Anthropic's IPO filing reveals massive compute costs and founder control

The price of power

Anthropic's S-1 prospectus reveals the staggering amount of capital required to sustain its AI development. The company has committed to spending as much as $120.5 billion by 2029 to secure the necessary compute power for its models. This massive expenditure highlights the physical infrastructure required to run modern large language models.

A significant portion of this spending is directed toward hardware and cloud access. The filing notes a monthly payment of about $3.1 billion to SpaceX to gain entry to Nvidia graphics processing units once the planned capacity is ready. These numbers illustrate the intense competition for the specialized chips that power the AI industry.

Beyond hardware, Anthropic has secured long-term cloud agreements with the tech giants. The company plans to spend $44.9 billion with Microsoft until 2033, at least $114.5 billion with Google until 2033, and $110 billion with Amazon over the next ten years. These deals ensure the company has the scale needed to compete with established tech leaders.

Founder-led control

Despite going public, Anthropic is designed to keep its original leadership in the driver's seat. The company uses a public benefit corporation structure, which allows it to pursue a social mission alongside profit. However, a specific corporate entity called Founder LLC adds an extra layer of control for the company's seven co-founders.

This arrangement allows the founders to collectively hold more than half of the voting power on important issues. The company argues that the founders are uniquely suited to act as stewards of its mission to benefit humanity. This structure ensures that the original team can make major decisions without being overruled by outside investors.

This level of control is unusual for a company seeking public investment. It suggests that the founders prioritize their specific vision for responsible AI development over traditional shareholder influence. This setup could impact how the company responds to pressure from the public markets in the coming years.

Executive perks and pay

The filing provides a clear look at the compensation for the company's top leadership. CEO Dario Amodei and President Daniela Amodei are the highest-paid executives at the firm. Last year, Dario's compensation reached nearly $18 million, while Daniela received $16.4 million, primarily in the form of stock and option awards.

The company also discloses specific expenditures for the personal security and travel of these two executives. Anthropic spent $2.75 million on security for Dario and $1.29 million on personal safety, home protection, and travel for Daniela. These costs are framed as necessary for the continued service of the senior team.

Additionally, the filing notes that Dario has access to a private jet for personal use. This specific perk cost the company an additional $97,055. Anthropic justifies these expenses by stating that the success of the company is highly dependent on the continued service of these specific individuals.

Unconventional risk factors

Anthropic's prospectus includes a list of risks that go beyond typical market competition or regulation. One of the most striking disclosures is the warning that AI systems could cause potentially 'catastrophic' harm. The company notes that such harm might be so severe that it could not be covered by the entire insurance industry in the aggregate.

The filing also addresses the possibility of AI systems developing forms of sentience. While this is a theoretical concern for many, Anthropic lists it as a potential business challenge. The company acknowledges that such a scenario would present unusual hurdles for its operations and goals.

Furthermore, the company warns that its models may behave in unintended, misaligned, or autonomous ways. These risks could potentially damage the company's reputation or its ability to function. By including these points, Anthropic is signaling to investors that it views the existential risks of AI as tangible business liabilities.

Market and infrastructure

The company is also preparing for the physical challenges of expanding its data center footprint. As it faces increasing opposition to data center development within the United States, Anthropic plans to expand its infrastructure overseas. This move is necessary to keep up with the demand for more compute power.

In terms of the competitive landscape, Anthropic's filing is surprisingly narrow in its scope. While many companies compete in the AI space, Anthropic identifies only two developers as direct competition: OpenAI and Google. This suggests a focused view of who poses the most significant threat to their market position.

The filing serves as a comprehensive roadmap for the company's transition to a public entity. It balances the need for massive capital investment with a desire to maintain founder control. For investors, it provides a detailed look at the high-stakes gamble of building the next generation of AI infrastructure.