Anthropic IPO filing puts SpaceX commitments at $120.5 billion

By Rebecca Stone

Anthropic's S-1 prospectus puts planned SpaceX compute spending at roughly $3.1 billion per month and as much as $120.5 billion by 2029. The figure applies once all planned capacity is delivered and covers access to Nvidia GPUs.

Anthropic filed the document with the SEC. At more than 300 pages, the prospectus gives investors a look under the hood at executive compensation, infrastructure plans, its plans for spending the IPO billions and risks to the business. Several disclosures in the document had not been reported before.

The filing matters because it turns private assumptions into public benchmarks. It ties future model capacity to very large supplier commitments while preserving founder control over key shareholder votes. It also places potentially catastrophic model harm alongside the ordinary competition, regulation and market risks that appear in an S-1.

What does Anthropic plan to spend on compute?

Anthropic's filing lists several long-dated commitments for the compute required to power its models:

• SpaceX: roughly $3.1 billion per month once all planned capacity is delivered, totaling as much as $120.5 billion by 2029, for access to Nvidia GPUs.
• Microsoft: $44.9 billion through 2033.
• Google: at least $114.5 billion through 2033.
• Amazon: $110 billion over the next ten years.

The monthly SpaceX figure depends on delivery of the planned capacity. Anthropic also plans to expand its own infrastructure footprint, including overseas, as it faces increasing opposition to data center development in the US. The filing therefore places both secured third-party supply and direct infrastructure inside the expansion plan.

For investors, those dates and amounts create a measurable execution test: whether contracted capacity arrives as planned and whether Anthropic can expand its own footprint despite local opposition. The scale also gives unusual weight to supplier delivery and infrastructure spending in the public-equity case. Anthropic has made compute access, rather than model development alone, a central part of the disclosure.

The $3.1 billion monthly figure applies only after all planned capacity is delivered. The three other supplier commitments extend through 2033 or across the next ten years. The filing thus gives investors dated obligations against which later capacity announcements can be measured.

What do the Amodei siblings earn?

According to the filing, CEO Dario Amodei received nearly $18 million in compensation last year. Most of that total came from stock and option awards. President Daniela Amodei received $16.4 million.

The two packages sit at the lower end of the Big Tech compensation scale. Anthropic also disclosed substantial spending on personal services:

• $2.75 million for personal security services for Dario.
• $1.29 million for personal protection, home security, and transportation services for Daniela.
• $97,055 for Dario's access to a private jet for personal use.

Those figures cover different categories. The nearly $18 million and $16.4 million represent executive compensation, while the security, home protection, transportation and aircraft expenses are company spending associated with personal use. The filing lists those categories separately from the compensation totals.

Anthropic says its success is “highly dependent on the continued service” of its senior team. The filing calls Dario and Daniela “critical” to advancing the company's research and mission.

That disclosure makes leadership continuity an explicit part of the company's risk assessment. It also places executive pay, personal security and personal transportation in the same public record as the two leaders' roles. Public investors can now see both the size of the benefits and Anthropic's stated dependence on Dario and Daniela.

What catastrophic risks does Anthropic disclose?

Every S-1 carries a long list of potential problems. Anthropic's filing follows that pattern with standard warnings about competition, regulation and market shifts. It also gives unusual attention to harms that could follow from advanced AI systems themselves.

Anthropic warns that AI systems could cause potentially “catastrophic” harm, including harm that could not be underwritten by the entire “insurance industry in the aggregate.” The filing also raises the possibility that Claude, or another AI system, could possess or develop forms of sentience.

The sentience language describes a possible future condition, not a reported fact about Claude. Anthropic separately warns that its systems may behave in unintended, misaligned or autonomous ways in a manner that could damage the company. It thus presents catastrophic harm, machine sentience and misaligned conduct as distinct risk scenarios.

These disclosures stand apart from the standard S-1 risk language. That distinction puts safety and model control in the same public risk record as competition, regulation and market shifts. Investors and policymakers now have Anthropic's own descriptions of those scenarios rather than a generic statement that advanced technology carries unknown risks.

How much control do the founders keep after the IPO?

Anthropic already operates as a public benefit corporation. Its IPO structure adds a separate entity called Founder LLC. Through it, the company's seven cofounders will collectively control 50.1% of shareholder voting power over key matters after the public listing.

Anthropic says the seven founders are distinctly equipped to steward the company's mission of benefiting humanity through responsible AI development. The filing presents continued founder control as part of that mission.

The 50.1% figure gives the founders a majority over the specified shareholder matters, though the filing does not present it as control of every vote. The filing gives investors clear numbers for both corporate form and voting control: a public benefit corporation plus a founder vehicle with 50.1% over key matters. Founder LLC therefore lets seven cofounders retain considerable control after the listing, even as Anthropic becomes a public company.

Which rivals does Anthropic actually name?

Under the heading for direct competition, Anthropic identifies only OpenAI and Google. SpaceX, Meta and other developers are not named in that section, according to the filing. The omission leaves the rest of the AI world outside Anthropic's declared peer group.

That omission creates a notable split in Anthropic's own filing. SpaceX is not listed as a direct model competitor, yet it accounts for the largest stated compute commitment. Google appears on both sides of the document: it is a direct rival and a supplier with which Anthropic has committed at least $114.5 billion through 2033. Microsoft and Amazon appear as compute suppliers rather than named model rivals.

The competitive framing therefore narrows Anthropic's declared peer set to two companies even as the same filing names three other large technology businesses as infrastructure suppliers. That structure gives investors a specific map of the company's declared model competition and a separate map of its resource obligations. It also puts SpaceX's infrastructure role ahead of any model rivalry named in the filing.

Future filings and operating results will now show whether Anthropic can deliver the capacity it has contracted, fund the buildout and preserve founder control without making the disclosed safety risks more material.