OpenAI’s Leaked Financials, SpaceX’s Valuation, and the AI Bubble

Ed Elson walks through OpenAI’s leaked 2025 financials with Ed Zitron, the reporter who broke them, then turns to Nicholas Owens of Morningstar on SpaceX’s valuation. The through-line: AI and space valuations have reached dot-com-level euphoria, increasingly detached from the underlying numbers and resting instead on the story a chief executive can tell.

Key ideas

1. OpenAI spent roughly $34bn to earn $13bn — and the comforting figures are spin. Revenue hit $13.07bn (up more than 250% year on year), but the company lost about $21bn from operations, with a headline net loss near $38.5bn. A source ‘familiar with the matter’ told the Financial Times the real loss, stripped of a one-off $30bn investor-rights charge, was closer to $8bn. Zitron dismisses that framing: you cannot spend $34bn to make $13bn and call it an $8bn loss.
2. Sales and marketing is the alarming line, not R&D. Marketing spend jumped 418% in a year to $5.73bn — 44% of revenue. For comparison, Facebook’s marketing peaked at 28% of revenue in 2008 and Google’s at 11% in 2003. The money is going not to build AI but to sell it, a sign the product does not yet pull its own demand.
3. A single partner supplied nearly a billion in revenue. About $867m (roughly 6.6% of revenue) came from SoftBank, apparently tied to a barely-launched product called Crystal Intelligence. Such concentration from one partner that did not lift costs suggests organic growth is softer than the topline implies.
4. SpaceX is priced at a story the numbers do not support. It trades around $2.6tn — as valuable as Amazon on less revenue than Macy’s. Morningstar’s discounted-cash-flow analysis values it at $780bn with a narrow moat, putting only a ~7% probability on both Starship reusability and a genuine cost edge for data centres in space holding at once. A tiny ~4% public float invites meme-stock dynamics, and large insider lock-ups are due to expire after Q2 earnings.
5. The whole market sits at dot-com-level euphoria. The Shiller cyclically-adjusted P/E of the S&P 500 is about 42 — the second-highest reading on record, behind only 1999’s 44. Valuations across AI and space increasingly hinge on narrative rather than fundamentals, with SpaceX ‘ground zero’ for an overdue correction.

Related

• Scott Galloway — host
• Ed Elson — host
• Ed Zitron — guest; broke the OpenAI financials story
• Nicholas Owens — guest; Morningstar equity analyst
• Narrative Valuation — valuations resting on a chief executive’s story rather than fundamentals
• Token Economics — the AI unit-cost problem underlying the losses