We’ve all been overwhelmed by the “AI is everywhere” narrative.
We weren’t asked if we needed AI. Businesses are just putting it into what feels like every product and service.
There has been a narrative about the earth-shaking possibilities of Artificial Intelligence. At the same time, it is hard to escape the feeling that something doesn’t add up.
Marrying those two things has been tough to do because we haven’t had anything publicly available.
Until now.
Anthropic has filed an S-1 for the world to see. The numbers are…
$16.3B in revenue. $80.7B in net loss.
Those are the two top-line numbers.
The company used non-GAAP numbers to make the financial argument seem okay.
This is the moment the AI bubble talk becomes real.
An S-1 is a marketing document…a story. This story is shaped to tell you what the company wants you to know.
Framed in a way that is best for them.
That’s why the $80.7B net loss gets the headlines.
It is designed to shock.
Job well done.
But it is a misdirection…a magic trick.
The number underneath the net loss is the one that matters.
The GAAP operating loss was $693M.
This is the cash consumed by the business. This is what the business really looks like.
The gap between those two numbers is nearly $80B. $72.4B of that is a non-cash accounting charge.
What does this mean?
Anthropic raised a lot of money selling convertible notes. These are essentially IOUs that convert to stock sometime in the future.
As the company’s valuation grew, the value of these IOUs grew with it.
Accounting rules require a company to record these increases in value as expenses on the company’s income statement, even though no cash has changed hands.
The larger the valuation, the larger the paper loss.
Think of it as a receipt for the hype of the AI bubble.
That’s part one.
The second part of the trick…
Anthropic reports adjusted operating income of $1.6B for the first half of 2026.
Not bad.
But this is a non-GAAP number. A company-defined metric with its own adjustments. In this case, it swings from a $1.8B loss one year earlier to a $1.6B profit today.
Three numbers stand out.
The net loss of $80.7B. That’s shocking.
An operating loss of $693M. That’s the reality.
An adjusted profit of $1.6B. That’s a comfort.
The first number makes the second number look small. The third number tricks you into believing that the second number isn’t that bad.
Three numbers. Three stories. You can decide which one you want to buy into.
The document doesn’t hide the cash burn. It reframes it as the minor cost of huge potential.
The magic trick isn’t creative accounting. The magic trick is the framing. The big number distracts from the real business. This sets up the sales pitch.
This is how Anthropic frames its losses.
Let’s also look at how they frame the revenue.
The $16.3B revenue is the star. This is used as a way for Anthropic to signal “product-market fit.”
“Look, there is a market for AI.”
This is their reason that the market should look past the cash burn.
But it is a trick…
The trick is the mechanism.
The business press calls it “circular financing.” In the dot-com years, they called it “round-tripping.” The older, telecom-era term was “vendor financing.”
The reality: the name changes. The mechanism doesn’t.
How does the mechanism work?
It is a circle.
Company A invests in Company B.
Company B takes that investment and buys products or services from Company A.
Company A books that money as revenue.
The money is moved around on a spreadsheet and magically called “growth.”
In this case, Microsoft and Nvidia invest around $15B in Anthropic. Anthropic commits to spending $30B on Microsoft’s cloud services and Nvidia’s chips.
The money goes out as an “investment” and comes back as “revenue.”
Magic.
We’ve seen this before. In the 1990s, Lucent and Nortel loaned billions to startup telecom companies so that these startups could buy their equipment.
Lucent and Nortel booked the sales as revenue…immediately.
When external capital dried up, the startups defaulted. Lucent’s revenue crashed 69%.
This is not growth. It is financial engineering.
The numbers aren’t fake. But they don’t prove a healthy business.
Why?
Because this is a closed ecosystem pretending to be an open market.
Another way of looking at this: when your biggest investors are also your biggest customers, you aren’t measuring demand.
You are measuring a business cycle.
Unfortunately, cycles always end.
All of this is premised on a bet.
The S-1 points out that Anthropic has made $450B in non-cancelable compute commitments.
They also tells us that they need hundreds of billions of dollars more.
The math: for every dollar in revenue, Anthropic owes twenty-seven dollars on future commitments.
This math is built on top of a business model that’s shown explosive growth once. And they have no plan that would lead investors to expect that growth to happen again.
This doesn’t seem like a calculated risk.
It seems like a bet that there will always be someone a little more reckless to buy the story before the bill comes due.
A greater fool.
We’ve seen this pattern before.
In 1929, the vehicle was the investment trust.
Goldman Sachs created the Goldman Sachs Trading Corporation, a pyramid of leveraged holding companies.
Goldman was building financial structures…not businesses.
The IPO was the moment the structure was sold to the public.
Then the crash came. 98% of the value was wiped out. Because there was no operating business underneath.
In 1999, the mechanism was the dot-com. Everything dot-com all the time.
Companies like Pets.com and Globe.com IPO’d with massive losses and no path to profitability.
The IPO wasn’t a milestone of maturity.
It was a prayer that the public markets would buy the story and keep the burn rate alive.
Revenue was real.
The business model…not so much.
When the external capital dried up, the businesses went failed.
In 2019, it was WeWork. A $47B valuation built on a story about WeWork being a tech company, not a real estate company.
But it was really built on a narrative.
First, offices. Then, tech. Eventually, vibes.
The IPO was going to be the exit. The market saw the numbers, the messy governance structure, and the psychobabble dressed up as strategy…and said, “No.”
SoftBank was in so deep that they couldn’t walk away.
They invested an additional $10B to take over a business they valued at $8B at the time the IPO was looking for $47B.
To save face. And to pray that something would save their investment.
An analyst made the point that SoftBank had more invested in the company than it was worth.
In this case, the market walked away, but a greater fool still stepped in.
In 2021, it was Rivian. A $150B valuation with almost no revenue.
The promise was “the next Tesla.” An affordable car for the masses…the R2.
On paper, Rivian was worth more than Ford or GM. But they had delivered very few cars.
The reality was the company was still selling cars at a loss. The stock crashed, losing around 90% of its value.
And we are still waiting for the next model to be the miracle that turns the company around.
Anthropic is running the same playbook. With a much different scale.
They are selling the public market on a story about AI revolution and infrastructure.
They ask that the market ignore that twenty-seven to one math problem.
The names change. The mechanism does not.
This makes the Anthropic S-1 a confession.
It tells you exactly what the business model requires…a miracle.
The S-1 admits that AI models may develop faster than the company can keep them safe. It admits that the models may take harmful actions on their own.
The document sanitizes this by calling these risks “self-preservation behavior.”
Amazing language. But it is pretty alarming that Anthropic highlights blackmail as a potential risk.
The S-1 admits the financial reality. One dollar in, twenty-seven committed to go out.
Anthropic is telling the markets that the technology could be an existential risk. They are telling the markets that the financial framework needs a miracle.
…And they are the only ones that can build this.
This is not strategy. This isn’t a calculated investment opportunity.
This is a prayer dressed up as a prospectus.
The S-1 is a confession that the business model depends on potentially existential risk, a future that may not happen, and finances that any investor should call unwise.
The scary thing: the market just might go for it.
Because the market has been trained to reward the story and not the business.
You aren’t the market.
You aren’t Anthropic. You don’t have Nvidia in your cap table. You don’t have the $112B in cash and cash equivalents necessary to fuel a “circular financing” cycle.
…And you definitely don’t have the luxury of having an $80B loss waved off as “creative accounting.”
Anthropic lives on a financialized business model.
Value creation is an afterthought.
This financialized business model is only extended to the chosen few.
If you try to run this playbook: subsidized growth, out-of-control cash burn, and a forecast dressed as a strategy…the market won’t forgive you.
Your math is different.
We know Anthropic is playing a game of circular financing, financialization, and selling a forecast.
The bigger question is: what game are you playing?
If you are hyping a forecast, you are waiting on a miracle.
And the market isn’t in the business of handing out miracles.