Anthropic's IPO: why cheap AI could pay off for you... maybe

Sam Altman (the CEO of OpenAI) and Dario Amodei (the CEO of Anthropic) want your money. They've both filed for IPOs.

Should you give them any of your hard-earned money?

We have time to decide, but let's think through this with what we know today...

ANTHROPIC: Sevenfold Revenue Pace In Seven Months

That's roughly what happened to Anthropic's annualized revenue pace.

It went from about $9 billion at the end of 2025 to more than $65 billion by July.

Now third-party tracker TickerTrends estimates Anthropic has reached roughly $76.8 billion in annualized revenue run rate.

But there is a wrinkle: TickerTrends estimates the latest month grew only 1.8%

In other words, after an extraordinary revenue ramp, the curve may finally be bending.

I'm not calling that a stall because these are outside estimates, not audited Anthropic numbers, and annualized revenue can exaggerate short-term changes.

But if you're thinking about buying Anthropic, or OpenAI, when they go public, I'm looking at something much more important...

Because something else is happening at the same time.

The price of using AI is collapsing. That could also create two enormous opportunities.

First, cheaper intelligence could cause us to use vastly more of it.

Second, useful intelligence can be worth much more than it costs to produce.

The Price Of Intellignence

We talked a lot about compute in our last issue: the giant clusters of Nvidia chips that power AI.

Think of compute as the raw material.

One way to measure what comes out the other end is the price of tokens, which are the small units of information AI models process when they read your question and generate an answer.

Silicon Data tracks what the market actually pays for one million of those tokens across hundreds of AI models.

In late May, its index was around $2 per million tokens. By early September, it had dropped below $1 for the first time.

In other words, the effective price of using AI models was cut roughly in half in a matter of months.

And the models themselves are getting better, too.

At first glance, that sounds terrible for OpenAI and Anthropic. And I'm sure you'll read some shallow analysis that concludes you should stay away from these IPOs...

But we're going to dive deeper today because we're focused on how technology actually evolves.

So, if machine intelligence keeps getting cheaper, how do the companies selling it ever earn extraordinary profits?

There are two answers.

ANSWER #1: We May Use A Lot More Intelligence

Revenue isn't just price. It is price multiplied by volume.

If the price of something falls by half but we use ten times as much of it, spending still increases 5X.

That could be exactly what happens with AI.

Today, most people still intentionally open an AI product: ask ChatGPT a question, have Claude write something or use an AI coding tool.

But intelligence is starting to appear inside everything else.

Software will use it in the background. Customer-service systems will use it. Voice assistants.

Cars, robots, search engines, financial systems and security tools will use it. And agents won't wait for humans to type most prompts.

They can work for minutes or hours doing things like searching, calling software, writing code and checking their own work.

That requires vastly more use.

In fact, this chart shows something interesting.

Among the companies spending the most on AI, the typical company in the top 1% is now spending roughly 8X as much as the typical company in the top 10%.

That doesn't prove AI providers have pricing power.

It shows something else:

The companies getting the most use out of AI are consuming dramatically more of it.

Falling token prices, in other words, don't necessarily mean a shrinking market. They can expand the market.

When computing became cheaper, we didn't stop buying computing.

That's not what technology wants. It wants to be everywhere.

Chips found their way into phones, cars, factories, televisions and eventually almost everything.

In fact, a Porsche in 1970 had zero onboard computers. In 1980 it had one. Today it has over 70.

AI will follow the same path.

Cheaper machine intelligence simply means we find exponentially more places to use it.

ANSWER #2: The Value of Intelligence

There is another reason falling token prices may not determine what OpenAI and Anthropic can ultimately charge.

The cost of intelligence is not the same as the value of intelligence.

A business doesn't care how many tokens Claude used to analyze a contract. It cares whether Claude found the clause that would have cost the company $2 million.

As a programmer, I don't care whether an agent used 50,000 tokens or 500,000. I care whether it completed the task successfully.

The commodity is tokens. The product is useful intelligence.

And those two prices do not have to move together.

If one or two AI systems become dominant, we could also see more pricing power.

Here's why...

The Real Prize Is A Share of The Payroll

Put yourself in the shoes of a business owner.

Suppose ten employees cost $1 million a year.

Now suppose four employees, working with AI, can reliably produce the same output as ten, and the AI costs another $200,000.

The owner spends $600,000 a year instead of $1 million. That's $400,000 in pure profit.

Now suppose the AI company's inference cost falls by half. Does it have to cut that $200,000 bill to $100,000?

Not necessarily.

If the system is still creating hundreds of thousands of dollars in economic value, the customer may happily keep paying $200,000. Or even more.

That's especially true if the AI has become deeply embedded in the company's records, workflow and institutional memory.

Competition should push prices down.

But if one or two systems become dominant, and switching means retraining agents, reconnecting data, and rebuilding workflows, those providers gain pricing power.

That is why today's $20, $100 or $200 AI subscriptions may severely underestimate the eventual price of useful intelligence.

A $100 seat could become a $1,000 agent.

A software subscription could become a fraction of the labor cost it replaces.

So there are two ways this market can get much bigger:

We can consume far more intelligence as it gets cheaper. Providers may eventually capture more of the value that intelligence creates.

The Bear Case Is Commoditization

There is still a bear case.

Producing AI becomes easy. Open-source systems get good enough. Customers can switch providers easily.

Token prices collapse faster than usage grows, while competition prevents the labs from capturing much of the value their models create.

In that world, AI could transform the economy without producing extraordinary profits for OpenAI or Anthropic.

The questions are: How fast will consumption grow as intelligence gets cheaper?

And who captures the difference between what intelligence costs to produce and what useful intelligence is worth?

But there is also another way...

Business Models Bring Revenue

Remember...

Google paired a breakthrough search engine with advertising and turned usefulness into an extraordinary business.

AI companies are still discovering their version of that.

OpenAI has subscriptions, enterprise customers, developers...

And advertising.

Sound familiar?

Anthropic increasingly sells Claude into valuable business workflows.

In each case, the technology matters...

But the business model determines how much of that value gets captured.

The bigger step may come when AI stops being priced primarily as software and starts being priced more like thought work.

The Machines Still Have To Be Paid For

Anthropic's growth has been astonishing. So have its bills.

According to the IPO prospectus Reuters has seen, the company spent about $7.3 billion on compute and infrastructure in 2025, against about $4.6 billion in revenue.

Its operating loss was still more than $8 billion.

You may also see a $42 billion loss for 2025 in the headlines. About $34 billion of that was an accounting charge tied to fundraising, not cash spent running the business.

And those 2025 numbers are already old. Anthropic's second-quarter revenue alone was about $11.5 billion, more than it made in all of 2025, and it has reportedly posted an adjusted operating profit.

But Reuters also reported that Anthropic has about $518 billion of infrastructure commitments stretching over years.

OpenAI is making a similarly enormous bet.

Its annualized revenue pace is approaching $70 billion. It has reportedly been seeking at least $30 billion in new funding at roughly a $1.4 trillion valuation.

That valuation is about 20 times its current annualized revenue pace, again, revenue, not earnings.

That's why these two arguments matter so much.

If inference gets cheaper while usage explodes, revenue can keep growing despite falling unit prices.

And if AI providers can capture part of the economic value their systems create, margins could eventually become extraordinary.

Neither outcome is guaranteed.

Price Still Matters

Give Anthropic $100 billion in annual revenue and a 20% net profit margin.

That produces $20 billion in annual earnings.

At the $2 trillion valuation some investors have floated for its IPO, you're paying 100 times those earnings. And those are imagined future earnings, not what the company earns today.

So yes, I want exposure to intelligence.

I also want a price that leaves room for reality.

For me, the IPO analysis comes down to five things:

First, usage growth.

How quickly does the amount of intelligence customers consume expand as prices fall and agents spread?

Second, revenue durability.

Is Anthropic's possible slowdown temporary, or are we seeing the beginning of a more mature growth curve?

Third, pricing power.

Can these companies charge for the value their systems create, or will competition force most of the savings to customers?

Fourth, unit economics.

As inference gets cheaper, how much of that efficiency stays with the provider?

Fifth, capital intensity.

How much cash must these companies spend before the economics finally turn?

Buffett For The Core, Asymmetry At The Edges

For roughly 80% of my portfolio, I still like Buffett's approach: own the SPY...

And for stocks, understand the business, look for durable advantages, care about the price and let time work.

Nvidia (NVDA), for example, has been alongside my SPY position for a long time.

With 20% or less, I look at other assets including asymmetric opportunities where the upside could be enormous relative to the capital I'm willing to risk.

AI belongs in that second bucket.

That doesn't mean I'll chase either IPO.

Yet.

What I'm Betting On

Models will improve. Agents will become more capable. Machine vision and world models will advance.

And cheaper compute will let us put intelligence in places where it would be absurdly expensive today.

My thesis is really two bets:

The world will consume vastly more intelligence as its price falls. And useful intelligence can be worth much more than the compute and tokens required to produce it.

If both are true, the market can become enormous even while the unit price keeps falling.

I found a way to invest in Anthropic and OpenAI pre-IPO, and I'm up 112% on Anthropic and 57% on OpenAI.

I also plan to add more in the future.

Future research and the IPO price will determine when and how much.

When Can We Actually Buy Shares On The Stock Market?

Anthropic's Nasdaq IPO appears to be first in line.

It confidentially filed for an IPO in June, and Reuters reports its debut will likely come after the November midterm elections.

My working estimate remains late November or December 2026 if the IPO market cooperates.

OpenAI has also confidentially filed, but CEO Sam Altman has ruled out a 2026 IPO.

Reuters says it is now expected to list in 2027.

We're tracking both for you.

When the filings become public, we'll dig into margins, cash burn, infrastructure commitments, risks, shareholder rights and dilution.

And I'll be watching two curves especially closely:

• How fast the price of intelligence falls...

• And how fast the world finds new ways to use it.

Always be prospering,

Harry Seldon
Moonshot Minute

P.S. #1: If you haven’t already, check out DD’s sell and buy alert below for Premium. It’s a 150% winner and a fresh new buy alert he sent out on Friday.

🔓 Premium Content Begins Here 🔒