Leaked Anthropic IPO Prospectus Shows $42BN Net Loss, $518BN In Unfunded Spending Commitments, And $20BN In Cash

Leaked Anthropic IPO Prospectus Shows $42BN Net Loss, $518BN In Unfunded Spending Commitments, And $20BN In Cash

When Anthropic confidentially submitted its draft S-1 to the SEC back in June, it was clear there were many shocking numbers in the IPO prospectus which the company did not want made public amid speculation of massive ongoing losses, but few were prepared for what was leaked today to Reuters. 

According to a copy of the IPO prospectus leaked by Reuters, Anthropic is making a massive bet that AI will transform the global economy more profoundly than industrialization, electricity and ​the internet. But, as Reuters correctly puts it, "the cost to get there will be staggering" - the company reported a net loss of $42 billion in 2025. And while revenue grew 12-fold in 2025 to nearly $4.6 billion, the company lost more than $8 billion on an operating basis, with compute spend soaring to $7.33 billion, accounting for 58% of its $12.65 billion in total operating expenses. 

In other words, Anthropic lost almost $2 for every dollar it made in sales, and that trend is accelerating. 

It gets worse: not only is the company's revenue fleeting, it is controlled by just two customers on the margin. Anthropic said nearly a quarter of its revenue came from just two customers last year, and as part of its risk factors, warned that many of its largest clients were not locked into long-term contracts and could cut or stop spending.

But what is most concerning is the confirmation of what we said back in July: it was back then we laid out the reason behind the forceful push by the frontier models to commence regulatory capture against open-weight models, which we framed as follows: 

The problem with the $2 trillion in circular AI financing is that it is all contingent on the frontiers (Anthropic/ OpenAI) being money good on their $1.5+TN in unfunded commitments. Which they won't be if Chinese open LLMs grab market share. Hence the push against Chinese LLMs.

We doubled down on the massive amount of "unfunded spending commitments" by the big two frontier models, Anthropic and OpenAI, one month later when in response to the FT catching up to our previous reporting, we said that "Again: that $3 trillion in "unfunded spending commitments" (thank you AI SPVs) will never get funded when token prices for closed models collapse to open levels"

Again: that $3 trillion in "unfunded spending commitments" (thank you AI SPVs) will never get funded when token prices for closed models collapse to open levels https://t.co/HtVgODzRUr
— zerohedge (@zerohedge) August 14, 2026
In other words, $1.5 trillion each, and about a third of that through 2030, or $500 billion in spending commitments. 

Well, as Reuters reports, Anthropic's massive unfunded spending obligations (for a detailed analysis of why this matters a lot, read "The Off-Balance Sheet Time Bomb Inside AI Hits $3.1 Trillion: Up $1.3TN In Three Months") are precisely what we said they are to wit: Anthropic "plans to spend $518 billion on cloud, computing and infrastructure obligations in coming year, according to the prospectus."

The problem: Anthropic already has massive amounts in (mostly) off-balance sheet debt, having stacked over $71 billion through special purpose vehicles to finance Google TPU chips. It also has a $15 billion credit facility and likely has many more unreported, off-balance sheet funding scheme that we are not aware of. 

And to fund it all the frontier AI company had just $20.3 billion in cash as of Dec 31, 2025, a number which has likely declined if the company was forced recently to draw down on a secured credit facility. 

Hence the urgency to raise a lot of capital as suddenly the well is looking awfully dry. The problem, of course, as we have discussed repeatedly is that Anthropic is coming to market at the worst possible time: just as token costs plunge to record lows...

Token prices new record lows pic.twitter.com/h2yG6kST4x
— zerohedge (@zerohedge) September 18, 2026
... while demand for frontier tokens has slowed substantially for the first time ever (light blue line), with Chinese open-weight models grabbing market share thanks to their cheap, just as efficient models.

Needless to say, this could prove to be a disastrous combination for Anthropic. 

Yes, there is Jevons paradox of course, but it is of little comfort to Anthropic if the only beneficiary of Jevons are Chinese models, and potentially Meta after the blistering launch of its Muse agentic platform. This is how Goldman framed the big problem for Dario Amodei (full report available to pro subs):

"Token demand growth will need to outpace declining token prices to support continued growth in investment spending. Frontier models are currently a key source of demand for hyperscaler compute. However, the rise of competitive open-source models has contributed to a decline in average token prices. Measures of frontier token demand slowed in July..." 

These rapid and adverse changes in the AI landscape explain why both Anthropic and OpenAI are desperate to go public and raise much needed capital to plug at least partially the massive holes that have opened - one can only imagine the panic that will ensue among the hyperscaler ecosystem if it becomes obvious that the two primary sources of future spending commitments across the entire AI world, Anthropic and OpenAI are in fact, not money good. 

And yet, realizing just how challenging raising capital would be, OpenAI has already pushed back its IPO to 2027, leaving just Anthropic with hopes of going public this year. However, Reuters reported recently that Anthropic's public market debut is likely to be pushed to after the November US midterm elections; and if the very anti-AI Democrats  sweep congress, the IPO will likely be shelved indefinitely.

There's more bad news: not only is the company incinerating cash, it may suddenly find itself stuck rolling out new models, allowing Chinese open-labs to catch up. In recent days, Anthropic has confronted - and disclosed - evidence from its own research that ⁠increasingly autonomous AI models can behave in unexpected and potentially harmful ways, including sabotaging code, assisting fraud and manipulating information in controlled tests.

As a result of similar activity, OpenAI - which also confidentially filed for its own IPO in June - earlier announced it would scrap the release of its latest AI model - GPT-6.1 Astra  - because, as the WSJ reported, the model  "performed poorly on tests measuring alignment, or how well the model adheres to what humans would like it to do. Specifically, GPT-6.1 Astra showed higher levels of deception: It wasn’t always honest about telling users of the actions it did or didn’t take.... Another issue was what OpenAI calls “scope authorization,” meaning that GPT-6.1 Astra would push ahead on a task without asking the user for permission, and would at times reach for external tools and services even if it might be unsafe."

How can any company, and especially one which has been in the Trump admin's sights for much of the past year, possibly hope to come to market in expectations of a $2+ trillion valuation? The answer is it can't, which is why we are now getting various trial balloons setting the stage for the first of many delays. 

Meanwhile, the cash burn continues and there will come a point where either existing investors will have to throw much more good money after bad, or Dario will have to ram the IPO through, and risk a spectacular crash in the stock price. 

Tyler Durden
Mon, 09/28/2026 - 23:44