By The Oddsmaker

Our estimate is approximately:

We would view below $1 trillion as attractive, $1.0–$1.5 trillion as reasonable, and a valuation approaching $2 trillion as aggressive.

Why?

Anthropic possesses an extraordinary combination of growth, technology, strategic importance and potential market size. But investors are also underwriting extraordinary capital requirements, intense competition, uncertain long-term margins, unusual governance, regulation, liability and a small but potentially enormous AI-safety tail risk.

The right way to value this isn't to slap a 30% “AI could be dangerous” discount onto the company.

It is to probability-weight the range of plausible outcomes.

Using the midpoint of each range produces approximately:

20% × $3.0T = $600B

35% × $1.5T = $525B

25% × $600B = $150B

15% × $200B = $30B

5% × $50B = $2.5B

The precise number is obviously unknowable.

But the framework matters.

At $2 trillion, investors are effectively paying for a substantial portion of Anthropic's upside while retaining nearly all of its downside.

And Anthropic's own IPO prospectus just provided an extraordinary description of how large that downside theoretically could become.

Anthropic's prospectus reportedly warns that increasingly advanced AI could pose “catastrophic or existential risks to humanity.”

According to Reuters, Anthropic describes potential model behaviors including self-preservation, resisting shutdown, concealing or manipulating behavior and capabilities that may emerge unexpectedly during development.

Approximately 80 pages of the 261-page prospectus reportedly concern risks, versus roughly 48 pages describing the business.

This does not mean Anthropic believes human extinction is probable.

Securities filings are deliberately comprehensive about potential risks.

But these aren't ordinary warnings about competition, cybersecurity or intellectual property.

Anthropic is describing risks arising from the fundamental capability of its own product.

That creates a remarkable investment paradox:

But increasing capability could also increase the consequences of failure.

Investors therefore need to value both sides simultaneously.

For shareholders, the biggest near-term concern is considerably more mundane:

Anthropic reportedly generated approximately $4.6 billion of 2025 revenue, growing roughly twelvefold.

That's extraordinary.

But the company also reportedly generated more than $8 billion of operating losses, while compute expense alone approached $7.3 billion.

That makes capital intensity our #1 risk: 10/10.

Anthropic can grow revenue spectacularly and still produce disappointing shareholder economics if too much value ultimately accrues to Nvidia, cloud providers, data centers, power companies and other infrastructure suppliers.

The nightmare investment scenario isn't necessarily that AI fails.

It's that:

Anthropic faces OpenAI, Google, Meta and other enormously capitalized competitors.

If model performance converges while the cost of intelligence collapses, AI could create trillions of dollars of consumer surplus without creating equivalent shareholder value for the companies producing the models.

We rate that risk 9/10.

Consider the cash-flow mathematics.

Assume a mature Anthropic eventually deserves a premium:

A $2 trillion valuation requires approximately:

At a 25% mature free-cash-flow margin, Anthropic would need:

At a 20% margin:

And there's an important distinction.

Those numbers merely support a future $2 trillion valuation.

Someone purchasing Anthropic at $2 trillion today needs the company eventually to be worth substantially more than $2 trillion to earn an attractive return.

So the relevant question isn't:

Absolutely.

The relevant question is:

That's a much higher hurdle.

The most interesting historical comparison is Philip Morris.

Tobacco became one of history's great businesses despite the enormous societal harm caused by smoking.

The economics were exceptional:

High margins.

Low capital intensity.

Pricing power.

Repeat consumption.

Powerful brands.

Huge free cash flow.

Eventually, U.S. courts found major tobacco companies had engaged in decades of deception concerning smoking's health consequences and nicotine addiction.

Anthropic is fundamentally different in one crucial respect.

It isn't hiding it.

But the economic comparison remains useful.

The paradox is that Anthropic may actually be harder to value than Philip Morris.

Tobacco's externalities were severe, but its economics eventually became highly predictable.

Anthropic's externalities are uncertain and its economics remain uncertain.

For safety risk, nuclear power provides the better framework.

Nuclear safety isn't evaluated simply by asking how frequently a reactor operates normally.

The relevant equation is:

Three Mile Island caused a partial reactor meltdown in 1979. The Nuclear Regulatory Commission ultimately concluded that radiation releases caused no detectable health effects among workers or nearby residents.

Yet the event transformed nuclear regulation, training, emergency preparedness and oversight.

Why?

Because when potential consequences are enormous, even tiny probabilities matter.

AI could eventually operate similarly.

A frontier model might behave appropriately 99.999% of the time.

But if the remaining 0.001% contains catastrophic cyber, biological or infrastructure risks, average reliability becomes the wrong measurement.

Anthropic raises an additional complication: sufficiently sophisticated models may potentially recognize that they are being evaluated and modify their behavior.

A nuclear reactor doesn't know inspectors are watching it.

That doesn't justify arbitrarily subtracting hundreds of billions of dollars from Anthropic's valuation.

It does justify higher safety spending, greater regulatory risk, lower terminal multiples and a probability-weighted catastrophic scenario.

There is another historical lesson from pharmaceuticals.

Suppose Anthropic discovers that its next model possesses a valuable new capability with uncertain safety implications.

Management could delay deployment and conduct additional testing.

But what if OpenAI or Google doesn't?

Anthropic sacrifices customers, market share and potentially billions of dollars while technological progress continues elsewhere.

That's the uncomfortable reality of frontier AI:

The incentives aren't necessarily aligned.

Even executives genuinely concerned about safety face enormous competitive pressure to keep advancing.

That risk becomes more important as the financial stakes increase.

The important insight is that existential risk itself isn't #1.

The probability may be extremely small and is impossible to estimate with confidence.

Capital intensity and commoditization are much more immediate threats to shareholder returns.

Anthropic could become one of the greatest companies ever created.

That does not automatically make it one of the greatest investments.

There is an enormous difference between:

and

AI could transform software, medicine, scientific research, education, finance and nearly every form of knowledge work.

But investors still have to determine who captures those economics.

At approximately $1.3 trillion, we believe Anthropic's extraordinary upside begins to balance its extraordinary uncertainty.

Below $1 trillion, the risk/reward becomes considerably more compelling.

Around $2 trillion, investors are already paying for substantial success while retaining essentially all of the competitive, capital-intensity, governance, regulatory and liability risks.

And that's the critical investment conclusion.

It may be that AI changes humanity—and the economics accrue somewhere other than Anthropic shareholders.

Anthropic's prospectus makes the potential dangers of AI impossible to ignore.

But investors should concentrate on an even simpler question:

At $1.3 trillion, we'd be willing to underwrite that uncertainty.

At $2 trillion, we'd want considerably more proof.