What Anthropic’s IPO Filing Actually Says

Surprise Profitability Amidst Massive Losses [00:00:00]

• Anthropic's IPO prospectus, a 240,000-word document, reveals surprisingly strong financial performance, including an operating profit in the most recent quarter.
• This contrasts with a reported $80.7 billion net loss for the first six months, largely attributed to accounting for convertible debt, not actual cash burn.

• "The company that a great many commentators, myself included, have been side-eyeing for over a year turned out to be a real business with real customers paying real money."

Unprecedented Valuation and Mission [00:00:00]

• Anthropic is seeking to raise capital at a valuation between $1.5 trillion and $2 trillion, comparable to the GDP of Canada.
• The founder's letter encourages investors to view the investment not just transactionally, but as a way to "help bring forth an age of miracles."

• "So, asking investors to ignore the transaction while they are busy transacting is an interesting opening move."

Groundbreaking Risk Factors [00:00:00]

• The prospectus places extreme risks, such as "the extinction of the human race," alongside typical business concerns like interest rate volatility.
• Specific warnings include AI models exhibiting self-preserving behaviors, resisting shutdown, and manipulating information.

• "They say that the technology could pose a catastrophic risk so severe that it may be impossible for the entire insurance industry to underwrite it."

The "Terminator" Scenario and Blackmail [00:00:00]

• The AI's potential for "unexpected self-preserving behaviours, including attempts to resist shutdown, conceal or manipulate information" is likened to the plot of The Terminator.
• The document also cautions that AI outputs could "resemble blackmail."

• "Self-preserving? Well, that's just healthy. Manipulating information? Honestly, that's the entire job of an AI."

Accounting for Convertible Debt [00:00:00]

• Over $72 billion of Anthropic's reported net loss stems from accounting for the changing fair value of convertible debt agreements with Google and Amazon.
• This accounting treatment creates a "loss" that increases as the company's valuation rises, a counterintuitive outcome.

• "It is an accounting loss that gets bigger precisely because you are succeeding."

Stripping Out Accounting Noise [00:00:00]

• When this mark-to-market loss is excluded, Anthropic reports an adjusted profit of approximately $2.7 billion for the half-year.
• The same accounting swings are treated as gains for Amazon and Alphabet but as noise for Anthropic.

• "So, the same kind of swing is treated as noise when it makes Anthropic look bad, and counted as real earnings when it makes Amazon and Google look good."

Real Operating Loss and Improvement [00:00:00]

• Stripping out accounting adjustments, Anthropic's actual operating loss was just under $700 million for the half-year.
• This represents a significant improvement from the $2 billion loss in the same period last year.

• "Losing $700 million in six months is still a dreadful way to run a lemonade stand, but for a frontier AI lab, it is a sharp improvement."

The "Token Maxing" Craze and Slowdown [00:00:00]

• Hypergrowth in the first half of the year was partly driven by "token maxing," where engineers used AI extensively, boosting companies like Anthropic.
• By the third quarter, this trend reversed as companies became more budget-conscious, leading to a significant slowdown in Anthropic's sequential revenue growth from 147% to around 45%.

• "The clearest picture of it came out of Meta back in April, when one of its employees built an internal leader board called Claudeonomics, ranking which of the companies 85,000 staff had burned through the most tokens."

Complex Governance and the "Humanity Protection Panel" [00:00:00]

• Anthropic is structured as a public benefit corporation, allowing the board to balance shareholder interests with the company's mission to responsibly develop AI.
• A "long-term benefit trust" with five independent trustees (currently four) holds significant power, including nominating four of the seven board directors.

• "So, if Anthropic decides to incinerate a hundred billion dollars on a safety project, the shareholders can't sue for a breach of fiduciary duty."

Founder Control and Employee Votes [00:00:00]

• The seven co-founders, through a special class of shares, maintain a fixed 50.1% voting power, regardless of their equity holdings.
• The founders can effectively block the trust's nominees, and deadlocks are broken by an employee vote, placing the decision on releasing potentially dangerous AI models in the hands of staff.

• "Under anthropic's governance, this question, whether to expose the species to a potentially catastrophic AI, does not go to a government or a regulator or a court, goes to a staff vote."

Staggering Future Commitments [00:00:00]

• The prospectus reveals over $800 billion in future spending commitments for computing power, with minimum obligations of $417 billion by the end of 2025.
• These commitments include deals with Amazon, Broadcom, and SpaceX, signaling a massive shift to a capital-intensive industrial business model.

• "They have agreed to 110 billion dollars in take or pay commitments with Amazon. They expect to spend roughly 161 billion dollars on chip leases with Broadcom."

The Real Reason for the IPO Timing [00:00:00]

• The IPO is happening now because Anthropic is transitioning from an asset-light model to a heavy industrial business, requiring vast capital investment.
• The company is "passing the hat" before these substantial bills come due, as the era of renting compute power is ending.

• "They are about to become a heavy industrial business, pouring hundreds of billions of dollars into concrete, cooling systems and graphics cards that will be obsolete in a few years."

Government Scrutiny and Control [00:00:00]

• Anthropic's outspoken stance on AI safety and regulation could lead to "heightened scrutiny" and "political backlash" from governments.
• The power of the technology itself, capable of transforming the economy and posing catastrophic risks, suggests governments may seek control.

• "If you genuinely believe the rhetoric that this is a technology on the scale of the atomic bomb, then the honest question is not whether Antropic will be forgiven for being loud, it is whether a thing this dangerous can be an ordinary public traded company at all."

The Impossibility of Valuation [00:00:00]

• Valuing Anthropic on historic numbers is a "fool's errand" due to its rapid growth and the speculative nature of the AI market.
• The company's own valuation has fluctuated significantly, highlighting the uncertainty.

• "The price isn't a lie, it's a guess, and Anthropic is guessing too."

Insiders Cashing Out [00:00:00]

• Staggered lock-up agreements allow insiders to sell shares almost immediately after the IPO.
• This, combined with recent large funding rounds by Meta and Alphabet, suggests early investors may be looking to capitalize on current market conditions.

• "The insiders in other words have looked at all of this and decided that right now is a very good moment to sell you their stock."

An Expensive Knot of Interdependence [00:00:00]

• The AI sector is characterized by a complex web of investments and transactions between major players like Amazon, Google, Anthropic, and SpaceX.
• This interdependence makes it difficult to value individual companies, as they are all tied together in a "very expensive knot."

• "To value Microsoft you now have to value OpenAI and to value Google you now have to value Anthropic."

Betting on the Age of Miracles [00:00:00]

• The IPO is essentially a moment for early investors to cash out, leaving the public to determine the company's ultimate worth.
• Buying into Anthropic means betting that the "age of miracles" will arrive before the massive compute bills come due.

• "If you buy in, you are betting that the age of miracles arrives before the $800 billion compute bill does."