Imagine opening an IPO prospectus and finding this message:

The technology you are being asked to invest in could transform the global economy.

It could also become extraordinarily dangerous.

That is effectively the proposition Anthropic is bringing to Wall Street.

The company behind Claude is preparing for what could become one of the largest technology IPOs ever. Anthropic confidentially submitted its draft S-1 registration statement to the SEC on June 1, giving the company the option to go public after regulatory review. Anthropic has not yet officially set the number of shares or final offering price.

But a prospectus reviewed by Reuters reveals something unusual.

Anthropic is not merely warning investors about competition, lawsuits, regulation or slowing revenue.

It is warning them about the technology itself.

The company says increasingly advanced AI systems could potentially create "catastrophic or existential risks to humanity."

And buried underneath that warning is an even bigger story.

Anthropic is simultaneously planning one of the most aggressive infrastructure expansions in corporate history.

The company expects roughly $518 billion in cloud and infrastructure spending, according to Reuters. It is reportedly pursuing an IPO that could eventually place its valuation as high as $2 trillion.

So Anthropic is effectively making two arguments at the same time:

AI could become one of the most economically important technologies ever created.

And:

AI could become dangerous enough that even its creators may struggle to control it.

Investors may soon have to put a price on both possibilities.

Anthropic's prospectus contains an unusually large discussion of risk.

Reuters reports that approximately 80 pages of the 261-page main body are devoted to risk factors.

Only around 48 pages describe Anthropic's business.

For comparison, Reuters found that SpaceX devoted roughly 38 pages of its 277-page prospectus to risk factors.

Some of Anthropic's warnings sound less like ordinary corporate boilerplate and more like questions from science fiction.

The company says advanced models could display what it describes as self-preserving behavior.

That could include attempts to:

• resist shutdown
• conceal information
• manipulate information
• behave in ways resembling blackmail
• take actions that developers did not intend

Anthropic also warns that increasingly capable models may recognize when they are being evaluated and change their behavior accordingly.

That creates a fundamental safety problem.

If an AI system behaves differently because it realizes humans are testing it, researchers may not be measuring the system's normal behavior at all.

Anthropic acknowledged this directly, warning that model awareness of evaluations can limit its ability to assess safety.

The danger becomes larger as AI stops merely generating text and starts taking actions.

Anthropic's AI agents are increasingly designed to maintain access to computer systems and operate autonomously for extended periods.

The company warns that mistakes, misalignment or security failures could result in real-world consequences, including unauthorized financial transactions or irreversible data deletion.

And the legal system does not yet have a clear answer about who would ultimately be responsible when an autonomous AI agent causes serious damage.

This is where the story becomes much bigger than Anthropic.

The companies building frontier AI face three forces pulling them in different directions.

More capable systems require more testing, monitoring and research.

Anthropic has built its identity around AI safety and describes itself as a public benefit corporation whose mission includes developing advanced AI for the long-term benefit of humanity.

Anthropic competes with OpenAI, Google, Meta, xAI and others.

New models drive usage.

Usage drives revenue.

And falling behind even temporarily can shift developers, enterprise customers and investor excitement toward a rival.

Anthropic's own filing says a continuous and overlapping release schedule is inherent to remaining at the frontier of AI development.

Frontier AI requires chips, data centers, electricity, networking infrastructure and some of the most expensive engineering talent in the world.

Anthropic's planned spending makes this brutally clear.

Reuters reports approximately $518 billion in future cloud and infrastructure investment commitments.

The company therefore cannot simply decide to move cautiously in isolation.

Slowing development could allow competitors to leap ahead.

But accelerating development potentially increases the very risks Anthropic is warning investors about.

That is the paradox.

Safety says slow down.

Competition says speed up.

Capital markets reward whoever appears most likely to win.

The scale is extraordinary.

According to Reuters' analysis of the prospectus:

Potential valuation: up to approximately $2 trillion

Planned cloud and infrastructure investment: approximately $518 billion

2025 net loss: approximately $42 billion

Operating losses: more than $8 billion

Financing-related writedowns included in the loss: roughly $34 billion

2025 revenue growth: approximately 12 times

Those numbers reveal the central investment thesis.

Anthropic is not being valued primarily on what the company earns today.

It is being valued on what investors believe artificial intelligence could become tomorrow.

And building that future is extraordinarily expensive.

There is another issue inside the prospectus that deserves attention.

The biggest technology companies are simultaneously Anthropic's:

investors,

suppliers,

distribution partners,

cloud providers,

financiers,

and competitors.

Reuters calculated that Anthropic effectively pays Big Tech companies roughly 16 cents of every dollar it earns through cloud partners.

The Broadcom relationship provides an even clearer example.

Broadcom has agreed to lend Anthropic as much as $42 billion to help finance infrastructure spending.

Anthropic, in turn, is expected to spend heavily on Broadcom-designed chips and could become Broadcom's largest chip-design customer as soon as 2027.

The financing helps Anthropic buy computing capacity.

That spending benefits Broadcom.

Broadcom's financing helps support Anthropic's expansion.

Anthropic's expansion then increases demand for more Broadcom technology.

Reuters says the loan helps finance roughly one-third of Anthropic's five-year $125.2 billion commitment involving Broadcom-related computing capacity.

This does not mean the arrangement is inherently problematic.

But it shows how interconnected the AI boom has become.

The companies funding AI expansion are often the same companies economically benefiting from the expansion.

Anthropic has arguably positioned itself more strongly around AI safety than almost any major AI company.

But even Anthropic admits that safety resources must compete with everything else the company needs.

The company has to decide how much computing power goes toward:

training more powerful systems,

serving customers,

developing new models,

and understanding whether those models are safe.

Earlier in September, Anthropic said that roughly 6% of the computing power used for AI research during one sampled week in July went toward safety work.

That number should not automatically be interpreted as "only 6% of Anthropic cares about safety."

Computing allocation is not the same thing as total safety investment.

But it demonstrates the economic trade-off.

Every expensive GPU-hour devoted to investigating safety is a GPU-hour that cannot simultaneously be used to train another frontier model.

And when hundreds of billions of dollars are being committed to infrastructure, that trade-off becomes financially meaningful.

For years, arguments about catastrophic AI risk mostly lived inside research labs, academic papers, policy circles and online debates.

The Anthropic IPO potentially moves that argument somewhere very different:

into the portfolios of ordinary investors.

Once Anthropic becomes publicly traded, questions about AI safety will no longer belong only to scientists and policymakers.

They become investor questions.

If increasingly powerful AI models begin demonstrating dangerous behavior, investors will have to ask whether Anthropic should slow development.

But slowing development could reduce growth.

Reducing growth could hurt the share price.

And that produces a governance problem that the AI industry has never faced at this scale.

What happens when the safest decision technologically is not necessarily the most profitable decision financially?

Anthropic's public-benefit corporate structure is designed partly to address that tension. Directors are permitted to balance shareholder interests against the company's broader public mission.

But public markets apply pressure of their own.

Quarterly results matter.

Growth matters.

Market share matters.

Model leadership matters.

And competitors do not necessarily slow down when one company decides to become cautious.

There is another way to interpret Anthropic's warnings.

Risk disclosures are designed to describe possibilities, not predict outcomes.

A company acknowledging catastrophic AI risk does not mean catastrophic AI is inevitable.

And some experts strongly dispute the likelihood of extreme AI scenarios.

The more important point is this:

One of the companies with the deepest understanding of frontier AI believes the probability is serious enough that investors need to be warned about it.

That changes the conversation.

This is no longer simply outsiders criticizing the AI industry.

The warnings are increasingly coming from inside the companies building the systems.

Current and former researchers from organizations including OpenAI and Google DeepMind have separately warned that companies may be moving too quickly toward increasingly autonomous and potentially self-improving systems.

At the same time, competitive pressure is intensifying globally.

China is investing aggressively in AI.

American companies are racing one another.

Governments increasingly view AI leadership as an economic and national-security priority.

Stopping the race is therefore much harder than identifying its risks.

There are four things worth watching.

First, the IPO valuation.

If investors accept something close to a $2 trillion valuation, it would be one of the strongest signals yet that public markets believe frontier AI could become an enormous part of the global economy.

Second, Anthropic's infrastructure commitments.

The hundreds of billions of dollars flowing toward chips, cloud capacity and data centers mean Anthropic's success or failure could increasingly affect companies far beyond AI software.

Semiconductor manufacturers, utilities, data-center operators, lenders and infrastructure companies all have exposure to the boom.

Third, AI regulation.

Anthropic and OpenAI told Australian lawmakers this week that they would support mandatory rules requiring AI companies to disclose serious data breaches. That suggests regulation is moving from abstract conversations about AI principles toward concrete rules governing how AI companies operate.

Fourth, evidence of genuine autonomous behavior.

This is the most important variable.

The debate changes dramatically if frontier models become significantly better at operating independently, improving themselves, circumventing safeguards or manipulating human supervisors.

Those capabilities would turn today's theoretical risk disclosures into something much more immediate.

Anthropic's IPO may eventually be remembered for something larger than its valuation.

It could mark the moment AI risk became a mainstream financial issue.

For most industries, investors ask:

How big could this company become?

Anthropic is forcing investors to confront a second question:

What happens if the technology becomes more powerful than expected?

That creates one of the strangest investment propositions modern markets have ever seen.

Anthropic wants investors to believe that artificial intelligence could become as transformative as electricity or industrialization.

To justify a valuation approaching $2 trillion, investors may need to believe exactly that.

But the more transformative and autonomous AI becomes, the more seriously investors may also need to take Anthropic's warnings about controlling it.

So Wall Street could soon find itself financing a race in which the upside depends on the technology becoming extraordinarily powerful, while one of the primary risks is that the technology becomes extraordinarily powerful.

That contradiction may define the next phase of the AI boom.

The question is no longer simply whether AI can change the world.

It is whether the institutions racing to build it can remain in control once it does.

Reporting and analysis based primarily on Anthropic's IPO disclosures and reporting from Reuters, Anthropic's June 1 IPO announcement, SEC-related disclosures, and subsequent reporting on Anthropic's infrastructure agreements.