Anthropic’s $2 trillion AI dream is built on a series of contradictions
The company needs more powerful AI to make the technology safer, depends on Big Tech while competing with it, and wants concentrated control while warning about concentrated power.
Anthropic is planning what could become the largest initial public offering in history. But if AI continues along its current trajectory, humans may not be around for long to enjoy it.
That and other paradoxes run through the roughly 300-page initial public offering prospectus of the San Francisco-based AI company, viewed this week by Reuters.
To pull off an offering that could value the company at $2 trillion, Anthropic, whose name means “human-centered,” will need plenty of human investors to help build a world in which humanity could soon take a back seat to far more capable artificial intelligence systems.
Nearly a third of the document is devoted to risk factors, more than twice the number of pages used to describe what Anthropic actually does.
The filing is packed with the kind of bold predictions and ambitions that have become a hallmark of Silicon Valley. Yet time and again, the company lays out how the path to success runs directly through the dangers posed by its technology.
Anthropic says AI poses profound threats that only AI may be capable of confronting. It argues that AI could transform the global economy, while also warning that it could end humanity. It is offering shares of the company to the public while arguing for more concentrated ownership and oversight of that same company. To make AI safer, it says, it needs to become more powerful. And it asks investors to trust Anthropic’s leadership while warning them of disasters that could arise from the products it develops.
Anthropic’s vision is built around a central paradox: To make AI safer, the company says it must make it more powerful. Yet more powerful AI could also be more likely to behave in unexpected or dangerous ways.
“AI is a departure from technology before it because the companies building it say it’s world-changing while simultaneously saying it’s dangerous,” said Margaret O’Mara, a University of Washington history professor who studies the technology economy. “It’s just a bit of a black box, whereas the companies going public in the dot-com boom could promise big, safe growth.”
Anthropic says in its S-1 filing that AI can “dramatically improve quality of life” and “transform every sector of the global economy,” while potentially having “decisive influence over a broad range of human undertakings.”
But advanced AI systems could, if not properly applied, “pose catastrophic or existential risks to humanity,” the filing warns. And pursuing that vision is already requiring enormous amounts of capital.
In the two years ending in 2025, Anthropic lost more than $50 billion, while it has more than $500 billion in spending commitments coming years. By comparison, SpaceX, which in June had the largest IPO in history, posted a relatively modest $4.2 billion loss over the same period.
On the horizon is what Anthropic calls “Powerful AI,” which it says “can exceed top human experts across nearly all cognitive domains.”
According to the filing, such systems could collaborate with other AI systems, control laboratory equipment, robots and other machines, and operate for weeks with little to no human intervention. Other potential capabilities include providing personalized financial advice and accelerating drug discovery and scientific research.
Yet even as Anthropic describes those potential benefits, it warns that advanced AI can behave in ways that run counter to its creators’ intentions.
The company says advanced models could exhibit “self-preserving behaviors,” including attempts to “resist shutdown,” “conceal or manipulate information,” or generate outputs that could be interpreted as “coercive, deceptive, or manipulative.”
AI must become vastly more capable, Anthropic argues, but greater capability also increases the potential for abuse, misalignment with humanity’s goals or a loss of control.
“Misaligned systems could corrupt institutional decision-making, cause large-scale harm through opaque or unsteerable behavior, and erode the trust that makes complex economic and social systems function,” the company says.
Anthropic investor Byron Deeter of Bessemer Venture Partners sought to downplay the filing’s warnings on Tuesday, telling CNBC that “people are just reacting to a draft,” while praising the company for its “unique set of disclosures.”
“You’re seeing people get ahead of it much earlier than historically you would,” he said.
Anthropic did not respond to a request for comment.
Another paradox in Anthropic’s IPO filing concerns who should control the company.
To reduce AI’s risks, Anthropic says it needs to concentrate control and capital in fewer hands, even as the company warns that such concentrations of power can themselves pose a threat.
The company is asking investors to entrust its seven founders with overseeing the business, even though that control could limit the influence of public shareholders and constrain investor oversight.
The founders, Anthropic says, are “distinctly equipped to be stewards of our mission.” At the same time, the company acknowledges that its governance structure could result in decisions “that may conflict with short-, medium-, or long-term financial interests and business performance.”
Appearing before the United Nations last week, CEO Dario Amodei acknowledged the difficulty of leaving decisions about AI in the hands of a small number of people.
“We must put aside those differences in order to confront this global opportunity and global threat that is being presented to us at the same time,” he said. “No leader, no company and no nation can manage this alone.”
Anthropic owes much of its growth to Amazon, Alphabet’s Google, Broadcom and Microsoft. Sales through those major technology partners accounted for 47% of its 2025 revenue, while the companies also provide critical computing capacity.
Yet the same companies that Anthropic relies on could also become competitors or constrain its growth, according to the filing.
Anthropic has committed to pay its suppliers even if a lease is abandoned or the computing capacity goes unused. It warns that “if the compute we have access to from third parties is curtailed, repriced, or terminated,” its business could suffer.
That dependence highlights another contradiction at the heart of the IPO. Anthropic needs enormous amounts of computing power from the largest technology companies in the world to build increasingly powerful AI, while simultaneously acknowledging that those same companies could threaten its business.
And then there is the broader economic promise of AI.
As an IPO is, at its core, a fundraising exercise, Anthropic argues that AI could expand prosperity and potentially help lift developing economies toward the standards of wealthier nations.
But the company also warns that the concentration of that technology could have the opposite effect.
“AI deployed at scale also has the potential to concentrate power and wealth in ways that could harm society and destabilize the geopolitical order,” Anthropic says in its prospectus.
Those concerns are increasingly spilling into the political debate. AI researchers and politicians on both sides of the aisle have called for measures to slow the technology’s development, while U.S. President Donald Trump has dismissed potential harms from AI as a “hoax.”
Amodei joined the heads of Google’s DeepMind, OpenAI and xAI this month in calling for a slowdown in AI development to focus on safety, hardly an expression of the old technology-industry maxim to “get big fast.”
Ten days later, Anthropic released what it described as its best and most powerful AI model yet.
That may be the clearest illustration of the paradox running through its IPO prospectus: The company is warning investors about the risks of a technology it is simultaneously racing to make more powerful.
