> ## Content Index
> Fetch the complete content index at: https://www.gonzocapital.net/llms.txt
> Use this file to discover other available public pages before exploring further.

# The Strange Financial Machinery Behind Anthropic
- URL: https://www.gonzocapital.net/anthropic-ipo-explained/
- Published: 2026-11-04T20:00:00.000Z
- Updated: 2026-11-04T20:00:00.000Z
- Description: Anthropic’s IPO filing reveals a company unlike almost anything public markets have seen before: nearly $900 billion in infrastructure commitments, suppliers who are also investors and competitors, unusual founder voting control, and a trust with the power to elect most of the board.
- Author: Carmelo Giuliano
- Tags: Artificial Intelligence, Economics, Finance, Investing, Startups, Technology, #Essay

*Nearly $900 billion in infrastructure commitments, computers it doesn't necessarily own, investors who are also suppliers, and a trust protecting humanity. Is this the weirdest IPO in Silicon Valley?*

I've read enough prospectuses to recognize the warm perfume of other people's money, but Anthropic's has a distinct undertone of burning insulation. Somewhere in the $895 billion thicket of infrastructure commitments, the familiar software company has grown hooves.

I was really hoping to avoid writing about artificial intelligence again this week. Failed again.

Honestly, it's becoming exhausting. Every few days we get another breakthrough, another trillion-dollar ambition, another billionaire telling us how the machines will transform civilization, or another prediction that nobody will have a job by Christmas 2028.

I like technology, obviously. At least I think I still do.

I've spent much of my adult life working with tech startups and investing in them. But the world is a bit of an odd place, and there are other things I'd like to write about from time to time.

Unfortunately, Anthropic has decided to go public. I think they decided that a long time ago, but I’ve just read their recent IPO filing, and it’s important to talk about it.

**WARNING:**

This may be on the drier side, drier than the Atacama Desert. That’s not to say it won’t be revealing or relevant for those of you who invest, build companies, or simply want to better understand the inner workings of the tech industry.

So why is it of paramount importance right now?

This won’t be a piece about Claude replacing you, taking control of your smart toaster, or becoming sentient. We can worry about those things another day. This IPO filing is a bit less glamorous but arguably just as, if not more, consequential.

So, let’s talk about the money.

Specifically, the extraordinary financial machinery underpinning one of the fastest-growing technology companies in human history, and what it tells us about the increasingly wacky economics of artificial intelligence.

For anyone unfamiliar with IPOs, here’s a quick explanation... 

An initial public offering is when a private company begins selling shares to public investors. 

Instead of ownership being largely restricted to founders, employees, venture capital funds, and other private investors, shares can become available to the public and traded on stock exchanges like Nasdaq or the New York Stock Exchange. 

Ahead of these momentous occasions, the company in question publishes a document called a prospectus, explaining its financial performance, business model, risks, ownership, and the rights investors will eventually receive after trading goes live. 

Think of it as the company's opportunity to tell prospective shareholders how wonderful it is, while also being legally obligated to explain the potential ways things could go horribly wrong.

I [spend enough time around venture capital and company valuations](https://www.gonzocapital.net/confessions-of-a-compromised-capitalist/) to find these documents oddly satisfying to peruse. They offer a glimpse beneath the polished investor presentations and optimistic projections that normally dominate the conversation.

And Anthropic's [November 2, 2026, SEC filing](https://www.sec.gov/Archives/edgar/data/2130357/000162828026068676/anthropicpbc-sx1ax1.htm?ref=gonzocapital.net) is quite a specimen to behold.

![](https://storage.ghost.io/c/f9/a7/f9a788de-665a-4e4b-9aa1-91f8e38e2279/content/images/2026/11/CleanShot-2026-11-04-at-19.07.28@2x.png)

Source: SEC.gov

Anthropic generated around $33.2 billion in revenue during the first three quarters of 2026, compared with a meager $2.5 billion during the same period last year. Its operating losses have narrowed dramatically, and Claude has become one of the most important products for enterprises and governments around the world.

This is an extraordinary achievement, whichever way you cut it. 2026 has been quite good to them.

But as I moved through the filing, five questions kept nagging at me. 

Each one led to another collection of financial disclosures, contractual arrangements, and corporate relationships that made me wonder what on Earth we're preparing to list on the stock market. 

SO, let’s dive into the big questions one needs to tackle to understand (or try to), what exactly is happening here.

## **What Does Anthropic Actually Own?**

Let's begin with the computers.

When most people imagine a software company, they picture engineers laboriously writing code while consuming unhealthy amounts of energy drinks. They imagine customers paying those dreaded subscription fees, and a product that can be seamlessly distributed to millions of people without ever having to set up an assembly line or build a factory.

That's part of what historically made software companies so attractive to investors. Build something valuable, distribute it widely, and hopefully enjoy growing profits without constantly investing enormous amounts in physical infrastructure. In fact, [I covered why AI is currently waging war on that old software model](https://www.gonzocapital.net/the-great-digital-jailbreak/) a few weeks ago if you’re curious about the death of software giants.

Well, Anthropic doesn't fit that model very well at all. It’s a beast of its own. Something new entirely.

Producing Claude requires colossal amounts of computing power, specialized chips, insane amounts of electricity, networking equipment, hyperscale data centers, and cooling infrastructure, among other things. 

We’re talking tremendous amounts of operational expenditure.

For the first nine months of 2026, Anthropic reported approximately $18.8 billion in compute and infrastructure expenses. That's roughly 57% of its revenue eaten up right off the bat.

To its credit, this represents a significant improvement over the previous year, when those costs far exceeded any revenue there was to speak of. Anthropic is growing at a pace that is beginning to make its infrastructure expenses more manageable relative to sales. For now.

Historically, Anthropic obtained substantially all its computing infrastructure from a handful of cloud providers. Rather than owning its entire physical computing operation, it largely paid companies like Amazon and Google for access to theirs.

That uneasy arrangement between bitter enemies has now evolved.

Anthropic has developed a strategy called Anthropic Direct, involving direct leases of AI accelerators, dedicated data center capacity, and facilities operated by third parties on its behalf. It also owns or procures certain equipment.

The company is trying to gain greater control and autonomy over its infrastructure economics without having to own the entire physical infrastructure.

Imagine an airline that leases much of its fleet, pays other companies to maintain the aircraft, rents airport facilities, and contracts for the fuel required to operate them. It can be an extremely successful airline. But understanding its finances requires knowing a great deal about its obligations to all those other businesses.

Anthropic presents to customers like a software company, while somehow in some ways resembling a vast industrial operation underneath.

That distinction is important because investors have historically valued software companies differently from capital-intensive businesses. Now we have some sort of weird centaur on our hands. A grotesque combination of the two.

Two companies generating the same revenue can be worth dramatically different amounts depending on how much money they must spend to produce that revenue and how much capital they need to keep growing.

Which brings me to the next question…

## **What Has Anthropic Promised to Pay For?**

While I was combing through the company's long list of financial disclosures, I encountered an amusingly outlandish number.

**$895.4 billion.**

That's the approximate aggregate spending represented by Anthropic's purchase commitments and lease agreements, overwhelmingly for computing and infrastructure.

If my math is correct, that’s almost… $900 billion.

![](https://storage.ghost.io/c/f9/a7/f9a788de-665a-4e4b-9aa1-91f8e38e2279/content/images/2026/11/image-2.png)

Source: [Prof G Media](https://x.com/edels0n/status/2117697019750010942?ref=gonzocapital.net)

For perspective, Anthropic was founded in 2021\. It's expected to incur about $447.5 billion of its committed spending before the end of 2030\. Most of the remainder stretches between 2031 and 2045.

You read that right…

**2045**!

Approximately $69.8 billion of those commitments is expected to come due during the twelve months following September 30, 2026.

I haven’t got the slightest clue which technological infrastructure we'll require in 2045\. Neither does anybody else, with any reasonable degree of certainty.

Twenty years ago, the iPhone didn't exist, Netflix was mailing scratched-up DVDs, and today's frontier AI models would have sounded like “stuff you read about in science fiction.”

Yet Anthropic has signed agreements that extend nearly two decades into the future.

Now, we need to carefully dissect that number.

The $895.4 billion is not debt, nor is it an invoice payable tomorrow. It represents various contractual commitments with different terms, conditions, and termination rights. 

About $213.8 billion relates to arrangements subject to termination rights or conditions that must be met before they commence. Some of those commitments might never turn payable.

There is also another $150 billion to $175 billion in estimated infrastructure payments associated with arrangements excluded from the headline figure because their dollar amounts aren't yet fixed. These estimates depend on future delivery and other contractual conditions.

$895 billion wasn't quite enough to fit everything in.

Some agreements also contain what are called *take-or-pay* provisions.

These provisions are common in industries like natural gas and electricity. A buyer reserves a certain quantity of something and agrees to pay even if they don't use all the capacity.

Imagine reserving an entire hotel for twenty years because you're convinced you'll have an enormous number of guests. If those guests somehow materialize, congratulations, you've secured the rooms you need. If they don't, the hotel still expects payment under the contract. They don’t care if you haven’t got any friends.

Anthropic has entered into massive agreements that involve obligations to pay for computing capacity regardless of actual utilization.

Certain dedicated power and data center arrangements even include 20-year terms, annual price increases, and uncapped fuel-cost pass-throughs.

Uncapped fuel costs. Imagine that… In a *software* company's IPO filing.

Anthropic is betting that securing enormous computing capacity today will protect it against future shortages and rising prices. Perhaps it will. But securing capacity at a particular price and on a particular schedule is only an advantage if that capacity remains economically useful.

But technology has this adorable little habit of changing. It’s sort of the point.

Hardware is improving, and will continue to do so. New model architectures will emerge. Competition will heat up as more combatants enter the arena. Absolutely nobody can guarantee the exact economics of computing ten or fifteen years from now.

Even weirder…

Approximately 79% of Anthropic's revenue during the first nine months of 2026 was consumption-based. Customers generally pay according to how much intelligence they use. Some enterprise contracts contain minimum spending commitments, but many customers can change their usage or reduce spending *without* corresponding long-term obligations.

Anthropic is locking in enormous portions of its future costs while the future revenue needed to cover them remains uncertain.

The danger is that it could be contractually committed to paying yesterday's prices for tomorrow's infrastructure, while competitors gain access to newer or cheaper technology.

If ever there were a bet that could make the entire city of Las Vegas blush, that would be it.

## **Who Supplies, Finances, and Owns Pieces of Whom?**

Let's talk about another company for a moment. Amazon.

Amazon [provides Anthropic with computing infrastructure](https://www.anthropic.com/news/anthropic-amazon-compute?ref=gonzocapital.net). It distributes Claude through AWS. It invests in Anthropic. It also develops AI products that compete with Anthropic's.

![](https://storage.ghost.io/c/f9/a7/f9a788de-665a-4e4b-9aa1-91f8e38e2279/content/images/2026/11/image.png)

Source: Amazon

Google [provides computing infrastructure](https://www.anthropic.com/news/expanding-our-use-of-google-cloud-tpus-and-services?ref=gonzocapital.net), distributes Claude through its cloud ecosystem, owns Anthropic securities, and competes through Gemini.

![](https://storage.ghost.io/c/f9/a7/f9a788de-665a-4e4b-9aa1-91f8e38e2279/content/images/2026/11/image-1.png)

Source: Google

Microsoft has its own combination of investment, infrastructure, distribution, and competing AI products.

Amazon even invested an additional $5 billion in Anthropic's Series H financing in May 2026\. 

Anthropic separately committed to spend at least $110 billion with Amazon between May 2026 and April 2036, subject to specified contractual exceptions.

During the first nine months of 2026, approximately 44% of Anthropic's revenue came from customers purchasing its services through cloud-provider marketplaces.

The same companies providing critical infrastructure to Anthropic can also invest in its success, receive payments from it, and help distribute the services that produce its revenue.

For those of you who aren’t as caught up in the Silicon Valley soap opera, this is a bit like your landlord investing in your restaurant, financing the kitchen equipment, managing much of your delivery business, and opening a competing restaurant directly across the street.

Everybody has a financial interest in everybody else. It’s a bit incestuous in some ways.

Each arrangement may make perfect commercial sense on its own. Taken together, though, they create dependencies and overlapping financial incentives that deserve a whole lot more scrutiny than the phrase "strategic partnership" normally receives.

If one domino can topple, it doesn’t bode well for the others.

All this definitely makes understanding the actual economics more confusing. 

Chief among the examples of this is SpaceX.SI, formerly xAI. Or whatever the current name is.

Anthropic has agreements to purchase access to substantial computing capacity from a company belonging to the same corporate group as a competing AI developer and bitter nemesis.

Those agreements have a stated maximum value of $120.5 billion through 2029, although they include important cancellation provisions and shouldn't be mistaken for guaranteed spending of that amount.

Still, I find this extraordinary.

We’ve arrived at a point where competing AI businesses are now significant infrastructure customers of one another, while their strategic partners own substantial stakes in the companies consuming and distributing their services. 

Is this all just one megaconglomerate at this point?

The lines between supplier, investor, customer, financier, distributor, and competitor are becoming blurrier than drunken eyesight.

The financial success of these companies is becoming difficult to untangle from the financial success of one another. For investors trying to work out where the risks actually sit, that's a rather serious problem.

## **Who Actually Controls Anthropic After the IPO?**

Now suppose you’ve made it all the way here, and somehow arrived at the conclusion that you’re thinking about deciding you want to buy shares in Anthropic at IPO or beyond...

It could be that you’ve bought in and believe the company will grow, its enormous infrastructure investments will pay off, and Claude will become more and more critical to the global economy as time approaches infinity.

Fine. You'll be able to buy Class A shares when the company lists on Nasdaq under the ticker ANT, assuming the offering proceeds as currently planned.

Those shares carry economic rights and one vote apiece.

But voting rights are not necessarily the same thing as having meaningful control.

Anthropic's founders have established a [special Class F share](https://www.sec.gov/Archives/edgar/data/2130357/000162828026068676/anthropicpbc-sx1ax1.htm?ref=gonzocapital.net), held through a Founder LLC, that carries 50.1% of the company's voting power on ordinary shareholder matters.

This doesn't mean they own 50.1% of the company's economic value. It's a distinction between owning shares and controlling votes.

Founder-controlled public companies aren't unusual. But…

Anthropic also established something called the Long-Term Benefit Trust, or LTBT.

This trust, through a special Class T share, generally has the authority to elect and remove a majority of the company's board of directors.

Its purpose is to help ensure that Anthropic pursues the responsible development of advanced AI for the long-term benefit of humanity.

So the founders control the ordinary shareholder vote, while a separate trust selects most of the board.

You can own a significant financial stake in Anthropic without being able to meaningfully influence who controls the company. If you disagree with the founders or the trust, voting your shares may accomplish very little.

Public shareholders participate financially, but corporate control is deliberately divided among institutions with different responsibilities.

And that's before we get into Anthropic's status as a [Delaware public benefit corporation](https://www.anthropic.com/public-benefit-report?ref=gonzocapital.net).

Under this structure, the company must balance shareholders' financial interests with the interests of people affected by its activities and its stated public-benefit mission.

The prospectus explicitly warns investors that pursuing that mission could reduce revenue, increase expenses, lower profitability, and cause shareholder dilution.

I actually somewhat respect the philosophy behind this. I can even understand why Anthropic would want safeguards against reckless decisions made in pursuit of short-term profits. 

But who decides what constitutes the long-term benefit of humanity, and who holds those people accountable?

The trustees largely appoint their own successors, and ordinary shareholders cannot simply vote them out if they disagree with their decisions. 

There are legal safeguards and some checks on their authority, but that's still an extraordinary amount of influence to place in the hands of a small group of people whom the public investors financing the company have no direct power to replace.

The whole thing strikes me as a rather peculiar interpretation of accountability.

Consider the tension that will exist in this bizarre arrangement.

Anthropic needs enormous amounts of capital to finance its ambitions. It wants public investors to purchase ownership stakes in the company, presumably because they expect financial returns.

At the same time, the company's governance structure deliberately prevents those investors from exercising conventional control and warns them that financial returns may sometimes take a back seat to other priorities.

Anthropic has also contributed shares representing approximately 5% of its fully diluted equity, under the filing's specified assumptions, to a wholly owned subsidiary called the [Public Benefit Fund](https://www.anthropic.com/news/public-benefit-fund?ref=gonzocapital.net). So, to be clear, it hasn't simply handed 5% of the company to an outside charity. 

The fund holds those shares and can eventually sell, transfer, or donate them to finance public-benefit initiatives.

But there's another interesting provision buried in the arrangement. Anthropic's board has approved a framework allowing it to issue additional shares to the fund, potentially representing up to another 2% of the relevant share base each time the stock appreciates 50% beyond its previous milestone. There are conditions attached, and the board retains discretion over whether to authorize these issuances.

That's quite a mechanism!

If Anthropic's stock performs extraordinarily well, the company may issue additional shares to support its philanthropic ambitions, diluting existing shareholders' economic interests. Those shares could eventually be donated, transferred, or sold to third parties.

I can appreciate the intention, but I'd imagine prospective investors will want to understand just how much of their future ownership could be affected.

But it's a remarkable proposition to present to investors:

Please provide capital for our extraordinarily expensive technological project. Our suppliers may also be our investors or competitors, our founders will retain special voting control, an independent trust will choose most of our directors, and our mission may sometimes require decisions that reduce your financial returns.

Perhaps investors will be perfectly comfortable accepting those terms. 

That's their decision.

But disclosure doesn't make an arrangement favorable. Consider that your warning.

## **What Kind of Company Are We Actually Valuing?**

This is the answer I think carries the most weight.

Whenever people discuss a company's valuation, the conversation tends to revolve around a single enormous number.

A company is worth $100 billion. Another is worth $500 billion. Someone suggests a trillion dollars or two trillion dollars, and after the initial shock wears off, we apparently got accustomed to that figure too.

But a valuation isn't a pile of money sitting in a bank account. It's the value investors place on the company's equity, based on what they believe its future economic prospects justify. 

For a public company, market capitalization is calculated by multiplying the share price by the number of outstanding shares. Simple arithmetic.

How that share price is determined is an entirely different matter, occasionally resembling a bizarre ritual in which founders and investors attempt to summon enormous valuations through sheer force of optimism. 

But determining whether that price makes sense requires understanding the business, its likely future profits, financial obligations, dilution risks, and the rights investors actually receive.

Traditionally, one of the simplest ways to assess a company's valuation has been the price-to-earnings ratio, or P/E. If a company earns $1 billion annually and the stock market values its equity at $20 billion, investors are paying twenty times its annual earnings. A P/E of 20\. Whether that's cheap or outrageously expensive depends on the business, its growth prospects, and a host of other considerations.

But what happens when the company isn't profitable? Anthropic has yet to demonstrate sustained net profitability, so a conventional positive P/E ratio isn't particularly useful here. Instead, investors must lean more heavily on revenue multiples, projected future earnings, and increasingly ambitious assumptions about what the business might eventually be worth.

And this is where I think things can start getting a little divorced from reality. We're placing enormous values on companies based on profits that haven't yet materialized, sometimes decades into a future that nobody can predict with any real confidence.

And revenue? That seems to have emerged as a substitute for earnings in half the conversations I hear about AI companies.

Anthropic's $33.2 billion in revenue over nine months is impressive, but that isn't $33.2 billion available to shareholders. It's the money coming into the business before accounting for the enormous costs of producing and delivering its services.

The computers, electricity, engineers, research, financing, and all the other expenses involved in keeping the operation running must be paid for.

What ultimately matters to investors is how much sustainable economic value the company can generate after accounting for those costs and the capital it needs to keep growing.

And this is where the distinction between a successful company and a successful investment transforms into something very, very important. Anthropic could become indispensable to the global economy, grow into one of the largest companies on Earth, and still deliver disappointing returns to people who bought its shares at the wrong price.

That's something that tends to get lost in the frenzy surrounding these valuations.

Anthropic's final IPO price and offering size haven't yet been specified in the amended filing, so we don't yet have a definitive IPO valuation to evaluate. That’ll likely come in the weeks or days approaching the actual listing date.

What we do have is a company unlike most conventional software businesses.

It develops valuable intellectual property and sells intelligence through software. It depends on enormous physical infrastructure investments, some extending decades into the future. Its major suppliers are also investors and distribution partners, and sometimes competitors. Its governance deliberately separates financial ownership from conventional corporate control.

How exactly can we possibly ascribe any sort of accurate value to something like that?

The standard comparisons won’t cut it here.

Anthropic isn't merely a traditional software company. Not even close.

Its enormous infrastructure requirements make it resemble an industrial titan in certain respects. Yet it doesn't own anywhere close to the entire physical system supporting its operations.

It is also a research laboratory, a strategic partner to some of the world's largest technology companies, and a public-benefit institution with unusual protections against ordinary market incentives.

We may just be witnessing the emergence of an entirely new kind of corporation. I'll entertain that possibility. 

But inventing a new category of business doesn't exempt anyone from the old arithmetic. Contracts still have to be paid, hardware still grows obsolete, and shareholders still have to contend with the consequences of decisions they may have very little influence over.

I suspect Anthropic's eventual IPO will be covered like a major sporting event. There'll be breathless commentary about the valuation, the opening share price, the fortunes being made, and what it all supposedly means for the future of artificial intelligence. The usual circus. 

Underneath all that excitement, though, will be the rather dull questions of what this company actually owns, what it owes, and whether the money coming in will ever justify the extraordinary commitments going out.

I have no idea what Anthropic's shares will be worth when they finally begin trading. It’s a fool's errand to pretend to know that sort of thing.

They may turn out to be a bargain. But after reading this prospectus, I'd be extremely cautious about accepting the inevitable argument that a company destined to change the world must also be a wonderful investment.

You could be absolutely right about Claude becoming indispensable, about Anthropic transforming entire industries, even about artificial intelligence fundamentally changing civilization. And still lose a fortune because you paid far too much for a piece of the company that made it happen. I can't think of a more expensive way to be right…