Hello and welcome to the latest premium edition of Where's Your Ed At, I appreciate any and all of your subscriptions. I work very hard on these, and they help pay for the costs of running Ghost and, well, my time investigating different things. If you're on the fence, subscribe! I promise it's worth it.
I also want to give a HUGE thank you to Westin Lee, a writer who has written about business and the use of AI, who was the originator of the whole "what if we used ARR to work out what these people make?" idea. He's been a tremendous help, and I recommend you check out his work.
If you're an avid reader of the business and tech media, you'd be forgiven for thinking that OpenAI has made (or will make) in excess of $10 billion this year, and Anthropic in excess of $4 billion.
Why? Because both companies have intentionally reported or leaked their "annualized recurring revenue" – a month's revenue multiplied by 12. OpenAI leaked yesterday to The Information that it hit $12 billion in "annual recurring revenue" – suggesting that its July 2025 revenues were around $1 billion. The Information reported on July 1 2025 that Anthropic's annual run rate was $4 billion – meaning that its revenue for the month of June 2025 was around $333 million. Then, yesterday, it reported that the run rate was up to $5 billion.
As a reminder, both of these companies burn billions of dollars – more than $5 billion each in 2024.
These do not, however, mean that their previous months were this high, nor do they mean that they've "made" anything close to these numbers. Annualized recurring revenue is one of the most regularly-abused statistics in the startup world, and can mean everything from "[actual month]x12" to "[30 day period of revenue]x12" and in most cases it's a number that doesn't factor in churn. Some companies even move around the start dates for contracts as a means of gaming this number.
ARR, also, doesn’t factor seasonality of revenue into the calculations. For example, you’d expect ChatGPT to have peaks and troughs that correspond with the academic year, with students cancelling their subscriptions during the summer break. If you use ARR, you’re essentially taking one month and treating it as representative of the entire calendar year, when it isn’t.
Sidenote: I want to make one thing especially obvious. When I described ARR as “one of the most regularly-abused statistics in the startup world,” I meant it. ARR is only really used by startups (and other non-public companies). It’s not considered a GAAP-standard accounting practice, and public companies (those traded on the stock market) generally don’t use it because they have to report actual figures, and so there’s no point. You can’t really obfuscate something that you have to, by law, state publicly and explicitly for all to see with crafty trickery.
These companies are sharing (or leaking) their annualized revenues for a few reasons:
- So that the tech press reports them in a way that makes it sound like they'll make that much in a year.
- So that the tech press reports a number that sounds bigger and better than the monthly amount. For example, calling a startup a "$100 million ARR" company (like vibe-coding platform Lovable) sounds way better than calling them an "$8.3 million a month company," in part because the number is smaller, and in part because, I imagine, it might mislead a reader into believing that's what they've made every month. Yes, saying the ARR figure does that already.
- So that investors will believe the company looks bigger and more successful than it is.
In any case, I want to be clear this is a standard metric in non-public Software-as-a-Service (SaaS) businesses. Nothing is inherently wrong with the metric, save for its use and what's being interpreted from it.
Nevertheless, there has been a lot of reporting on both OpenAI and Anthropic's revenues that has created incredible confusion in the market that benefits both companies, making them seem far more successful than they really are, and giving them credit for revenue they are yet to book.
Before I dive into this — and yes, before the premium break — I want to establish some facts.
OpenAI:
- Raised: $31.9 billion. Other reporting reports $61.9 billion, but it includes $40 billion of its March 2025 round, of which only $10 billion has been wired, and $20 billion of which is contingent on OpenAI's conversion to a for-profit. And that’s without mentioning the other variables in play — like whether Softbank will be able to fulfill their side of the deal. This number is also further muddied by OpenAI raising another $8.3 billion this week - months ahead of when they were meant to - but this is also supposedly part of that $40 billion.
- Valuation: $300 billion.
- Projected 2024 revenues: $4 billion, per The Information, though a September 2024 story from the New York Times suggested that it had predicted $3.7 billion internally. CNBC appears to confirm the "$3.7 billion" number is correct, but this article was also written on September 27 2024, and the year ends December 31 2025.
- Projected 2025 revenues: per Bloomberg, OpenAI projects to make $12.7 billion this year.
Anthropic:
- Raised: $20.7 billion, with current plans to raise as much as $5 billion more.
- Valuation: If the round above closes, $170 billion (I believe it will).
- Projected 2024 revenues: $918 million, per The Information.
- Projected 2025 revenues: per The Information, Anthropic's base case (likeliest, according to their reporting) revenue projection is $2 billion, with the "optimistic" projection being $4 billion.
The intention of either reporting or leaking their annualized revenue numbers was to make you think that OpenAI would hit its projected $12.7 billion revenue number, and Anthropic would hit its "optimistic" $4 billion number, because those "annualized revenue" figures sure seem to have the word "annual" in them.
A sidenote about ARR, and a potential way my analysis is actually too kind: In this analysis I have assumed that OpenAI and Anthropic's revenues have always gone up.
Annualized revenue is a one month snapshot of a business. Though I have no way of proving it — which is why I don't try! — but there is always a chance that one or more of the months I discuss here was lower than the following. If I had to speculate, I’d wager that the summer months — those outside the normal academic calendar — see lower subscription revenue for these companies. Nevertheless, we do not have that information, and thus will not factor it into the analysis. But even one "off" month would be bad for either Anthropic or OpenAI.
I will add that we've never had real reporting about OpenAI or Anthropic's actual total year revenues before, which is why I am doing my best to work it out.
Yet through an historic analysis of reported annual recurring revenue numbers over the past three years, I've found things to be a little less certain. You see, when a company reports their "annual recurring revenue," what they're actually telling you is how much they made in a month, and I've sat down and found every single god damn bit of reporting about these numbers, calculating (based on the compound growth necessary between the months of reported monthly revenue) how much these companies are actually making in cash.
My analysis, while imperfect (as we lack the data for certain months), aligns closely enough with projections that I am avoiding any egregious statements. OpenAI and Anthropic's previous projections were fairly accurate, though as I'll explain in this piece, I believe their new ones are egregious and ridiculous.
More importantly, in all of these stories, there was only one time that these companies shared their revenues — when OpenAI shared its $10 billion runrate in May, though the July $12 billion ARR leak is likely intentional too. In fact, I believe both were an intentional attempt to mislead the general public into believing the company was more successful than it is.
Based on my analysis, OpenAI made around $3.616 billion in revenue in 2024, and so far in 2025 has made, by my calculations, around $5.266 billion in revenue as of the end of July.
This is also a slower growth rate than it’s experienced so far in the year. Going from $5.5 billion in annualized revenue in December 2024 to $10 billion annualized in May 2025 was a compound growth rate of around 12.7%. The "jump" from $10 billion ARR to $12 billion ARR is 9.54%. While I realize this may not seem like a big drop, every single penny counts, and percentage point shifts are worth hundreds of millions (if not billions) of dollars.OpenAI has been projected to make $12.7 billion in revenue in 2025. Making this number will be challenging, and require OpenAI to grow by 14%, every single month, without fail. For OpenAI to hit this number will require it to make nearly $2 billion a month in revenue by the end of the year to account for the disparity with the earlier months in the year when it made far, far less.
How Exactly Is OpenAI Calculating Annualized Revenue?
I also have serious suspicions about how much OpenAI actually made in May, June and July 2025.
OpenAI roughly doubled its revenue in the first seven months of the year, reaching $12 billion in annualized revenue, according to a person who spoke to OpenAI executives.
Yet the New York Times, mere days later, reported $13 billion annualized revenue:
OpenAl's business continues to surge. DealBook hears that the company's annual recurring revenue has soared to $13 billion, up from $10 billion in June — and is projected to surpass $20 billion by the end of the year.
First and foremost, it’s incredibly fucking suspicious that two very different numbers were reported here so closely, and even more so that the June 9 2025 announcement of OpenAI hitting $10 billion in annualized revenue was not, as I had originally believed, discussing the month of May 2025.
This likely means that OpenAI is not using standard annualized revenue metrics - which would traditionally mean “the last month’s revenue multiplied by 12,” and instead choosing “if all the monthly subscribers and contracts that are currently paying us on this day, June 9 2025, were to be multiplied by 12, we’d have $X annualized revenue.”
This is astronomically fucking dodgy. For the sake of this analysis, I am assuming any announcement of annualized revenue refers to the month previous. So, for example, when OpenAI announced they hit $10 billion in annualized revenue, I am going to assume this is for the month of May 2025.
This analysis is going to favour the companies in question. If OpenAI “hit $10 billion annualized” in or around June 9 2025, it likely means that their May revenues were lower than that. Similarly, OpenAI “hitting” $12 billion in annualized revenue (announced end of July 2025) - which I have factored into my analysis - is considered the revenue they hit in July 2025.
In reality, this is likely to credit them more revenue than they deserve. If June’s annualized revenue was $10 billion, that means they made $833 million, rather than the $939 million I credit them with for the month.
One cannot hit $12 billion AND $13 billion annualized in one month unless you are playing extremely silly games with the numbers in question, such as moving around when you start a 30 day period to artificially inflate things. In any case, my analysis for OpenAI’s revenue for August is around $13.145 billion - so in line with a “$13 billion annualized” figure.
In any case, I am sticking with my analysis as it stands. However, the timing of these annualized revenue leaks now makes me doubt the veracity of their previous leaks, in the sense that there’s every chance that they too are either inflated or used in a deceptive manner.
Based on these numbers, OpenAI's current growth rate is around 9.54% — and at that current pace, it will finish the year at around $11.89 billion in revenue. This is an impressive number, meaning it’d be making over $1.5 billion a month in revenue by December 2025 — but such an impressive number will be difficult to reach, and mean it has something in the region of $18 billion annualized revenue by the end of the year.
I also question whether it can make it, and even if it does, how it could possibly afford to serve that revenue long-term.
In Anthropic's case, I am extremely confident, based on its well-reported annualized revenues, that Anthropic has, through July 2025, made around $1.5 billion in revenue. This is, of course, assuming that their annualized revenue leaks are for calendar months, and if they're not, this number could actually be lower.
This is not a question of opinion. Other than April, we have ARR for every single month of the year. Bloomberg is now reporting that Anthropic sees its revenue rate "maybe [going] to $9 billion annualized by year-end," which, to use a technical term, is total bullshit, especially as this number was leaked as Anthropic is fundraising.
In any case, I believe Anthropic can beat its base case estimates. It will almost certainly cross $2 billion in revenue, but I also believe that revenue growth is slowing for these companies, and the amount of cash we credit them as actually making is decidedly more average than "annualized revenue" would have you believe.