Weird article. “Our annualized revenue as of March 2026 is $19b” is not “telling the public we’ve made $19b.” And the discrepancy the author points at is actually more like revenue to date “exceeds $5b” vs revenue to date looking like $6.7b. I hate to tell you, but $6.7b exceeds $5b.
If you look at the table, the implied revenue for March 2026 was $1.58b. So isn't this just a case of the $5b number being from one month earlier than the $6.66b number? Ed dismisses this, but it seems to be the obvious answer - the CFO quote is from March 9, 2026, so 70% of the March 2026 revenue presumably had not yet been earnt. If you subtract that out (or even the whole month, which would also make sense), it checks out: you get something between $5.08b and $5.54b, reasonably describable as "exceeding $5b".
Ah, OK. It still seems reasonable they might report the number in the court affadavit with one month lag for various reasons. The root cause in the discrepancy is just that Anthropic claims to be (and appears to be) on a ridiculous tear, with +35% MoM growth. OP and Ed both seem to dismiss this as impossible, but it seems to align with Anthropic's recent desperate search for more capacity.
I expect that no figure ever tended to the courts this way is actually correct literally to the date of filing. It's whatever numbers were in the most recent accounts when it was signed off.
If the last number you saw was ~$5b as of last month, claiming revenue "exceeding $5 billion to date" doesn't seem unreasonable - it's still true.
Could've possibly made a stronger claim, but the affidavit may have been drafted/written before February's numbers were determined, even if those numbers were in by the time it'd been reviewed by their legal team and submitted.
Zitron’s narrative around AI revenues is that somehow the people Anthropic/OpenAI etc are pitching to, people who meet the sophisticated or wholesale investor tests (i.e., the only people actually able to trade on this information) do not know what ARR means, how it’s different from revenue, and are unable to read financial statements for themselves. And that he is seemingly the only one with insight, with the partial bits of information he has access to, that can tell them the truth of the situation.
I used to love his stuff. Until he just wouldn’t stop beating this drum so breathlessly. Now I wonder how much I’d suffered Gell-Man amnesia with the other content of his I’d previously enjoyed.
> somehow the people Anthropic/OpenAI etc are pitching to, people who meet the sophisticated or wholesale investor tests (i.e., the only people actually able to trade on this information) do not know what ARR means
This is… not unrealistic. Plenty of such investors have fallen for clear scams.
To be honest I'm pretty sure he is right about "AI players are basically throwing out numbers because the real one are not so shiny (yet?)".
What I'm not really fan about is the communication style and the "trust me bro" approach, which is pretty annoying ... but I guess it's because it's easier to make the view counter go up.
Maybe for people used to deal with financial announcements, but from my ignorant and naive point of view, if I read that claim I think instinctively that they made $19b during the last 12 months, not that they are going to maybe make it if they keep the same rate as the last month for the next 12 months.
The problem is, annualized revenue doesn’t work when your income has a 3 standard of deviation month to month. It is standard for other fields/businesses but these tend to have stable month to month revenue.
It’s not deviating up and down. It’s deviating upward. It is necessarily going to wildly overstate the previous 12 months’ revenue while wildly understating the next 12 months’ revenue. There is no way to describe exponential growth in a single number that doesn’t do this. This is why adults with a brain look at the series.
If anything, it’s a very conservative estimate. Short of a major turn of events it seems very unlikely Anthropic’s revenue growth is going to slow to zero.
It's certainly not a perfect number, but what else are you going to do for a fast growing business? If it does 1B month 1, 5B month 2 and 10B month 3, how does an investor extrapolate the next 12 months? Obviously it's a forecast and will be imperfect without more data. Any potential large investor will be given more financials to help them determine if the 10B was a 1 time event or if it is actually recurring. The harder part is understanding the growth rate.
The be fair to Zitron he claims that enterprise customers are likely paying up front so it won't continue in future months. But now we're into accrual for the future revenues which further complicates the analysis.
Even if an enterprise customer was prepaying, that would only show up on a cash balance statement, not as revenue for the month it was collected in. Yes, this is based on accrual in accounting terms. But because the revenue isn’t recognized immediately, collecting prepayment in February shouldn’t skew ARR reported in March.
Per GAAP you are correct, but many of these ARRs are qualified as non-GAAP. So yes, a real analysis should do what you're saying but are they is the question. Also, if they are buying tokens with no time limit, that will complicate the accrual forecast. This could be why all of the downstream investor companies like Amazon are pushing their employees to tokenmax.
As I understand it: in a sworn affidavit (in federal court no less) the norm is to state the tightest accurate floor. A CFO sitting on $6.7B in recognized revenue writes "exceeding $6 billion" or "approaching $7 billion," not "exceeding $5 billion." That's the basis for the 25–35% overstatement claim. This is about legal/financial language norms. It's not OK to under-represent just the same as it would not be OK to over-represent. But let's pretend this is an unusual circumstance (it's not) and that the understatement was a one time thing (it wasn't either based on past statements). So...
Then you have the speedometer/odometer math. Just December 2025 through early March 2026, at Anthropic's own ARR-implied monthly rates, sums to roughly $4.25B ($750M + $750M + $1.17B + $1.58B). If lifetime revenue is barely "exceeding $5B," that leaves under ~$750M for everything before December 2025. But... Anthropic was already reporting $1B ARR in January 2025. This is the argument and "6.7 > 5" doesn't address it because now we're seeing the CFO play the other side of the coin.
So no matter which way you look at it Anthropic's CFO has lied publicly and in very large numbers. You don't get to spin numbers, especially sworn numbers, in a way that benefits you. The only way that can even remotely work is that you're consistent. This is sloppy and all over the board which implies Anthropic wants to hide something.
> You don't get to spin numbers, especially sworn numbers, in a way that benefits you
If it's "a filing where Anthropic was trying to impress a federal court with its commercial scale", then wouldn't the understatement ("exceeds $5B" rather than "exceeds $6B") be to their detriment?
My assumption would be that the February numbers were not yet determined when Krishna Rao was initially writing the affidavit (say, one week before its submission date, if legal review/revisions take some time), which would make "exceeds $5B" the strongest claim that could be safely made.
> As I understand it: in a sworn affidavit (in federal court no less) the norm is to state the tightest accurate floor.
All three of the numbers in the affidavit are nice round multiples of five billion dollars, so the simpler assumption might be they just rounded looser than you think? He also told the court they'd raised over $60 billion, when the precise number was actually $69.1B (as of Feb 12), so they would appear to have taken the 1 significant figure floor for the numbers in the court filing.
Also, the revenue curve probably looks like well behaved exponential growth when Anthropic plots it out. Given that they're making speculative claims in a fast-changing environment anyway, I personally would probably cope if they included extrapolated growth in their annualised revenue figures. Not sure what the custom is, but as the article notes we're dealing with made-up numbers and wild guesses from the accountant's perspective, so it doesn't sound like there is any particular reason to just multiply the current revenue by 12 without making some basic adjustments for the growth rate they're seeing internally.
Revenue recognition for private companies generally is less precise than for public companies, which in the US are obligated to report under GAAP, which uses a different indicator than annualised revenue so public companies are comparable.
It makes sense to scrutinise Anthropic’s revenue in the lead up to IPO on those grounds; their AR figure simply isn’t comparable to revenue numbers from other firms.
However it doesn’t make sense to be sensational about this - iirc reporting GAAP revenue is a necessary condition of going public so the chickens will come home to roost one way or another.
This article has a point to make, but the framing of the title is extremely misleading. The $5B figure is total revenue, and was framed as a not less than figure. The $19B figure was annualized revenue run rate. A figure that has grown rapidly month over month. These discrepancies are described clearly in the text of the article.
In the current environment of circular money flows “revenue” is a fuzzy number. That’s part of the challenge here. There’s concerning gaps between the money flowing in circles and net-new money entering the ecosystem.
Yep. My employer has invested billions in Anthropic, and now there is a strong executive mandate for everyone (regardless of role) to use Claude for everything. The one metric that is being tracked per employee is how much our Claude usage costs. I'm not in a developer role, so I'm wasting time and money making Claude do useless busy-work just to keep my numbers in a range where I'm not in management's focus. I know I'm not alone in my company, and I know the experience is not unique to my company.
The money people are absolutely desperate to make this whole mess work.
uhhhh isn't the most logical explanation that the "exceeding" number is intentionally vague? Just like literally every other number in the CFO's declaration ("spent over $10 billion"... "raise more than $60 billion").
Helps to look at source documents. Declarations like this are written to be literally true, but not precise enough to reveal more information than they have to.
Weird article. “Our annualized revenue as of March 2026 is $19b” is not “telling the public we’ve made $19b.” And the discrepancy the author points at is actually more like revenue to date “exceeds $5b” vs revenue to date looking like $6.7b. I hate to tell you, but $6.7b exceeds $5b.
"Exceeds $5b" is a weird way to say "5$b +34%" ...
Don't get me wrong, I'm not a big fan of Zitron, but the narrative around revenues and profitability around AI labs are quite misleading.
If you look at the table, the implied revenue for March 2026 was $1.58b. So isn't this just a case of the $5b number being from one month earlier than the $6.66b number? Ed dismisses this, but it seems to be the obvious answer - the CFO quote is from March 9, 2026, so 70% of the March 2026 revenue presumably had not yet been earnt. If you subtract that out (or even the whole month, which would also make sense), it checks out: you get something between $5.08b and $5.54b, reasonably describable as "exceeding $5b".
The $19b ARR is from March 3rd, so it is based on the revenue from February.
Ah, OK. It still seems reasonable they might report the number in the court affadavit with one month lag for various reasons. The root cause in the discrepancy is just that Anthropic claims to be (and appears to be) on a ridiculous tear, with +35% MoM growth. OP and Ed both seem to dismiss this as impossible, but it seems to align with Anthropic's recent desperate search for more capacity.
> they might report the number in the court affadavit with one month lag for various reasons
According to the article the total revenue figure in the affidavit is “to date”, so basically 8th or March.
I expect that no figure ever tended to the courts this way is actually correct literally to the date of filing. It's whatever numbers were in the most recent accounts when it was signed off.
Here[0] the original affidavit, IANAL but I'm pretty sure "up to date" in this context means "8th of March".
[0] https://storage.courtlistener.com/recap/gov.uscourts.cand.46...
If the last number you saw was ~$5b as of last month, claiming revenue "exceeding $5 billion to date" doesn't seem unreasonable - it's still true.
Could've possibly made a stronger claim, but the affidavit may have been drafted/written before February's numbers were determined, even if those numbers were in by the time it'd been reviewed by their legal team and submitted.
Zitron’s narrative around AI revenues is that somehow the people Anthropic/OpenAI etc are pitching to, people who meet the sophisticated or wholesale investor tests (i.e., the only people actually able to trade on this information) do not know what ARR means, how it’s different from revenue, and are unable to read financial statements for themselves. And that he is seemingly the only one with insight, with the partial bits of information he has access to, that can tell them the truth of the situation.
I used to love his stuff. Until he just wouldn’t stop beating this drum so breathlessly. Now I wonder how much I’d suffered Gell-Man amnesia with the other content of his I’d previously enjoyed.
> somehow the people Anthropic/OpenAI etc are pitching to, people who meet the sophisticated or wholesale investor tests (i.e., the only people actually able to trade on this information) do not know what ARR means
This is… not unrealistic. Plenty of such investors have fallen for clear scams.
To be honest I'm pretty sure he is right about "AI players are basically throwing out numbers because the real one are not so shiny (yet?)".
What I'm not really fan about is the communication style and the "trust me bro" approach, which is pretty annoying ... but I guess it's because it's easier to make the view counter go up.
If I were on the stand and therefore liable for perjury, I'd also try to give safe answers with some buffer.
Maybe for people used to deal with financial announcements, but from my ignorant and naive point of view, if I read that claim I think instinctively that they made $19b during the last 12 months, not that they are going to maybe make it if they keep the same rate as the last month for the next 12 months.
"Annualized revenue" is a projection, and is known to be a projection.
Yes but my mother-in-law doesn't understand it, so it's a lie.
The problem is, annualized revenue doesn’t work when your income has a 3 standard of deviation month to month. It is standard for other fields/businesses but these tend to have stable month to month revenue.
It’s not deviating up and down. It’s deviating upward. It is necessarily going to wildly overstate the previous 12 months’ revenue while wildly understating the next 12 months’ revenue. There is no way to describe exponential growth in a single number that doesn’t do this. This is why adults with a brain look at the series.
I would venture that's it's the whole reason they use it. because it doesn't work.
If anything, it’s a very conservative estimate. Short of a major turn of events it seems very unlikely Anthropic’s revenue growth is going to slow to zero.
Interesting. I’m going to start describing my “annualized impact” in my performance self-reviews in terms of all the things I project that I’ll do.
Go for it. Be aware of what happens to your stock when expectations you set do not materialize.
Dunno about that, somehow Tesla stock doing pretty well despite setting a lot of expectations that haven't materialized.
Investors don't want results anyway, they want a dream. Until the IPO which is the final boss stage where you are selling forward looking profits
It's certainly not a perfect number, but what else are you going to do for a fast growing business? If it does 1B month 1, 5B month 2 and 10B month 3, how does an investor extrapolate the next 12 months? Obviously it's a forecast and will be imperfect without more data. Any potential large investor will be given more financials to help them determine if the 10B was a 1 time event or if it is actually recurring. The harder part is understanding the growth rate.
The be fair to Zitron he claims that enterprise customers are likely paying up front so it won't continue in future months. But now we're into accrual for the future revenues which further complicates the analysis.
Even if an enterprise customer was prepaying, that would only show up on a cash balance statement, not as revenue for the month it was collected in. Yes, this is based on accrual in accounting terms. But because the revenue isn’t recognized immediately, collecting prepayment in February shouldn’t skew ARR reported in March.
Per GAAP you are correct, but many of these ARRs are qualified as non-GAAP. So yes, a real analysis should do what you're saying but are they is the question. Also, if they are buying tokens with no time limit, that will complicate the accrual forecast. This could be why all of the downstream investor companies like Amazon are pushing their employees to tokenmax.
Wall Street be like that (and worse). Caveat emptor.
You're missing a very big part of the argument.
As I understand it: in a sworn affidavit (in federal court no less) the norm is to state the tightest accurate floor. A CFO sitting on $6.7B in recognized revenue writes "exceeding $6 billion" or "approaching $7 billion," not "exceeding $5 billion." That's the basis for the 25–35% overstatement claim. This is about legal/financial language norms. It's not OK to under-represent just the same as it would not be OK to over-represent. But let's pretend this is an unusual circumstance (it's not) and that the understatement was a one time thing (it wasn't either based on past statements). So...
Then you have the speedometer/odometer math. Just December 2025 through early March 2026, at Anthropic's own ARR-implied monthly rates, sums to roughly $4.25B ($750M + $750M + $1.17B + $1.58B). If lifetime revenue is barely "exceeding $5B," that leaves under ~$750M for everything before December 2025. But... Anthropic was already reporting $1B ARR in January 2025. This is the argument and "6.7 > 5" doesn't address it because now we're seeing the CFO play the other side of the coin.
So no matter which way you look at it Anthropic's CFO has lied publicly and in very large numbers. You don't get to spin numbers, especially sworn numbers, in a way that benefits you. The only way that can even remotely work is that you're consistent. This is sloppy and all over the board which implies Anthropic wants to hide something.
> You don't get to spin numbers, especially sworn numbers, in a way that benefits you
If it's "a filing where Anthropic was trying to impress a federal court with its commercial scale", then wouldn't the understatement ("exceeds $5B" rather than "exceeds $6B") be to their detriment?
My assumption would be that the February numbers were not yet determined when Krishna Rao was initially writing the affidavit (say, one week before its submission date, if legal review/revisions take some time), which would make "exceeds $5B" the strongest claim that could be safely made.
> As I understand it: in a sworn affidavit (in federal court no less) the norm is to state the tightest accurate floor.
All three of the numbers in the affidavit are nice round multiples of five billion dollars, so the simpler assumption might be they just rounded looser than you think? He also told the court they'd raised over $60 billion, when the precise number was actually $69.1B (as of Feb 12), so they would appear to have taken the 1 significant figure floor for the numbers in the court filing.
Weird is far too generous. It’s a travesty of thinking.
Also, the revenue curve probably looks like well behaved exponential growth when Anthropic plots it out. Given that they're making speculative claims in a fast-changing environment anyway, I personally would probably cope if they included extrapolated growth in their annualised revenue figures. Not sure what the custom is, but as the article notes we're dealing with made-up numbers and wild guesses from the accountant's perspective, so it doesn't sound like there is any particular reason to just multiply the current revenue by 12 without making some basic adjustments for the growth rate they're seeing internally.
Revenue recognition for private companies generally is less precise than for public companies, which in the US are obligated to report under GAAP, which uses a different indicator than annualised revenue so public companies are comparable.
It makes sense to scrutinise Anthropic’s revenue in the lead up to IPO on those grounds; their AR figure simply isn’t comparable to revenue numbers from other firms.
However it doesn’t make sense to be sensational about this - iirc reporting GAAP revenue is a necessary condition of going public so the chickens will come home to roost one way or another.
This article has a point to make, but the framing of the title is extremely misleading. The $5B figure is total revenue, and was framed as a not less than figure. The $19B figure was annualized revenue run rate. A figure that has grown rapidly month over month. These discrepancies are described clearly in the text of the article.
In the current environment of circular money flows “revenue” is a fuzzy number. That’s part of the challenge here. There’s concerning gaps between the money flowing in circles and net-new money entering the ecosystem.
Yep. My employer has invested billions in Anthropic, and now there is a strong executive mandate for everyone (regardless of role) to use Claude for everything. The one metric that is being tracked per employee is how much our Claude usage costs. I'm not in a developer role, so I'm wasting time and money making Claude do useless busy-work just to keep my numbers in a range where I'm not in management's focus. I know I'm not alone in my company, and I know the experience is not unique to my company.
The money people are absolutely desperate to make this whole mess work.
Ai written, needlessly verbose in sections. 2/10 did not finish
uhhhh isn't the most logical explanation that the "exceeding" number is intentionally vague? Just like literally every other number in the CFO's declaration ("spent over $10 billion"... "raise more than $60 billion").
Helps to look at source documents. Declarations like this are written to be literally true, but not precise enough to reveal more information than they have to.
Maybe Claude just hallucinated the answers. :)