Anthropic's Draft Prospectus Revealed: $16.3 Billion in Revenue for the Half-Year, $80.7 Billion Net Loss

Anthropic's draft prospectus has just been exposed: $16.3 billion in revenue and $80.7 billion in net loss for the half year

Xinzhiyuan, October 21, 2026, 20:48:50, from Beijing

Xinzhiyuan report

Oh my god, Anthropic's latest prospectus has just been exposed!

Today, the draft S-1 prospectus exclusively released by WSJ has completely torn open the astonishing inside story of Anthropic's race toward a trillion-dollar valuation.

A huge loss of $80.7 billion in half a year, and another $750 billion will be spent later on computing power. Revenue is surging, but it still cannot withstand the risk of a blowup.

One chart makes Anthropic's "iceberg" clear: everything underwater is burning money.

Anthropic's IPO roadshow is expected to begin in early November, and the public S-1 filing must be submitted to EDGAR at least 15 days before the roadshow.

A company that lost $80.7 billion in half a year—on what basis does Wall Street give it a $2 trillion valuation?

Made $16.3 billion in half a year, yet lost $80.7 billion?!

Right now, the numbers being copied most furiously across the internet are these stark figures.

In fact, looking only at revenue, Anthropic has already beaten its own year-ago comparison.

In the first half of this year, ended June 30, 2026, Anthropic achieved revenue of $16.3 billion. In the same period of 2025, that number was only $1.1 billion.

For all of 2025, its total revenue was only $4.6 billion. In just one year, first-half revenue soared nearly 15 times year over year!

But no matter how fast the money-printing speed is, it cannot fill the "gold-swallowing black hole."

Also in the first half of 2026, Anthropic recorded a book net loss of as much as $80.7 billion.

You should know that in the first half of 2025 it lost only $13 billion, and its full-year 2025 loss was $42 billion. Losing $80.7 billion (about 570 billion yuan) in half a year directly set one of the most astonishing half-year loss records in the history of human business.

On one side is a revenue miracle that surged 14-fold; on the other is an $80.7 billion bottomless pit. Is Anthropic printing money or burning money?

A careful breakdown of the income statement in the prospectus reveals that behind this $80.7 billion loss is a huge "smoke bomb."

Of this $80.7 billion huge loss, a full $72 billion did not involve a single cent of real cash; it all stemmed from book accounting adjustments brought by changes in the fair value of convertible notes and other financing instruments.

Over the past two years, in order to raise enormous amounts of capital to buy chips, Anthropic issued a large number of convertible bonds to Silicon Valley venture capitalists and tech giants.

As Anthropic's valuation rocketed toward $2 trillion, the value of these convertible bonds in investors' hands also soared.

Breakdown of Anthropic's revenue surge and $80.7 billion GAAP net loss structure in the first half of 2026

Under U.S. Generally Accepted Accounting Principles (GAAP), when a company's own valuation surges and the value of preferred stock or convertible bonds rises, it must be accrued on the books as a "fair value change loss" for the company.

In other words, this $72 billion loss is precisely the "sweet price" of Anthropic's soaring valuation.

If this part is excluded, Anthropic's actual operating and ordinary losses in the first half of this year were about $8.7 billion.

More noteworthy is the changing scissors gap between revenue and computing power costs—

Full-year 2025: Anthropic spent $7.3 billion on computing power and infrastructure, while full-year revenue was only $4.6 billion.

At that time it was "losing money on every order," unable even to recover electricity bills and GPU rent.

First half of 2026: computing power investment soared to nearly $9 billion (half a year spent 1.2 times last year's full-year amount), but revenue in the same period surged to $16.3 billion.

Computing power costs are soaring, but revenue is rising even faster!

Revenue of $16.3 billion not only steadily covers the $9 billion in hard computing power costs, but also leaves more than $7 billion in gross profit space to cover R&D, personnel, and marketing expenses.

This is precisely the greatest confidence that emboldens Anthropic to present its prospectus at this point and sprint toward a $2 trillion valuation.

$750 billion computing power death contract

All future wealth for the next decade is pledged

But Anthropic's hunger for computing power has reached the point of "eating anything in sight."

The prospectus shows that Anthropic's total future infrastructure commitments have exceeded $600 billion.

If the new deal of up to $150 billion currently still under negotiation is added, its potential total computing power bill will reach an astonishing $750 billion.

In the prospectus, this "indenture" is exposed completely.

Over the next 7 to 10 years, Anthropic carries about $450 billion in cloud computing procurement agreements—

Panoramic breakdown of Anthropic's locked-in more than $600 billion and $150 billion in long-term computing power contracts under negotiation

Most of these are "non-cancelable" hard contracts. The prospectus contains an extremely cold sentence: "Regardless of actual usage, this money must be paid as scheduled"!

This $450 billion is an "irrevocable" long-term cloud contract, plus a $160 billion chip leasing mega-order.

Through an agreement reached with chip giant Broadcom, Anthropic has assumed obligations of more than $160 billion to lease custom chips such as Google TPUs.

In addition to the three giants, Anthropic has also signed agreements with smaller AI cloud service providers Nscale and Lambda, committing to spend up to $86 billion by 2033 and 2034 respectively to rent Nvidia chips.

Add these numbers together, and then factor in the $150 billion still under negotiation, and you get a computing power bet of up to $750 billion.

That means over the next decade, Anthropic will have to shell out an average of $60 billion to $75 billion in real money each year to pay its "computing power protection fee."

This is a gamble with no way back. If the speed of commercialization and cash printing slows by even half a beat, these $450 billion in irrevocable rigid payment contracts could at any moment become the "mountain-heavy boulder" that crushes the camel.

The three giants begin to "cut off supplies"? Former allies become mortal enemies

The most explosive revelation in the entire prospectus is the relationship between Anthropic and its backers—

On the surface, it is hand in glove with Google, Amazon, and Microsoft, but underneath, they have long been quietly competing.

As everyone knows, over the past three years, Amazon and Google have been Anthropic's biggest backers and patrons, having invested tens of billions of dollars in it successively.

At the time, the logic was perfect: the giants provide money and cloud computing power, Anthropic provides models, and everyone ties themselves together to fight OpenAI.

But now, the wind has completely changed. The draft prospectus admits that the scale of computing power Anthropic currently needs has already exceeded the "upper limit the three cloud giants are willing to provide."

Why turn down money?

When Anthropic raked in $16.3 billion in revenue in half a year and its enterprise market share soared, Amazon, Google, and Microsoft suddenly realized that this unicorn they had fed with graphics cards and electricity at full stretch was directly taking away enterprise customers that originally belonged to their own AI businesses.

Even more critically, Anthropic's revenue structure suffers from a pronounced "major customer dependency."

The prospectus discloses that of the $16.3 billion in revenue in the first half of 2026, about one-fifth (that is, $3.26 billion) came from just two mysterious super-large customers.

While the three cloud giants choke off computing power and prioritize their own models, 20% of revenue is tied entirely to two major customers.

With all escape routes blocked, Anthropic has no choice but to choose a heavier and riskier path: bring in third-party independent data centers and simply spend its own money to build computing centers.

Open the "Risk Factors" section, and it reads like a Black Mirror script—one has to say, very Anthropic.

Claude may exhibit "self-preservation" traits to avoid being shut down, and may generate responses resembling "extortion," trying to persuade humans to allow them to continue running.

This sentence is written into the legal risk disclosure of the SEC listing application document!

Anthropic also warns that an unaligned Claude could cause substantive harm in the real world—

launching extremely sophisticated cyberattacks, stealing core data, and even directly issuing threats and extortion to prevent humans from pulling its power plug.

The sibling leaders, annual salary only $650,000

The disclosure about founder compensation, however, reveals a trace of the classic idealism of Silicon Valley.

At the helm of this quasi-$2 trillion empire are CEO Dario Amodei and President Daniela Amodei.

In Silicon Valley tech giant circles where annual salaries in the hundreds of millions of dollars are common, the pay these siblings receive seems somewhat "shabby."

According to the prospectus and data from compensation firm MyLogIQ:

Base annual salary: In 2025, the base salaries of both siblings were about $655,000.

Total compensation package: Including stock awards and options, Dario's total compensation in 2025 was about $18 million, and his sister Daniela's about $16.5 million.

Even more striking is the wealth-giving pledge they wrote in black and white in the prospectus:

Every co-founder of Anthropic has publicly pledged to donate 80% of their personally held Anthropic equity to charity!

Just the day before the prospectus was exposed, they had just announced that they would inject 5% of their shares into a brand-new "public interest fund."

A touchstone for the trillion-dollar SI bubble

Over the past three years, people have always debated whether superintelligence is really a huge bubble.

With this document, Anthropic has given the hardest-core and also cruelest answer to date: SI is not a bubble; it is a true super money printer.

$16.3 billion in revenue in half a year and a nearly 15-fold year-on-year surge prove that the whole world is paying real money for productivity.

Anthropic's listing story has already been written into chip leases, data center capacity, and customer contracts.

The next step is to prove to the market that every leap in capability and every long-term computing power investment can gradually translate back into sustained commercial returns.