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

Summary

Alex Kantrowitz and Ranjan Roy focus first on a Wall Street Journal-style accounting puzzle: roughly $3 trillion in AI-related commitments across nine major tech companies is sitting off balance sheet, largely through leases, joint ventures, guarantees, and financing vehicles. They use Meta’s Hyperion data center in Louisiana as the case study, highlighting $27 billion of debt financing, about $12.3 billion of initial lease commitment, and $347 billion of future lease obligations that have not yet kicked in. Roy argues the structure is rational because it shifts risk to external pools of capital, but both hosts say it makes leverage and true AI exposure much harder for investors to assess. The episode then turns to Anthropic and OpenAI, with Anthropic’s revenue growth looking explosive while OpenAI deals with executive departures, strategic shifts, and tougher comparisons. They close by making the case that travel is one of the best practical evals for AI because it combines planning, logistics, and changing constraints in a way that is useful but not catastrophic if it fails.

Notes

• Big Tech’s AI buildout is framed as a hidden leverage story, not just a capex story.
• Nine top tech companies are said to have about $3 trillion of off-balance-sheet AI commitments.
• The reported off-balance-sheet obligations are growing faster than the roughly $600 billion of traditional capex over the past year.
• The mechanism is external financing vehicles, not direct corporate spending.
• Blue Owl is cited as a famous example of the outside investor structure.
• Meta’s Hyperion campus is the showpiece example of the structure.
• Hyperion is described as not appearing on Meta’s balance sheet despite the construction debt.
• Meta initially agreed to a four-year lease beginning in 2029 with renewal options up to 20 years.
• Meta also guaranteed bondholders whole if it exits early, but has not booked a liability because it says payment is not probable.
• Ranjan Roy argues the money is coming from pension funds, private equity, and other institutional pools.
• The financing is tactically attractive because it preserves corporate cash.
• Meta’s aggregate initial lease commitment for Hyperion is about $12.3 billion.
• Meta disclosed about $347 billion of future lease obligations that have not started yet.
• The hosts compare that $347 billion figure with Meta’s roughly $1.39 trillion market cap.
• Meta has about $91 billion in cash and liquid investments, which narrows near-term liquidity concern.
• Google is said to have about $187 billion in cash and liquid investments.
• Roy says Wall Street should not be assumed to be fully on top of the issue.
• The hosts say off-balance-sheet complexity makes leverage harder for investors to model.
• The legal structure may be permissible, but it creates disclosure opacity.
• Roy says the largest companies are all making the same levered AI bet.
• The systemic-risk concern is correlated exposure across the biggest players.
• Alex Kantrowitz describes the spending as a call option on AGI.
• The upside case is powerful AI that can do jobs and cure cancer.
• Roy insists the key question is what the unwind looks like if the financing story breaks.
• The $347 billion figure is a long-dated lease commitment, not a simple immediate loss number.
• Roy separates the fate of AI technology from the fate of data-center financing.
• He says compute demand could fall because of innovation even if AI itself succeeds.
• Anthropic is said to be on track to annualized revenue above $65 billion by the end of July.
• That is described as more than 7x the pace at the end of last year.
• Anthropic’s Q2 revenue is said to be $11.5 billion versus $787 million in the comparable period.
• Roy wants month-by-month revenue and a graph, not just a run-rate headline.
• Kantrowitz suggests the Anthropic leak may be strategically timed against OpenAI’s IPO path.
• OpenAI reportedly told investors revenue grew 18% quarter over quarter.
• OpenAI’s losses reportedly widened at the same time.
• OpenAI is said to have made about $7 billion in the quarter.
• OpenAI has had a string of executive exits, including Denise Dresser, Brad Lightcap, and Fidji Simo.
• Roy says the enterprise strategy matters because those departing leaders were supposed to run it.
• Alex says OpenAI is compressing startup phases into a very short period.
• Roy pushes back that these companies are already valued like giant public companies, not startups.
• Alex defines the startup label by product immaturity and lack of a settled offering.
• OpenAI’s enterprise endgame is still unresolved and could span Codex, knowledge work, cloud, wearables, or home devices.
• Anthropic is seen as pulling ahead of OpenAI in the current cycle.
• Travel is presented as a strong AI evaluation domain because it forces information-heavy planning.
• The hosts describe travel as a Goldilocks-style eval: meaningful, but not catastrophic if it fails.
• Roy endorses more practical consumer evals like TripAdvisor and Google Flights style tests.
• Roy used AI to plan a trip to Denia, Spain from detailed family constraints.
• AI also helped Roy manage rental-car and passport logistics.
• Kantrowitz says he used ChatGPT for a multi-island Indonesia itinerary plus Dubai.
• He set up a daily 6 a.m. briefing that searched chat, Gmail, and calendar.
• Roy notes the free version of ChatGPT was dramatically worse than the paid version for travel planning.
• AI is making people more ambitious about complicated travel because it reduces fear of failure.