31 Comments
User's avatar
Mattman26's avatar

Great piece, thanks. I wonder if the recent coordinated messaging of “we need to slow down the AI train so as not to destroy humanity” is driven by realization of the points you make here.

rtko's avatar

Exactly

Lukas Terhechte's avatar

Its probably also a try to shield them from liability / copyright claims on the data they used (stole) to train their model

Allan Richard Wasem's avatar

And then there's the need to stop the cheaper Chinese LLMs from blowing the end-user market to Hell and Gone!

NLNW's avatar

There. You've said it.

Michael Lipman's avatar

So many top notch people have said it already. Chanos, Kedrosky, etc. The market monkeys and talking heads ignore it. I think it’s going to take a Scott McNealy moment.

Bob Nixon's avatar

Chris, you continue to provide a strong case for a correction in the market. I believe that continued passive inflows to the S&P is propping up the Mag 7 more than is justified based on current financials.

DLB's avatar

One thought I have had for over a year now is; "how will all this money that is being spent on A.I. show a profit that covers all the investment that is being made?" Using the rule of 10, the investments made should return profits within 10 years to match what has been invested. With the fierce competition that exists in A.I. today, including what exists in China and other places around the globe, I just don't see how everyone makes the necessary returns in the next 10 years on what they have invested today.

Add to that the fact that increased competition eventually leads to lower prices. Then one has to consider how today's product will perform against the new ideas that come out in future, making today's product obsolete. To me it's like investing your money in the old DOS operating system for your computer only to see Windows arrive just a couple of years later.

One other point that bothers me goes back to when the word processors became standard issue on all computers. In the mid 90's I was watching my son's 5th grade class when the subject of spelling came up. The teacher made an announcement that stunned me, "we don't worry about spelling anymore, the computer will correct our mistakes". My question today is will A.I. make the average person less intelligent as he or she now depends on A.I. to answer all their questions instead of studying most any subject?

The Color Of Air's avatar

You ask a lot of pertinent questions. Well done.

Mike's avatar

A modern day Frankenstein experiment.

Petty Rage Machine's avatar

I want you to take what I’m about to say constructively, in the best way possible, because it’s going to sound harsh - but you need harsh because someone might actually invest off of this report: from someone who should know better, this is bad faith at best and complicit puppetry at worst. Both point to an agenda, one worse than the other.

Did you ever once question why the Reuter’s leak ONLY included 2025 financials? This is a question someone that understands S-1 filing rules and journalism should be asking before anything else.

The only reason someone would leak selectively, exclude financials that WOULD HAVE BEEN REQUIRED BY LAW, is to allow for bad faith reports like yours to flourish and to cause a negative spin and a hopeful downturn on this unstoppable market.

“Regulation S‑X Rule 3‑12 (17 CFR 210.3‑12) sets the requirements for how current a registrant’s financial statements must be at the effective date of a registration statement or the mailing date of a proxy statement, including when interim or audited year‑end updates are required. “

None of Reuters' three stories on the draft (financials, risk factors, governance) contains a 2026 financial figure.

• Financials story: everything is FY2025 with 2024 comparisons, including a cash balance as of December 31, plus the $518B in forward commitments.

• Governance story: its pay data comes from a 2025 compensation table.

• The only 2026 items are outside context, such as the $965B valuation from May.

• The one 2026 operating stat, about 6% of research compute going to safety in a sample week in July, came from a separate Anthropic statement this month. Reuters also notes the filing itself didn't disclose safety spending.

So I’ll help you.

Leaked FY 2025 revenue 4.6Bn

Reported Q1 2026 revenue from sourced investor documented. 4.73Bn.

If you skimmed what I just wrote. Read it again.

From Granite:

Anthropic Revenue and Run Rate Growth Explained

“Private valuations follow revenue when growth is this fast. Anthropic's run rate, the annualized figure it projects from recent performance, is what turned a $183 billion company into a near-trillion-dollar one inside a year.”

Date Run rate

End of 2025 (Dec) $9B

Feb 2026 $14B

Mar 2026 $19B

Apr 2026 $30B

Mid-May 2026 $47B

End of July 2026 $65B

I want you to understand that I wasted my time in correcting this atrocity of a story you put up and I did it with a multitude of LLMs and sourced data from all over to the point that this is what Google’s off the cuff response to the post was when I ran it through the filter:

“Anthropic's explosive 2026 revenue run rate signifies a hyper-growth trajectory that renders historical 2025 financials largely obsolete for evaluating its current valuation. When a company's annualized revenue scales from $9 billion to over $65 billion inside of seven months, static, backward-looking reporting fails to capture the actual commercial velocity driving its market value toward $1 trillion. Your observation regarding Regulation S-X Rule 3-12 is correct; leaking exclusively historical 2025 tables selectively isolates a massive capital investment phase while omitting the real-time revenue engine that justifies it.”

The story you should have run was “why would Reuters knowingly and willfully leak an incomplete report?”

Because the real report shows the “Emperor” has a 15 inch pecker and is draped in diamond encrusted gold plate mail.

Attila Rebak's avatar

There may also be an additional pressure on the economics described here: energy costs. As QTR has discussed in previous posts, the AI infrastructure buildout requires enormous amounts of electricity, with some hyperscale projects demanding multiple gigawatts of power and, in some cases, considering dedicated generation because existing grids may not accommodate the load.

At the same time, his recent discussion of the Iran conflict has highlighted the risk of higher oil and broader energy prices. If rapidly growing AI power demand coincides with tighter or more expensive energy supply, the issue is therefore not only whether sufficient electricity can be made available, but at what cost. Higher energy and associated infrastructure costs could put additional pressure on already challenging AI economics and potentially push the point at which these enormous investments generate adequate returns further into the future.

Herschel Savage's avatar

I remember criticizing Bezos for being an idiot, pouring money into AWS for 7 years and losing money every year. Here was a guy who sells books online, what the fuck does he know about IT infrastructure? Not one of my best decisions to divest. And he wasn't even working on technology that might help cure cancer someday. The AI bubble will definitely deflate, but I'm holding for 3 more months.

Dbigkahunna's avatar

Sí. Mañana. Always mañana.

Radicaltroll's avatar

I'll wait to you are proven yet again to be wrong and not say so.

TheSovereignRevolution's avatar

The whole scheme reminds me of Wimpy: I'd gladly take a hamburger today and pay you Tuesday.

In this case, I'd gladly build a data center and pay you someday (maybe).

Steve S's avatar

If the emperor is AI I might agree, but not if Chris refers to the market in general. AI is a tool, our tool, we are not and will not become its tool. The Matrix was a movie, and even in that movie Keanu Reeves comes out ahead and can defeat the strongest AI.

As for individual AI stocks, remains to be seen which will succeed. When Amazon first came to market it was considered to be way over capitalized considering it was a small online bookstore branching out to sell other stuff, with little earnings. Remained that way for several years before becoming profitable.

The 50D MA EPS of the S&P 500 was $11.76 on 9/25/26; $10.20 on 10/25/24; for a 15% gain and steady lower left to upper right graph pattern. From 4/10/26 to present it shows steeper elevation gain, indicating a rise in earnings. Important to note these are earnings, not revenues. SPACEX is not part of the S&P 500, nor with Anthropic be once it becomes publicly available. Given short term blips and dips caused by factors other than earnings, I continue to be in the market.

If the sky is falling I'm well stocked with a raincoat, sturdy roof, and ample provisions and firearms in my basement to keep me going hopefully long enough for the air to clear and sanity to return. In such an occasion, I guarantee earnings of the S&P 500 will be down.

zaporozhe's avatar

CAT's rolled over, URI, TRP, XLU, all rolled over. Looks like some hedging, issues allied to AI are wilting.

Fast money's gonna party to the end, I like to look at the adjacent stuff.

Thomas Hauber's avatar

The only people who will make money from AI will be those involved with the companies picking over the bones of this generation of bankrupt AI companies.

Michael's avatar

Another great piece, thanks Chris

Bob Brown's avatar

This interview is interesting and relevant:

https://www.macrovoices.com/1549-macrovoices-549-matt-barrie-ai-gent-provocateur

He points out that rather than allow the AI data centers access to their data, it is not difficult for them to setup local company AI that is secure and that NVIDIA is already selling into that market. If this continues the data centers will have very low usage.