered to CapEx market volatility and the valuation leaves insufficient margin of safety.
- Geopolitical & Reputational Risks:
- The debate surrounding Airwallex and founder Jack Dorsey involves allegations of Chinese ownership and IP risks, illustrating the increasing sensitivity of cross-border M&A to geopolitical concerns.
- Speakers noted that large transactions (e.g., Fin) have previously required the removal of Chinese IP to close, validating concerns even if Airwallex's specific cap table claims are disputed.
- The public controversy highlights a strategic deficiency for founders: the need to build an "army of advocates" rather than relying on a single CEO to defend against aggressive social media attacks.