▶ YouTube Summaries

• YT :: https://www.youtube.com/watch?v=S79u6BtYvQ0
• Original title :: Anthropic’s business is insane

Prime reacts to Anthropic's S-1 IPO filing, then plays a largely uncut interview with TheStockGuy recorded before the filing was amended. Q3 numbers, missing at first, show growth slowing from ~100% to ~50% quarter over quarter. The amended filing blames the end of "token maxing". The key risk is ~$900B of take-or-pay compute commitments, which only work out if Anthropic keeps shipping the best models. Prime also criticizes the $1.5B book-piracy settlement and the opaque founder-controlled governance. TheStockGuy covers circular hyperscaler financing, the risk of cheap Chinese models once intelligence plateaus, and index rule changes that force retail exposure to mega-IPOs.

Prime's take on the S-1

• The original S-1 skipped Q3 despite being filed 26 days after the quarter ended, which fuelled "AI is ending" speculation. The amended filing shows Q3: revenue growth fell from ~100% (Q1 → Q2) to ~50% (Q2 → Q3). Estimates put it at ~$16B vs ~$11B.
• Anthropic attributes the slowdown to the end of the token-maxing fad. Prime finds it more telling that revenue still grew after companies (including some Fortune 100s that dropped Claude over cost) cut AI spend.
• ~$900B in committed compute spend spread over ~19 years, on take-or-pay terms: unused compute is still paid for in full. It is manageable only as long as demand holds.
• Real risks: losing the model lead (no one is loyal to a model), and an "intelligence plateau" where good-enough cheaper models — especially Chinese open-weight ones — win on price.
• A $1.5B line item for the book-piracy settlement (~$3,000 per book, ~500k books). Prime calls this hypocrisy for a company that presents itself as a moral leader.
• Governance: the Long-Term Benefit Trust and a founder vehicle holding one share with 50.1% voting control, so outsiders cannot see which of the seven people voted for what.
• Prime's view: a bubble in the hype, but real revenue and broad adoption make an outright collapse unlikely. Not financial advice.

Interview with TheStockGuy

• Missing Q3: likely withheld to avoid showing a slowdown before the IPO; markets punish even strong but slower growth (as with Google's ad revenue that day).
• Frenemies and circular money: ~48% of enterprise revenue flows through Microsoft, Google and Amazon, who are both competitors and infrastructure providers. Hyperscalers invest in Anthropic, which spends the money back on their cloud under guaranteed contracts, and their stake grows in value. The same pattern holds for Oracle/OpenAI, CoreWeave and others — "everyone passing the same money around".
• Profitability: the S-1 states in bold that Anthropic "may not generate sufficient revenue to achieve or sustain profitability". Much of the reported net loss is on paper (e.g. convertible notes), unlike OpenAI's cash burn. Revenue relies on staying the best model, and government and defense contracts are the stickiest revenue.
• Long-term threat: once public data is exhausted, intelligence plateaus and cheaper models at ~90% less cost take the mass market. Google is positioned well because it holds the most private data.
• Governance: a Delaware incorporation is litigation-friendly. Founder control through a single LLC share works as a buffer that diffuses blame and credit. Comparable structures: Elon's ~60% voting control at SpaceX, and Meta's split into a non-voting share class so Zuckerberg can sell to cover California's proposed wealth tax without losing control.
• Index rule changes: after SpaceX, Nasdaq allows immediate index inclusion with no price-discovery period, a ~5% float weighted 2–3x, and no lockup. This forces 401(k)s and passive funds into these IPOs (SpaceX, Anthropic, then OpenAI). The S&P and Dow declined to follow.
• Market backdrop: the US market is propped up by steady inflows (401(k)s, pensions) and by the top 10% of earners. Asian markets have sold off heavily, and $1.2T in margin debt recalls the 1920s and South Korea's margin-call spiral — sector rotations are likely rather than a crash.