Anthropic would like you to believe it is a company on the verge of a triumphant public debut. The headlines write themselves: $16.3 billion in half-year revenue, a jaw-dropping 1,258% year-over-year growth rate, 85.4% gross margins, and a projected $1.5 trillion to $2 trillion valuation that would make it one of the most valuable companies on Earth overnight. The Claude maker is, by any surface-level metric, the hottest story in tech.

But dig into the 300-page Form S-1 that Anthropic filed with the SEC on Monday, past the glossy growth figures and the carefully worded risk factors, and a far more uncomfortable picture emerges. This is not a company choosing to go public from a position of strength. This is a company that has run out of private options, is staring down the barrel of the largest corporate rental obligation in tech history, and needs public market cash to avoid a slow-motion collapse under the weight of its own commitments.

The most revealing number in the entire filing is not the revenue. It is $430 billion — the total value of Anthropic's non-cancelable computing and infrastructure commitments, spanning up to 20 years. These are contractual obligations to Broadcom ($161 billion), Google ($115 billion), Amazon ($110 billion), and Microsoft ($45 billion). Anthropic does not own a single data center, a single chip, or a single cooling unit. It has signed the largest corporate rental lease in the history of technology, and the landlord always gets paid.

The Math That's Not Mathing

Anthropic's annualized revenue run rate sits at roughly $32.6 billion. Its annualized compute obligation, assuming a straight-line distribution of the $430 billion commitment, comes to approximately $21.5 billion per year. That means infrastructure alone devours 66% of every dollar that comes through the door before a single engineer is paid, a single marketing campaign is launched, or a single legal bill is settled. The margin for error is razor-thin.

And those payments are not spread evenly. The S-1 discloses a back-loaded schedule, meaning the annual bills start relatively modest but balloon dramatically by late 2027 and 2028. Anthropic is not facing a steady drain — it is facing a payment cliff.

• $430 billion in non-cancelable compute commitments over up to 20 years — an average of roughly $21.5 billion per year, with payments back-loaded to spike in 2027–2028.
• 66% of current annualized revenue (~$32.6 billion) consumed by infrastructure obligations alone, before payroll, R&D, marketing, or legal costs.
• $112.4 billion in cash on hand — a buffer that sounds enormous until you realize it could evaporate in under five years if revenue growth flatlines while the payment schedule accelerates.
• 20–25% of total revenue concentrated in just two unnamed corporate clients, with no long-term contracts to keep them from walking away.

Anthropic reported positive operating cash flow of $3.93 billion in the first half of 2026. It generated $974 million in operating profit in Q2 alone. But those figures describe a company at the peak of its growth momentum, before the back-loaded compute bills arrive in force. They do not describe a company that can comfortably absorb a $21.5 billion annual fixed cost if the AI market cools, if token prices collapse under competitive pressure, or if its two biggest customers — widely reported to be enterprise platforms running Claude through coding tools like Cursor and GitHub Copilot — decide to switch to OpenAI or build their own internal models.

The Customer Concentration Death Trap

Anthropic's $16.3 billion in half-year revenue is not a broad, diversified consumer safety net. It is a pipeline balanced on a knife's edge. Just two unnamed corporate clients account for roughly 20% to 25% of total revenue — somewhere between $3.2 billion and $4 billion. The top five clients bring in a full quarter of the company's entire revenue pipeline.

Worse, Anthropic's revenue is consumption-based — pay-per-token API billing — not locked into multi-year contracts. If those mega-clients decide to switch providers or route their workloads to internally hosted open-source models, a massive chunk of Anthropic's revenue could vanish with the flip of a software switch. There is no contractual safety net. There is no churn protection. There is only the hope that the biggest customers in the AI economy keep choosing Claude over its rivals.

Amazon and Google Win Twice

Anthropic's largest investors — Amazon and Google — are also its largest infrastructure landlords and its primary distribution marketplaces. Nearly half of Anthropic's sales (47–48%) flow through Amazon and Google Cloud marketplaces. Anthropic records the full sales amount as top-line revenue, then pays its landlords roughly a 16% channel distribution fee as a sales and marketing expense. The money flows in a circle, and the tech giants always take their cut.

If Anthropic thrives, Amazon and Google win twice — through equity appreciation and through massive guaranteed cloud revenues. If Anthropic fails, Amazon and Google still own the $430 billion physical data center footprint, which they can immediately rent out to thousands of other companies. The public retail investors who buy Class A shares in the IPO, however, would be left holding an empty software shell.

The IPO Is a Lifeline

The obvious question is why Anthropic didn't simply raise another private round from Amazon or Google. The answer is that it couldn't. Regulatory bodies in the U.S. and U.K. have aggressively cracked down on Big Tech exercising control over AI startups. Amazon and Google legally could not inject tens of billions more into Anthropic without triggering antitrust investigations. And both companies have their own public shareholders to worry about — they cannot keep diluting their own earnings to bankroll a single startup's hardware bills while spending $50+ billion a year each on their own internal AI infrastructure.

So Anthropic turned to the public markets. The timing is deliberate. By going public now, while trailing growth looks spectacular and AI investor enthusiasm is at its peak, Anthropic can raise an estimated $50 billion to $80 billion in fresh capital — effectively doubling its cash reserves before the back-loaded compute payments hit full force. The IPO is not a celebration of profitability. It is a cash injection timed to land before the bill comes due.

The $430 Billion Obligation Is the Real Debt

It is worth being precise about what the headline-grabbing $80.7 billion net loss actually represents. The vast majority of it — over $78 billion — is a non-cash accounting charge triggered by the rising fair value of convertible bonds as Anthropic's private valuation skyrocketed. It is a paper loss, not a cash burn. The company actually added $3.93 billion in net cash from operations during the first half of 2026.

But that accounting artifact has become a convenient distraction from the real financial threat. The $80.7 billion loss is not the problem. The problem is the $430 billion in non-cancelable commitments that do not appear on the balance sheet at all, because accounting rules classify them as operational obligations rather than debt. A retail investor looking only at the core balance sheet would see a pristine company with zero traditional debt and $112.4 billion in cash. The danger is buried in the footnotes.

The Verdict: Anthropic Is Structurally Broke

Anthropic is not a company in crisis today. It has real revenue, real customers, real gross margins, and real cash flow. But it is a company that has structured itself in a way that leaves almost no room for error. It has locked itself into the largest rental obligation in tech history, concentrated a quarter of its revenue in a handful of clients with no long-term contracts, and built a circular financial relationship with the very tech giants that own its physical infrastructure and control its distribution channels.

The IPO gives Anthropic a window — perhaps two to three years — to convert its explosive growth into a sustainable business before the back-loaded compute payments arrive in force. If revenue keeps compounding at triple-digit rates and the customer base broadens beyond a handful of mega-spenders, Anthropic can survive and even thrive. If growth slows, if token prices collapse, if OpenAI undercuts it, or if those two unnamed mega-clients walk away, the $112.4 billion cash cushion could evaporate in under five years, and the company would be left owing Broadcom, Google, Amazon, and Microsoft $21.5 billion a year with no way to pay.

This is why the IPO matters. Not because Anthropic is celebrating a triumph, but because it is buying time. The public markets are being asked to absorb the risk that private Big Tech investors could no longer legally carry. If the AI boom continues, early investors will look brilliant. If it doesn't, thousands of mutual funds, ETFs, and retail investors will be left holding the bag while Amazon and Google collect their rent checks and keep the data centers.

Anthropic is not broke in the sense of having empty bank accounts. It is broke in the sense that it has spent years signing contracts it cannot cancel, accumulating obligations it cannot easily pay, and building a business that must grow explosively just to stay in place. The IPO is not the finish line. It is the starting gun on the most expensive race in corporate history — and the clock is already ticking.