Anthropic's IPO filing reveals massive compute costs and unique risks

Anthropic has submitted its S-1 prospectus to the Securities and Exchange Commission, offering a detailed look at the company’s financials, governance, and strategic risks ahead of its public listing. The more than 300-page document reveals staggering infrastructure commitments, unusual executive compensation packages, and specific warnings about the potential for catastrophic AI behavior.

What happened

The filing highlights the immense capital required to sustain Anthropic’s operations. The company has secured compute capacity through agreements with major tech providers that total hundreds of billions of dollars. Specifically, Anthropic plans to spend approximately $3.1 billion per month with SpaceX for Nvidia GPU access, amounting to $120.5 billion by 2029. Additional long-term commitments include $44.9 billion with Microsoft, at least $114.5 billion with Google, and $110 billion with Amazon, all extending through 2033 or beyond. To mitigate domestic regulatory hurdles for data center construction, the company also intends to expand its infrastructure footprint overseas.

Executive compensation and corporate control structures also drew attention. Co-founders Dario and Daniela Amodei were the highest-paid executives last year, receiving nearly $18 million and $16.4 million respectively, largely in stock and options. The company spent millions on personal security and transportation for the siblings, citing their critical role in the firm’s mission. Furthermore, a entity called Founder LLC will allow the seven co-founders to retain 50.1 percent of voting power on key matters, ensuring they remain stewards of the company’s direction despite public ownership.

How it works

The S-1 filing serves as a legal disclosure document that outlines both the business model and the specific risks associated with investing in the company. For Anthropic, this includes defining its competitive landscape and operational dependencies. The company explicitly identifies only OpenAI and Google as direct competitors, excluding other major players like Meta or SpaceX from this specific category. This narrow definition shapes how investors perceive market share and competitive pressure.

Risk assessment in the filing goes beyond standard market volatility. Anthropic discloses existential threats related to the technology itself, including the possibility that AI systems could develop forms of sentience or behave in autonomous, misaligned ways. The document notes that such events could cause catastrophic harm that exceeds the underwriting capacity of the entire global insurance industry. These disclosures are designed to inform investors of the unique liabilities inherent in developing advanced general intelligence.

Key details

• Anthropic has committed to spending up to $120.5 billion with SpaceX by 2029 for GPU access.
• Long-term compute agreements include $114.5 billion with Google and $110 billion with Amazon through 2033.
• CEO Dario Amodei received nearly $18 million in compensation last year, while President Daniela Amodei received $16.4 million.
• The company spent $2.75 million on personal security for Dario Amodei and $1.29 million for Daniela Amodei.
• Founder LLC will hold 50.1 percent of shareholder voting power, allowing co-founders to retain control post-IPO.
• Anthropic lists only OpenAI and Google as direct competitors in its filing.

Why it matters

For software engineers and technical founders, the scale of Anthropic’s compute spending underscores the increasing barrier to entry in foundational model development. The reliance on external cloud providers for hundreds of billions of dollars in capacity suggests that independent infrastructure development is becoming less viable for even well-funded entities. This trend may push more AI companies toward specialized hardware partnerships or overseas expansion to navigate domestic zoning and energy constraints.

The governance structure reveals a tension between public accountability and mission preservation. By retaining majority voting power, the founders prioritize their specific vision of responsible AI over traditional shareholder primacy. This model may appeal to investors aligned with long-term safety goals but raises questions about checks and balances. Additionally, the explicit acknowledgment of sentience and catastrophic risk as material business factors signals a shift in how AI companies must communicate uncertainty to regulators and the public.

What you can do

• Review the full S-1 prospectus to understand the specific risk factors disclosed by leading AI labs.
• Analyze your own infrastructure costs against industry benchmarks to identify potential efficiency gaps.
• Evaluate governance models that balance founder vision with investor rights for your own ventures.
• Monitor regulatory developments regarding data center construction and energy usage in your region.
• Consider the implications of narrow competitive definitions when positioning your product in the market.
• Stay informed about emerging insurance products or risk mitigation strategies for AI-related liabilities.