A more skeptical take on Anthropic's numbers

In yesterday's e-mail, I reviewed AI giant Anthropic's highly anticipated S-1 filing for its IPO and concluded:

Anthropic is a very impressive, market-leading company that's growing faster than perhaps any business of its size in history. It deserves a huge valuation, though the right number is impossible to know.

I always like to consider – and present to my readers – a balanced picture. So today, I'd like to share a more skeptical take from Mia Silverio of Scott Galloway's Prof G Media (for paid subscribers)...

She notes that companies choose when to go public – and Anthropic may have chosen this moment because "its business may never look this good again." For example:

Anthropic's S-1 shows the company at its strongest point. Revenue grew 3x faster than costs in the first half of the year. The firm posted its first operating profit in the second quarter. It holds $112.4 billion of cash against $6.2 billion of debt.

But all of that is changing. Revenue growth is slowing. The big three cloud platforms through which Anthropic gets nearly half of its revenue are charging increasingly onerous fees, which will erode operating margins. And it has signed close to $1 trillion in compute deals, most of which haven't yet reached Anthropic's balance sheet or income statement.

S-1s are legal disclosures, but they're also marketing documents, and every company paints a flattering self-portrait. Anthropic's problem is that it's changing so fast its portrait will be abstract by next year.

Silverio acknowledges that revenue is skyrocketing, and there are two other positive factors:

Its customer base has also become less concentrated. In 2023, one buyer made up 43% of revenue. In the first six months of 2026, the biggest customer accounted for only 11% of sales.

The business is getting more efficient as it grows. In the first half of 2025, compute costs were twice Anthropic's revenue; in the first half of this year, they were 55% of it.

However, she highlights a risk many people are missing:

Nearly half of all its revenue routes through one of three gatekeepers: Amazon's AWS, Google Cloud, or Microsoft's Azure. This dependence has grown over time, from 11% in 2023 to 48% in the first half of 2026...

Anthropic's dependence on the big cloud providers is dangerous for three reasons:

1. Gatekeepers can continue to raise the price of access. Anthropic's filing warns that the fees the gatekeepers charge may continue to fluctuate with "changes in the terms of our channel partner distribution arrangements."

2. They have a reason to increase said fees. Amazon, Google, and Microsoft are Anthropic's gatekeepers but also its competitors. The filing admits that the cloud platforms "simultaneously serve as our investors, customers, distribution partners, and competitors," and warns they may make decisions "that disadvantage our models, platform, and applications relative to their own offerings." In other words, Anthropic is meeting its customers in enemy territory.

3. Changing how customers buy Claude could send them to competing models. If Claude gets harder to access through a client's existing cloud account, they could easily switch to Google, Amazon, or Microsoft's in-house large language models. Anthropic's filing warns of this: that a change to one of these relationships could disrupt "customer access to our models."

Most importantly, she documents the massive contractual commitments Anthropic has signed:

Until this year, Anthropic rented nearly all its computing power from Amazon, Google, and Microsoft, and paid for it as the capacity became available. That's a big part of why its costs look so manageable: It didn't have to pay for data centers up front. But renting from competitors is risky, as the cloud providers may put their own compute needs first.

So Anthropic is changing how it buys compute. It's signing larger, longer cloud contracts, and it's leasing chips and data centers of its own. In total, those deals could cost as much as $1 trillion over the next two decades, and the bulk of that bill hasn't reached the balance sheet or income statement (yet)...

Anthropic spent $8.9 billion on compute and infrastructure in the first half of 2026, but it has committed to about $850 billion more, mostly in contracts it can't cancel. The prospectus doesn't give a year-by-year payment schedule, but several of the largest deals overlap from 2027 to 2029. That suggests Anthropic's compute costs will rise to an estimated $132 billion to $155 billion a year starting next year. Its compute costs may not look this low relative to revenue again for a while.

Lastly, she points out that revenue growth is slowing:

Anthropic's growth is already slowing. Revenue rose 144% from the first quarter to the second and then 45% from the second to the third, the Financial Times reported. Anthropic specifically warned that growth "may decline significantly and more rapidly than we anticipate" and noted that only about $8 billion of its next 12 months of revenue is under contract.

Silverio concludes that the financial picture in Anthropic's S-1 isn't indicative of its future:

Going public in 2026 lets Anthropic sell the asset-light version of itself. This is a company with remarkable growth now, operating leverage now, and low debt levels now.

That is all about to change. Once its lease commitments start, Anthropic's financial statements will reveal what it really is: an overleveraged startup with extraordinarily high costs and without direct access to its customers.

I'm not arguing that it is an inherently bad company, but rather trying to illuminate the fact that investors are being asked to bet on Anthropic given a past that is unrepresentative of its future.

Given these circumstances, there is no time like the present. Anthropic seems to agree.

On one of Scott Galloway's other platforms, Prof G Markets, Ed Elson interviewed Steve Eisman of The Big Short fame, who made similar points about Anthropic's filing.

Lastly, Michael Burry, also of The Big Short fame, raised a number of concerns in a Substack post last night (for paid subscribers).

He started by addressing Anthropic's slowing revenue growth:

Revenue growth during Q1 and Q2, the peak tokenmaxxing quarters, was 132% and 144% sequentially quarter-on-quarter.

Anthropic's third quarter results are not clearly stated, though the filing says revenue was 86 times the December 2024 quarter's $197 million, which puts it near $16.9 billion. The company also told investors $16.8 billion at its October 14 briefing (Bloomberg, FT).

Either way, that is up about 45% sequentially, a slowdown from peak tokenmaxxing...

The slowdown was acute in August and September, when Anthropic revenue grew only 3% to 7% a month...

Burry also highlighted the circular financing with Amazon (AMZN), Microsoft (MSFT), Google parent Alphabet (GOOGL), and Broadcom (AVGO):

These companies are not just compute monoliths to Anthropic. These big hyperscalers, especially Google and Amazon, are Anthropic's lenders, vendors, channel providers, and shareholders...

While some may see all the circularity and connectedness as security on the way up, and that is not wrong, it does leave the entire organism catastrophically vulnerable either on a pause or on the way down.

This is very similar to what happened with vendor financing for the telecom infrastructure/internet buildout in 1998-2002. And it is what happened with the computer leasing in the 1960s. Computers were big. The internet was big, but people keep making the same mistakes with every new thing.

Silverio, Eisman, and Burry are right to be cautious. But everything, of course, depends on Anthropic's valuation.

I think it's going to be huge – at least $1.5 trillion – which will likely inflate the AI bubble even further, though I suspect not for very long.

I've said it before, and I'll say it again: As an investor, you don't have to be a hero. Said another way: You don't have to attend every argument you're invited to.

When Anthropic's stock starts trading roughly a month from now, I don't think it will be a good long – or a good short. And while I believe the AI sector is in a bubble, I'm not shorting or buying puts on any AI stocks.

To be sure, there will be plenty of losers in the sector. But my friend Jeff Brown, founder of Brownstone Research, has identified one of the likely winners...

It's a new company behind an AI "super chip" that's 50 times faster than Nvidia's (NVDA). And this breakthrough could open the opportunity of the decade for investors who get in early – before a pivotal announcement on November 5.

I went on camera with Jeff to share all the details in a special presentation that aired this morning. If you missed it, don't worry – you can watch a replay right here.

Best regards,

Whitney

P.S. I welcome your feedback – send me an e-mail by clicking here.

P.P.S. By all accounts, regardless of your political leanings, the midterm elections on Tuesday are extremely important – so I encourage you to vote. Susan and I did so on the first day of early voting on Saturday: