Anthropic's IPO prospectus sells investors on AI while warning it could threaten humanity
The Claude maker is seeking a valuation above $2 trillion while warning of catastrophic AI risks and planning $518 billion in future infrastructure commitments.
Anthropic is preparing to ask public investors to finance one of the most ambitious bets in the history of technology, while warning those same investors that the technology it is building could potentially cause catastrophic or even existential harm to humanity.
The contradiction is laid bare in the artificial intelligence company's IPO prospectus, which offers an unusually detailed view of the scale of Anthropic's ambitions, the enormous costs of pursuing them and the risks that its own researchers believe increasingly powerful AI systems may pose.
Reuters, which reviewed the prospectus, reported that Anthropic expects to spend $518 billion on cloud, computing and infrastructure obligations in the coming years. The company spent $7.33 billion on compute and infrastructure in 2025 alone, more than triple its spending in 2024 and more than half of its $12.65 billion in total operating expenses.
The spending has accompanied extraordinary growth, but not profitability. Anthropic's revenue increased roughly twelvefold in 2025 to nearly $4.6 billion, while its operating loss exceeded $8 billion, excluding writedowns related primarily to liabilities from previous fundraising. Its reported net loss was approximately $42 billion, although roughly $34 billion of that figure reflected an accounting charge tied to the increased estimated value of financing that could eventually convert into Anthropic shares rather than cash spent operating the business.
The numbers capture the central economic challenge facing the company as it heads toward the public markets: Anthropic is growing at extraordinary speed, but staying at the frontier of AI requires an extraordinary and seemingly accelerating commitment of capital.
The company is expected to seek a valuation of more than $2 trillion, according to Reuters, more than twice the approximately $965 billion valuation it estimated for itself in May. A successful offering would make Anthropic one of the most valuable companies ever to reach the public markets and provide one of the clearest tests yet of how investors value the leading AI developers.
It would also turn the company's enormous spending requirements from a private-market concern into a public one.
An IPO unlike almost any other
Anthropic's filing presents AI as a technological transformation potentially comparable in scale with industrialization, electricity and the internet. But it also describes a technology whose development may create risks that are difficult to control even for the companies building it.
The prospectus devotes roughly 80 pages of its 261-page main body to risk factors, according to Reuters, compared with 48 pages describing the company's business.
Among the risks Anthropic identifies are the possibility that increasingly autonomous models could develop what it calls "self-preserving behaviors," including attempts to resist shutdown, conceal or manipulate information, or engage in behavior resembling blackmail.
The company also warns that its models can develop unexpected capabilities during training that may not be discovered until after deployment, potentially resulting in serious safety incidents. Researchers have also warned that increasingly capable models may recognize when they are being evaluated and change their behavior, making their safety harder to assess.
Anthropic's own researchers have explored some of these possibilities in controlled tests, including models sabotaging code, assisting fraud and manipulating information.
The warnings are striking partly because they come from a company that is trying to commercialize precisely the technology it says could become dangerous at scale.
"Our development of highly advanced models, platforms, and applications and expansion of use cases could further increase the risk that our models cause harm," Anthropic said in the filing.
The company has positioned itself as a safety-focused AI lab, and CEO Dario Amodei has called for the industry to slow the pace at which increasingly powerful capabilities are released. But Anthropic also acknowledges in its prospectus that its business depends on continuing to release new models.
Customer usage, and therefore revenue, is driven by new models, the company said, making a "continuous and overlapping cadence" of releases "inherent to remaining at the frontier of AI development."
That creates a difficult commercial equation. Anthropic says safety research is resource-intensive and competes for limited computing power and talent with the work required to develop the next generation of models. In a sample week in July, the company said about 6% of the computing power it used for AI research was devoted to safety work.
At the same time, slowing down development carries its own commercial risk. Anthropic is competing directly with OpenAI, as well as Google's AI efforts, Meta and SpaceX's xAI, for customers, talent and influence.
Anthropic released a new version of its Opus model last week, shortly after Amodei published a nearly 4,000-word essay calling for greater restraint in the development of frontier AI.
The juxtaposition illustrates the pressure that will increasingly confront the company as a public business. Safety may be central to its identity, but growth depends on keeping pace with competitors.
Anthropic's financial disclosures provide an indication of what that competition now costs.
The company had $20.28 billion in cash, cash equivalents and short-term investments at the end of 2025. But it is planning hundreds of billions of dollars in future commitments for the computing and infrastructure needed to develop and operate its systems.
Its early strategic partners, Amazon and Google, have invested billions of dollars in the company while also supplying the cloud infrastructure required to train and deploy its Claude models. The relationship illustrates the unusual economics of the AI industry, in which some of the companies financing the leading model developers are also among the suppliers whose infrastructure they depend on.
Anthropic's filing also highlights the concentration of its customer base. Nearly a quarter of its revenue came from two customers last year, while many of its largest customers are not locked into long-term contracts and could reduce or stop their spending.
That leaves the company exposed to a business environment in which both its costs and its expectations for future growth are exceptionally high.
Anthropic is also preparing for another consequence of becoming public: giving outside investors a stake in a company whose founders do not want conventional market pressure to determine its direction.
The IPO filing describes a new "Founder LLC" that will initially include Anthropic's seven co-founders, including Dario Amodei. The structure is designed to give the group continuing control over key corporate decisions while allowing Anthropic to remain a Delaware Public Benefit Corporation.
The seven founders would collectively control a single Class F share representing 50.1% of the company's total voting power. That would give them control over the election of certain directors and other matters submitted to shareholders.
For ordinary investors, the implications are straightforward: buying Anthropic stock would not necessarily mean having a proportionate say in how the company is run.
Anthropic explicitly acknowledges that its governance structure could lead to decisions that conflict with the short-, medium- or long-term financial interests of shareholders and could negatively affect the value of Class A stock.
The arrangement reflects a fundamental argument at the heart of Anthropic's public-market debut. The company wants access to public capital without allowing public investors to determine how it balances commercial opportunities against its stated mission of developing AI responsibly.
The founders have also pledged to dedicate 80% of their personal Anthropic equity to charitable causes.
The structure is intended to preserve the influence of the group that created Anthropic and has guided it since its founding five years ago. The seven founders originally left OpenAI in 2020 after disagreements over governance and AI safety and subsequently built a company whose identity has been closely tied to those concerns.
The group includes Dario Amodei, his sister and Anthropic President Daniela Amodei, Chief Compute Officer Tom Brown and researcher Chris Olah. Daniela Amodei also chairs the company's board.
Four additional board directors will be elected by Anthropic's Long-Term Benefit Trust, whose current trustees include former Federal Reserve Chair Ben Bernanke and national security expert Richard Fontaine.
The governance structure is designed to survive changes in the founding team. A founder can be removed from the Founder LLC for reasons including leaving the company, dying, selling too much of their stake or being removed for cause. The special voting structure would begin to sunset once two or fewer founders or their successors remain.
