This episode of Prof G Markets (June 17) is hosted by Ed. The first half connects with independent journalist Ed Zitron (of the 'Where's Your Ed' newsletter and the 'Better Offline' podcast) — he just obtained and exposed OpenAI's financials, data already independently verified by the Financial Times. The core numbers: last year's revenue was about $13.07 billion (+250% YoY), but full-year spending was about $34 billion, with a $21 billion operating loss and a $38.5 billion net loss, ending the year with about $22 billion in cash. There's a dispute over 'exactly how big the net loss is': counting interest income/expense pushes it to $60 billion; while a 'person familiar' the FT reached says that, due to the rising valuation, there was about a $30 billion 'investor rights' charge (akin to stock-based comp), and after adjustment 'the true net loss is only about $8 billion.' Zitron scoffs at this: 'You spent $34 billion to make just $13 billion — just look at actual costs and actual revenue, don't let this GAAP witchcraft twist you around.'

He names a few numbers most worth watching: ① sales and marketing expense of $5.73 billion, up a staggering 418% YoY, at a full 44% of revenue — for comparison, Facebook's marketing peak (2008) was 28% of revenue, Google's (2003) 11%; 'they're not spending money to build AI, they're spending money to sell AI'; ② R&D of $19.18 billion (he thinks OpenAI deliberately stuffs a lot of money into R&D, framing marketing as 'this is temporary, it can be stopped later,' when in fact every cost item is soaring); ③ administrative/personnel alone is $1.57 billion; ④ about $867 million (6.6% of revenue) came from a single partner, SoftBank's mysterious Crystal Intelligence project — which has barely materialized, yet was paid nearly a billion, hinting OpenAI's growth isn't as fast as outsiders assume.

On 'can it be sustained,' Zitron is very pessimistic: OpenAI essentially has just one product — ChatGPT and the API (plus a few side lines) — with no real path to 'invest more to earn more'; it already disclosed in the Musk lawsuit that compute alone will cost $50 billion in 2026, from which he extrapolates it could burn $80-90 billion this year; it just raised $122 billion at the start of the year, 'obviously not because it's near profitability.' He bluntly says: with a cost structure this severe, he doesn't think any AI lab can be profitable, and Anthropic's financials would look just as ugly; the market seems to think it won't be this bad, 'but I told you so — of course it's this bad.' The two also mock that Sam Altman lacks Musk's 'reality distortion field' and knack for storytelling to capital markets — Altman even admitted on stage that 'AI cost is a big problem right now,' considered a cardinal roadshow sin.

The second half connects with Morningstar equity analyst Nicholas Owens — who actually did a DCF valuation of SpaceX. Background: on its third day public SpaceX became the world's fifth-largest company, matching Amazon's market cap (about $2.6 trillion), yet with less revenue than Macy's; that day it announced a $60 billion (all-stock) acquisition of AI coding startup Cursor intraday, jumped 15% at one point, closed up 5%, and is up 48% cumulatively from Friday's offer price. Owens's fair value is only $780 billion (probability-weighted across three scenarios), with a moat rating of 'narrow,' less than half the market price: the mature rockets + Starlink business is worth about $611 billion, near a wide moat (its cost curve leads rivals by a decade); but repackaging the company as an 'AI company' (buying xAI, claiming a $26-28 trillion TAM) relies on 'space data centers' that are too speculative — by his math, the probability that Starship achieves 'hourly/daily high reusability' AND 'space data centers have a cost advantage over the ground' both hold is only about 7%. Even giving the moonshot scenario full probability, it's worth just $169/share, versus ~$200 now and an offer price of just $135. He thinks the current price is driven more by supply and demand: the public float is only 4%, an extremely small float → it has become a meme stock, compounded by structural buying from passive index funds; and after Q2 earnings (late July/early August), insider lockups expire, and hundreds of billions of dollars of stock — more than the IPO issuance — will hit the float, the real supply-demand test — these early shareholders who've held for over a decade with essentially zero cost basis 'don't care about 160 vs 200; the price is set by the marginal seller.' The Cursor acquisition is essentially neutral to valuation (fair value 63→62), but raises the value of the moonshot scenario.

In closing, Ed characterizes it alone: the stock market has entered 'crazy town.' The Shiller cyclically-adjusted PE (CAPE) has risen to 42x, the second-highest in history, behind only 1999's 44x — 'this isn't an opinion, it's a fact: we've basically entered dotcom-level valuation territory.' He stresses not every stock will crash (Microsoft, Meta look relatively cheap to him), but certain areas must pull back, with the 'epicenter' being space, especially SpaceX; FOMO is extremely strong right now, 'the AI mania has officially arrived.'

(For AI investors: ① this is the rarest-ever look inside a frontier large-model company's financials — high revenue growth vs. out-of-control costs, 44% of revenue on marketing, 'suspicious revenue' from a single partner — directly bearing on the sustainability of the closed-model camp and the pricing of the coming OpenAI IPO; ② SpaceX's 7%-probability framework is a good tool for de-mystifying the 'space data center / AI narrative premium'; ③ CAPE at 42x is a hard metric for judging where the overall AI bubble stands. Can be read against the views of Cathie Wood, Dalio, and Druckenmiller in this same library.)