Anthropic Just Went Public at $2 Trillion. Now Follow the Money.

Well, that escalated quickly.
Anthropic, the company behind Claude, began trading on the Nasdaq under the ticker ANT on November 19, 2026. It raised approximately $81.3 billion in the largest initial public offering in history, opening at $111.50 per share against an $83 IPO price.
That opening price briefly put the company's valuation above $2 trillion.
Two. Trillion. Dollars! For a company founded in 2021.
Normally this is where everyone starts celebrating the incredible success of artificial intelligence, the dawn of a new technological economy, and whatever other breathless headline Wall Street has prepared for the occasion. And sure, it's an extraordinary accomplishment. But I'm considerably more interested in what's happening underneath that valuation. Because the most revealing number in Anthropic's IPO documents isn't $2 trillion.
It's $900 billion.
And once you understand what that number represents, the offering starts looking like something much bigger than a company going public.
On April 21, I Said This Wasn't About Claude
On April 21, 2026, I published an article titled Amazon Just Put $25 Billion on Anthropic. This Isn't About Claude.
The argument was straightforward. Amazon wasn't simply investing in a company building a popular chatbot. It was positioning itself to control a substantial piece of the infrastructure that the next generation of artificial intelligence would depend on.
The investment was connected to an enormous cloud-computing commitment, specialized Trainium chips, and a strategy that would make Amazon both a financial beneficiary and a critical supplier of Anthropic's growth.
I called it infrastructure economics dressed up as an AI investment.
I also suggested that Amazon's capital commitment might reduce Anthropic's immediate pressure to go public.
I got that last part wrong.
On November 19, here we are, watching Anthropic ring the Nasdaq bell.
But the underlying infrastructure argument? The IPO documents make it considerably more interesting.
The $900 Billion Elephant in the Data Center
According to Anthropic's final SEC prospectus, the company has approximately $900.9 billion in contractual purchase and lease commitments, substantially all associated with computing and infrastructure. That isn't a typo.
Nearly nine hundred and one billion dollars.
These are commitments spread across many years. Approximately $451.6 billion is expected to be incurred before December 31, 2030, with substantially all the remainder extending from 2031 through 2045.
But here's the part that matters right now.
Approximately $70.7 billion falls due within the 12 months following September 30, 2026. And it gets worse than that, in a way almost nobody is quoting yet. The prospectus also describes a separate block of additional TPU compute, with no fixed dollar commitment, expected to be delivered between 2028 and 2031. Anthropic estimates the aggregate future payments for that compute at roughly $125 billion to $150 billion, and explicitly notes those obligations are *not* included in the contractual obligations table.
So the real forward picture isn't $900.9 billion. It's north of a trillion.
To put all of this in perspective, Anthropic reported $33.2 billion in revenue during the first nine months of 2026, up from $2.5 billion over the comparable period in 2025. That's growth of 1,206% in a year. This is a company growing at a pace that would have sounded absurd just a couple of years ago. And its infrastructure obligations are growing at an equally extraordinary scale.
Here's the Number That Actually Matters
Now compare two figures from the same document.
Anthropic's remaining performance obligations as of September 30, 2026 were $13.8 billion. That's the contracted revenue sitting in backlog, and the company expects about 75% of it, roughly $10.4 billion, to be recognized over the following 12 months.
Against that, $70.7 billion of infrastructure commitments come due in the same window.
Read those two numbers together, slowly.
On the revenue side, about ten and a half billion dollars of contracted work. On the obligation side, about seventy billion dollars of contracted payments. The gap doesn't close with backlog. It closes only if new consumption arrives at an extraordinary rate, continuously, from customers who haven't signed long-term contracts. That's not a scandal. It's how consumption businesses work.
Anthropic's revenue is overwhelmingly usage-based, and usage is growing faster than almost any enterprise product in history. But it does mean the company is making decade-long fixed commitments against demand that has to keep materializing on its own. That isn't automatically a bad business model. It is an extraordinarily consequential bet. And there's already evidence of what it costs. Anthropic's third-quarter GAAP operating loss was $567 million, and the prospectus attributes a large part of that swing to $5.0 billion of compute procured under short-term commitments, which currently costs significantly more than comparable compute bought under multi-year agreements.
Sit with that. The long-term commitments exist because spot compute is expensive. And the spot compute is expensive because the long-term capacity hasn't come online yet. The company is paying a premium in the present to bridge itself to a future it has already bought.
Now Look at Who's Getting Paid
This is where the story becomes genuinely fascinating. According to the prospectus, Amazon and Google together hold approximately 28% of the company's outstanding capital stock following the offering and concurrent private placement.
Amazon holds about 14.6%.
Google holds about 13.4%.
Those holdings are nonvoting shares, which is important. They're substantial financial interests, not equivalent voting control. But ownership is only one component of these relationships. Anthropic has committed to spending at least $110 billion with Amazon between May 2026 and April 2036. And at least $121.3 billion with Google between April 2026 and July 2033.
Both arrangements carry minimum spending obligations regardless of actual usage. If Anthropic's spend falls short, it pays the difference as a true-up, subject to contractual exceptions. Think about that structure for a moment. Amazon invests in Anthropic. Anthropic commits to buying enormous amounts of computing infrastructure from Amazon. Amazon earns revenue supplying that infrastructure. Amazon also owns a substantial piece of Anthropic, meaning it participates financially if Anthropic's valuation rises. And Amazon can distribute Claude through its cloud ecosystem, earning additional revenue from customers using Anthropic's technology.
Google has a remarkably similar arrangement. It's an investor, a supplier, a distributor, and a competitor. Anthropic says essentially that in its own prospectus, identifying Amazon, Google, and Microsoft as partners occupying multiple roles simultaneously.
These companies aren't just competing in the AI economy. They're financing, supplying, distributing, and investing in different parts of the same economy.
And every one of those relationships can generate a different financial benefit.
The Landlord, the Tenant, and the Investor Are Sometimes the Same Company
Here's an analogy.
Imagine I invest billions of dollars in a business.
That business agrees to rent enormous amounts of property from me for the next decade. I make money from the rent. I potentially make money as an investor. I might make money distributing the business's products. And the business's growth helps justify expanding the infrastructure I own.
That's a pretty attractive arrangement for the landlord.
Now replace property with computing infrastructure. Replace the tenant with Anthropic. Replace the landlord with Amazon or Google. It's an oversimplification of the actual contracts, but it illustrates something important.
The cloud companies can benefit from multiple layers of the same economic activity. And critically, some of those benefits don't require Anthropic itself to achieve spectacular profit margins.
The supplier gets paid for infrastructure. The distributor collects its fees. The investor participates in potential equity gains. Those are different financial mechanisms with different risks and timelines.
Of course, the investor can lose money if the valuation collapses, and cloud providers face enormous costs building all that infrastructure. Nothing here guarantees a profit. But the economic interests look very different depending on where you're standing.
Is This Circular Financing?
That's the question everyone is eventually going to ask. It's worth being precise, because there's a difference between interconnected financing and an artificially manufactured economy.
Anthropic's growth is real in the sense that it reports substantial revenue from customers using its services. Enterprise adoption is expanding. AI coding tools, automation systems, research applications, and business integrations are creating legitimate demand. I use these technologies. I build with them. I see their usefulness firsthand. This isn't a discussion about whether artificial intelligence creates value.
It absolutely can.
The question is how much of the industry's capital expansion is being supported by independently generated customer demand, and how much depends on continued financing and commercial commitments among the same interconnected companies.
Consider the relationships.
An infrastructure company invests in an AI developer. The AI developer agrees to buy infrastructure. The infrastructure company uses those commercial commitments to justify additional investment in facilities, chips, power, and networking. The AI developer's growing revenue and infrastructure access help support a higher valuation. Investors then supply additional capital against expectations of future growth.
And around we go.
That isn't evidence of fraud, nor does it mean the same dollar is being improperly counted as revenue multiple times. These can be perfectly legitimate commercial transactions. But the arrangement creates interdependencies that make the industry's ultimate economic returns harder to evaluate. And when hundreds of billions of dollars are involved, those dependencies matter.
The $2 Trillion Valuation Doesn't Mean $2 Trillion in the Bank
Another distinction worth making. Anthropic's opening market valuation was approximately $2 trillion. That doesn't mean the company raised $2 trillion.
The offering raised $81.34 billion in gross proceeds from shares sold, including a portion sold by existing stockholders rather than by Anthropic itself. A concurrent $26 billion private placement added another financing component. Google and NVIDIA each committed approximately $10 billion, AMD about $2.5 billion, and Broadcom, Microsoft, and Micron roughly $1 billion each, alongside TSMC.
That's more than $107 billion in combined gross offering and placement value. Which, by itself, is extraordinary. But market capitalization is the market's valuation of the company's equity, not cash available to pay its bills. And the financial statements still matter. Anthropic reported approximately $17 billion in revenue for the third quarter of 2026.
It also reported a $567 million operating loss for that quarter under generally accepted accounting principles, against positive adjusted operating income of $2.5 billion.
That adjusted figure is worth a second look, because it's thinning. Adjusted operating margin fell from 19% in the second quarter to 15% in the third, and the prospectus points squarely at the short-term compute bill. For fairness, the nine-month picture is improving on the bottom line: the operating loss narrowed to $1.3 billion from $5.2 billion over the comparable period in 2025. But investors are valuing this company substantially on what they expect it to become.
Future demand.
Future efficiencies.
Future margins.
Future profits.
And those expectations now have to support one of the most expensive infrastructure expansion cycles in technological history.
Here's Another Fascinating Detail: Who Actually Controls Anthropic?
There's another important distinction buried in all this. Anthropic may now be publicly traded, but that doesn't mean public shareholders control the company in the ordinary sense.
Its founders retain a special Class F share structure providing 50.1% of the company's general stockholder voting power through a founder-controlled entity. The Class F shares carry no economic rights, only votes.
Amazon and Google's substantial holdings are nonvoting. So the financial ecosystem looks something like this.
The founders retain voting control. Strategic investors hold enormous economic interests. Infrastructure partners receive massive commercial commitments. Public shareholders purchase exposure to Anthropic's future performance. And the ultimate customers have to generate enough demand to make the economics work.
That's a remarkably complex arrangement for a company most people still associate primarily with a chatbot. It also introduces an interesting new tension. Anthropic has consistently positioned itself as a company particularly concerned with AI safety and responsible development. Now it must navigate those commitments while being subjected to the expectations, scrutiny, and volatility of public capital markets. That doesn't mean safety automatically loses. It does mean the financial incentives are becoming much more visible.
The Question Isn't Whether AI Will Be Huge
It will be.
At least, I believe the evidence overwhelmingly supports that direction.
We're already seeing AI transform software development, research, education, content production, business processes, and an expanding list of things that barely existed as practical applications three years ago.
The technology is genuinely consequential. But technological importance and financial sustainability aren't the same thing.
The internet was transformative. That didn't mean every internet company in 1999 was worth its valuation. Railroads transformed economies. That didn't mean every railroad investment was profitable. Entire industries can be economically revolutionary while individual investors, suppliers, and companies experience wildly different outcomes. And that's the distinction I think we're going to spend the next several years confronting.
Who owns the computing infrastructure?
Who controls distribution?
Who captures the operating margins?
Who carries the long-term financial obligations?
And, perhaps most importantly, who supplies the money that ultimately sustains the entire system?
Follow the Money. Then Follow It Again.
On April 21, I argued that Amazon's Anthropic investment wasn't primarily about Claude. It was about securing strategic control over the infrastructure layer of the AI economy.
As of November 19, that infrastructure story is a publicly measurable financial experiment. We have a company that briefly commanded a valuation exceeding $2 trillion on its first day of trading. We have approximately $900.9 billion in contractual commitments, plus another $125 billion to $150 billion of expected compute that isn't even in that number.
We have cloud providers simultaneously acting as investors, suppliers, distributors, and competitors. And we have $13.8 billion of contracted future revenue against $70.7 billion of near-term obligations.
That last pairing is the whole story. Everything else is commentary. I'm not arguing that Anthropic is a bad company. I'm not arguing that artificial intelligence is a bubble. And I'm certainly not arguing that the technology isn't valuable. I'm arguing that the financial architecture around artificial intelligence has become almost as important to understand as the technology itself. Because the smartest model doesn't necessarily capture the most value. The company selling the infrastructure might. The company controlling customer distribution might. The company financing the entire arrangement might. And the companies making those long-term commitments might discover that owning the future is considerably more expensive than building it.
In April, I asked who owned the terrain that artificial intelligence runs across. As of November 19, that question has an even more interesting companion. Who is going to pay the rent?
And perhaps the more important question.
When all that money finishes circulating through the AI economy, how much actual profit is left, and who gets to keep it?
That's the number I'll be watching. Not ANT's share price.
Sources and Technical References
Fortune, Amazon and Google's Anthropic Holdings and Infrastructure Relationships (November 19, 2026)
Rich Washburn is a technologist, strategist, and Founder & Chief AI Architect of ARIA AI Labs, working at the intersection of AI, infrastructure, communications, and capital. He also serves as Managing Partner and Chief AI Officer at Eliakim Capital.





Comments