Anthropic's $30T Assumption & OpenAI Confirms IPO | Why Customer Service & Robotics are Overinflated

Poolside AI Acquisition Dynamics

• NVIDIA paid $6 billion to license Poolside's "model factory" and invested an additional $1 billion, valuing the company at a $12 billion pre-money price.
• The deal involved transferring 109 engineers to NVIDIA's Nemotron team to build open-source models.
• Poolside's founders could not raise $2 billion required to purchase 40,000 GPUs or fund their own massive data center, forcing the sale despite their belief in the US open-source market.
• Investors exited with a 15x return on the acquisition, a figure the hosts deem insufficient for true seed funds aiming for 50x–100x returns in the current era.
• Poolside's pre-money valuation effectively sat near $600 million post-dilution, implying that a 15x return from that stage is mathematically low compared to the potential trillion-dollar outcomes of frontier models.
• The acquisition highlights a market reality where capital intensity at the frontier is so high that only hyperscalers (Microsoft, Google, Amazon, NVIDIA) and Apple can finance state-of-the-art models, pushing VCs to exit earlier or accept lower multiples on "failure" bets.

NVIDIA's Ecosystem Investment Strategy

• NVIDIA is allocating its massive free cash flow (projected near $70 billion) into its ecosystem to secure demand for its chips, including funding Neocloud, Poolside, and Perplexity.
• NVIDIA is reportedly joining a new funding round for McCor (Mercury), a data labeling/training data company, which is valued at $20 billion with a goal of crossing $2.2–2.5 billion ARR.
• Unlike direct chip financing, McCor represents a strategic investment to secure high-quality training data, though the hosts note its gross margins are currently lower than pure-play AI software.
• The investment logic assumes that if 2026–2027 model training budgets reach $200–300 billion, a 5%–10% share of that spend would justify a $20 billion revenue run rate for data providers.
• Kroll data indicates that in the current AI M&A environment, gross margins above 30% do not command premium multiples; high growth and strategic fit are the primary value drivers.
• NVIDIA's strategy is described as "vendor financing," where the chip maker provides capital to ensure customers have the compute and data necessary to build models that purchase more chips.

OpenAI Public Market Trajectory

• OpenAI CFO Sarah Fry announced to employees that the company will go public in 2027, a timeline driven by the need to secure capital and compete with Anthropic's projected IPO.
• The announcement came after Q1 and Q2 showed a 5%–6% revenue growth (approx. 18% QoQ), which would annualize to roughly $30 billion if sustained, leaving OpenAI significantly behind Anthropic's estimated $60 billion run rate.
• The public statement is viewed as an existential defense mechanism to reassure chip vendors (like NVIDIA and Broadcom) of future demand, preventing them from scaling back chip orders.
• Hosts predict OpenAI's IPO valuation will be lower than Anthropic's due to the shift in market ranking and the increasing competition from open-weight models.
• The "absence of choice" narrative is cited as a primary driver; OpenAI must go public to prove profitability and scale, whereas competitors have already established clearer paths to revenue dominance.

Market Volatility and Investor Lessons

• Citadel's Ken Griffin unwound 80% of his short position in Leo Ashkenazi's book (likely a reference to a specific market failure or bet), highlighting the risks of leverage in volatile markets.
• The Korean Stock Exchange (KOSPI) rose 56.46% year-over-year, yet individual investors suffered massive losses due to extreme volatility and the need for precise timing.
• A key lesson from Griffin's success is that while long-term directional bets can be profitable, leveraging capital (4x leverage in Ashkenazi's case) requires being correct at every step, not just in the long run.
• The hosts contrast this with NVIDIA's business model, where they can be right "eventually" without leverage, whereas vendor financing requires customer growth to materialize immediately to repay the capital.

The "Token Addiction" and Enterprise Budgeting

• Enterprises are described as "token addicted," with usage patterns becoming entrenched across coding, legal, and finance sectors.
• Stripe's internal letter likened intelligence to capital: it is fungible, requires allocation, and demands spending controls similar to financial budgets.
• CFOs face a dual pressure: preventing token cost runaway (as seen in the "performative AI" phase of 2025) while retaining top talent who demand AI agents to remain productive.
• Hosts predict 2027 will be the year of the "backlash" where companies must decide between cutting headcount to fund AI token usage or accepting reduced EPS to maintain productivity.
• The "what if it all goes right" mental model is emphasized: even if a business model (like open weights or robotics) has theoretical flaws, the rapid adoption and revenue potential in the current cycle make it a rational bet.

Skepticism on Future Investment Categories

• Customer Support/CX: Predicted to become a commodity and largely obsolete as agents merge with sales and marketing functions; no exits expected for pure-play CX software.
• Humanoid Robotics: Hosts express skepticism regarding general-purpose humanoids, arguing the vision exceeds current dexterity and cost realities, favoring focused robotics (e.g., Locus Robotics) instead.
• Professional Services AI: Doubts exist regarding the ability to transform accounting or law firms into high-growth AI unicorns, citing the difficulty of standardizing high-discretion, relationship-driven work.
• Personal Productivity Agents: While inevitable, hosts question the massive market size, noting that most consumers are not as efficiency-obsessed as Silicon Valley elites, limiting the addressable market for hyper-personal agents.

Public Market Re-ranking

• Private companies like Stripe, Databricks, and Palantir are outperforming most public software peers, with Stripe accelerating to 41% growth and 71% billing growth.
• The hosts suggest that if these private assets went public, they would fundamentally re-rank the S&P 500, pushing traditional software companies into a "distant memory" category.
• Hugging Face is rumored to be a potential acquisition target, with a valuation of ~$13 billion driven by its strategic value as an open-weight model repository for enterprises seeking counterweights to closed frontier models.
• OpenRouter (acquired by Stripe) exemplifies the peak valuation moment for infrastructure enabling open-weight models, with the hosts noting that "selling" at this transition point is optimal.

Geographic and Demographic Wealth Concentration

• The AI boom is concentrating wealth in Silicon Valley, with property prices in San Francisco (e.g., The Avalon) reaching $10,000/month, forcing a permanent ratcheting up of the cost of living.
• The hosts note that unlike the 1999 or 2000 cycles, this cycle is more concentrated, requiring fewer people to generate more revenue, which will likely exacerbate inequality and displacement in other regions.
• The "average American" is not trying to be efficient; the market for personal productivity tools may be overestimated by Silicon Valley founders.