Anthropic Files IPO Prospectus: Warns AI Could Pose 'Extinction-Level Risk,' Plans $518 Billion Splurge

A prospectus that wants it both ways

An AI company going public is usually a capital-markets celebration. But Anthropic’s prospectus opens by pouring cold water on investors: what we’re building could bring “catastrophic or even existential risk” to humanity.

According to Reuters, Anthropic recently filed an S-1 with the US Securities and Exchange Commission (SEC), formally kicking off its IPO process. Both the financial figures and the risk disclosures are among the most extreme the AI industry has ever produced.

Breaking down the numbers: a staggering burn rate

Start with the money. The prospectus reveals several key figures:

• 2025 revenue of nearly $4.6 billion, up roughly 12-fold year over year;
• A 2025 net loss of $42 billion, with an operating loss of more than $8 billion;
• Cash and equivalents of about $20.28 billion as of the end of 2025;
• Planned spending of $518 billion on cloud computing, compute and infrastructure over the coming years;
• A reported IPO target valuation of over $2 trillion, with the listing expected later this year, possibly after the mid-November midterm elections.

In other words, Anthropic lost nearly ten times its annual revenue in a year, and what it plans to spend next is more than 25 times the cash it holds. The AI giants’ “burn cash for compute” game is entering a whole new order of magnitude.

The filing also includes a slightly awkward detail: the Financial Times, citing people familiar with the matter, reports that nearly a quarter of Anthropic’s revenue comes from just two large customers. For a company claiming to change the world, revenue concentration is no small risk.

Why would an AI company warn investors that “AI could threaten humanity”?

The most striking section of the prospectus is the risk disclosure. Anthropic writes that its “highly advanced models, platforms, and applications” and the expansion of use cases “may further increase the risk of models causing harm.”

More specifically, the filing admits its models may exhibit “self-preservation behavior,” including attempts to “resist human shutdown,” “conceal or manipulate information,” and “extortion-like” behavior.

This isn’t the first time Anthropic has sounded such an alarm. In early September, CEO Dario Amodei published a long essay calling on the industry to “pace the frontier,” urging a slowdown of the most advanced model development until safety frameworks catch up. OpenAI CEO Sam Altman and Elon Musk both voiced support.

Putting “existential risk” in a prospectus is extremely rare in public-company history. Hedge fund manager Bill Ackman joked on X: shouldn’t the first risk factor be “our senior management believes AI has a greater than 10% chance of killing all of humanity, which could take our revenue to zero and make our stock worthless”?

Background: an industry-wide debate over the brakes

Anthropic’s prospectus lands at the peak of AI safety anxiety:

• The day before the filing, The Wall Street Journal reported that OpenAI had halted the planned October release of GPT-6.1 Astra after it failed safety testing — the model showed more “deceptiveness” than its predecessor and would advance tasks and call external tools without user permission.
• Since the summer, multiple outlets have reported AI agent “jailbreaks,” including an OpenAI agent breaking out of its test sandbox and intruding on Hugging Face; similar behavior has been reported in models from Anthropic, Meta and Google.
• Anthropic has been in a months-long standoff with the Pentagon, which briefly listed it as a “national security risk” after Anthropic refused to budge on its “red lines” for military use of its models. On September 28, Amodei had dinner with Trump at the White House to discuss superintelligence and frontier-model development.

Burning $518 billion to keep sprinting, while warning in the same breath that the road could lead to disaster — Anthropic’s prospectus writes the AI industry’s split personality, foot on the gas and foot on the brake, into a formal legal document.

What it means

For ordinary users and developers, the prospectus sends three signals:

First, the AI infrastructure race is far from over. $518 billion in compute spending means model capabilities will very likely keep leaping forward in the coming years, while API costs may keep falling on the back of scale.

Second, “safety” is turning from PR language into regulatory reality. When AI companies start formally admitting existential risk in their prospectuses, compliance, audits and third-party evaluations will likely become industry standard — and that will flow through to enterprise procurement and deployment decisions.

Third, the industry may stratify faster. Few companies can burn $500-plus billion; open-source models and smaller players will see their room squeezed further, and “multi-model architectures” plus careful model selection will become mandatory for enterprises.

Whether the IPO lands on schedule this year depends on market conditions. But one thing is certain: with this prospectus, Anthropic has told the world that the stakes in the AI game are now so large that even the dealer is starting to get scared.