Big Tech’s Insane Hidden AI Spending, Ranking Anthropic vs. OpenAI, AI For Travel Debate

Episode Summary

Executive Summary: The episode argues that Big Tech’s AI buildout is far larger and riskier than headline capex suggests because much of it sits in off-balance-sheet structures. It then compares Anthropic’s rapid revenue growth with OpenAI’s slower progress and executive turnover, and closes by making the case that travel is a surprisingly strong real-world eval for AI because it combines complex, consequential, but reversible decision-making.

Main Topics: Off-balance-sheet AI financing in Big Tech (Priority: 5/5): The hosts unpack how Meta, Google, Oracle and others are using joint ventures, leases, and third-party capital to fund AI infrastructure without showing the full exposure on their balance sheets. Systemic risk and leverage in the AI boom (Priority: 5/5): They argue these structures resemble a levered, concentrated bet on AI demand, with potential downside that may be hidden from investors until much later. Anthropic’s surging revenue and IPO readiness (Priority: 4/5): Bloomberg-reported numbers show Anthropic’s revenue accelerating dramatically, strengthening its case for a public listing and suggesting it may be pulling ahead of OpenAI. OpenAI’s slower growth and organizational churn (Priority: 4/5): OpenAI’s revenue growth is still strong, but it trails Anthropic and is accompanied by high-profile executive departures, raising questions about execution and product-market focus. What AI ‘outcomes’ really matter (Priority: 3/5): The hosts debate whether AI should be framed around broad future outcomes like prosperity, misuse, or failure, versus the more immediate business mechanics of financing and execution. Travel as an AI eval use case (Priority: 4/5): They conclude that travel planning is an ideal stress test for AI because it requires summarization, coordination, and iterative decision-making in a context with real but manageable stakes.

Key Arguments: Big Tech’s apparent AI capex understates total exposure because large obligations are shifted into leases, JVs, and financing vehicles that do not fully appear on balance sheets. Meta’s Hyperion project is a prime example: Meta is the builder, tenant, and guarantor, but the debt and campus sit with Blue Owl-linked entities rather than on Meta’s books. This financing is tactically rational for companies because it preserves cash and avoids visible dilution of their financial statements, even if it obscures risk from investors. The structure may be legal, but it makes it harder for public-market investors to understand total leverage and could create future systemic problems if AI demand disappoints. Anthropic’s numbers suggest much faster momentum than OpenAI’s, implying the company may be better positioned ahead of its expected IPO. OpenAI is facing internal turbulence and relatively slower revenue acceleration, which the hosts connect to competition pressure from Anthropic and product focus issues. Travel is a strong AI benchmark because it combines abundant information, shifting constraints, and practical consequences, making it a realistic but not catastrophic environment for testing agentic capabilities.

Data Points: Off-balance-sheet commitments across top tech companies: About $3 trillion - Wall Street Journal estimate of AI-related obligations not reflected on balance sheets Traditional capex by top tech companies: About $600 billion over the past year - Compared with much larger off-balance-sheet commitments Meta cash and liquid investments: $91 billion - Used to illustrate that Meta has large cash reserves even while using off-balance-sheet structures Google cash and liquid investments: $187 billion - Cited as another example of a cash-rich company Meta Hyperion initial lease commitment: About $12.3 billion - Initial commitment disclosed for the Louisiana data center project Hyperion financing debt: $27 billion - Debt financing the project, not shown on Meta’s balance sheet Meta total lease obligations not yet kicked in: $347 billion - Reported total obligations for leases that have not begun yet as of June Meta market capitalization: $1.39 trillion - Used to compare against the scale of off-balance-sheet obligations Anthropic annualized revenue run rate: $65 billion - Bloomberg-reported run rate by end of July before IPO Anthropic quarterly revenue: $11.5 billion - Reported quarterly revenue figure used to show rapid growth Anthropic comparable prior-period pace: $787 million - Revenue pace in the corresponding period in 2025 Anthropic growth multiple: More than sevenfold - Run rate growth from end of last year to July OpenAI quarterly revenue growth: 18% QoQ - Revenue grew from the first quarter to the second quarter OpenAI revenue in quarter: About $7 billion - Mentioned as still strong but lagging Anthropic Anthropic timing to IPO: As soon as this fall - Expected public listing timing discussed on the show OpenAI IPO timing: Not until 2027 - Reportedly said by Sarah Friar in an all-hands

Pivotal Quotes: "As these off-balance sheet commitments become frequent, larger, and more complex, it's becoming increasingly difficult for investors to assess companies' total potential leverage." — Morgan Stanley accounting analysis (quoted in story): The quote summarizes the core investor-risk concern about hidden AI obligations "This is creative financial structuring." — Alex Kantrowitz: Used to describe how Big Tech is funding data centers through JVs, leases, and guarantees "I think the travel is actually an excellent proving ground." — Alex Kantrowitz: Argument that travel is a strong real-world AI eval because it mixes information overload and practical stakes

Implications: Investors should treat AI infrastructure as far riskier than reported capex suggests. Anthropic appears to be scaling faster than OpenAI, while travel may become a useful benchmark for judging whether AI is genuinely useful in everyday high-complexity tasks.