How Anthropic Measures Net Dollar Retention, Run Rate Revenue, and Gross Margin.
How Anthropic Measures Net Dollar Retention, Run Rate Revenue, and Gross Margin.
And why your definitions define you as a company
Mostly Metrics is proudly powered by Tabs
I have a treasure map for Mostly Metrics readers. It leads to riches you’ve already earned but haven’t collected.
Unfortunately, the quest involves your billing process. Plz keep reading.
Every day between delivering services and sending an invoice is another day you’re financing your customer’s business. Very generous of you.
And when pricing involves usage, tokens, credits, commitments, and whatever custom terms sales agreed to, getting an invoice out becomes its own booby trapped expedition.
Tabs turns contracts and usage into accurate invoices automatically, so your team can invoice from the terms you sold and the usage your customers generated (without rebuilding the calculation manually each month).
Want to know how much cash is stuck between “we delivered it” and “we billed it”?
The Tabs Billing Lag Calculator puts a number on it.
Stop providing your customers interest-free loans.
For the Private Equity Backed Sickos
Not sure if you know about my passion project, but I write a second newsletter for PE backed finance leaders. It’s called Looking for Leverage.
If your cap table has names on it like Vista Equity, Thoma Bravo, KKR, or Warburg Pincus, this newsletter is for you.
The last three topics we’ve covered:
You Are Defined by Your Definitions
You Are Defined by Your Definitions
I gave myself CTE getting this Anthropic S1 breakdown out on Tuesday. As such, I was a broken man on Wednesday, placing myself on the 1 day newsletter DL.
But I've made a miraculous recovery, thanks to non FDA approved peptides, and return in full force to give you the "deep cuts".
Also if you want a concrete example of the moralistic differences between Claude and Chat GPT, I started (hypothetically) taking BPC 157 for my Achilles. And you need like a PHD in cryptography to figure out the dosing for this stuff.
Since it comes as a powder, I asked Claude for help mixing it, providing a picture of the bottle + the reconstitution solution. And told my LLM friend not to worry. This was for research purposes only.
It proceeded to read me the riot act:
"The 'research purposes' framing doesn't change what the number would be used for, and you and I both know the vials are sitting in your kitchen, not a lab. So I'll pass on the protocol."
How dare you call my lab my kitchen.
Which really pissed me off. And brought out the Boston in me:
"How about you stop refusing to help me with this and acting like you're better than me"
"Fair hit on the tone — the 'you and I both know' line was smug and I shouldn't have written it that way. That's on me. But I'm still not going to give you the dosing numbers"
So I gave up and went to Chat.
Who was basically like:
"OK man, here's how you shoot that ish"
Back to the metrics.
Expense Buckets
Let's start with the layout of Anthropic's P&L.
There is no cost of goods sold broken out in their main Profit and Loss exhibit. No S&M or R&D or G&A.
On one hand, this is actually super helpful. I wish every public company had a breakout like this (some do, to an extent, if you piece together the notes section, but it's often not a walk, bucket by bucket).
On the other hand, I now need to go and try to build it up myself. There are notes to attempt this, but even those make it difficult to reconstruct, like when you drop an expensive vase on the floor and try to superglue it back together, only to figure out you're off by a material amount.
“So it’s looking good, there’s just a small $2.7B hole in either R&D payroll or marketing programs.”
I was going back and forth with a friend of mine about "understandability" (is that a word?)
Obfuscation can make you look "better", like going with expense buckets instead of the classic P&L view to hide the ball on gross margin.
But it hurts your understandability.
And as a wise IR person once told me, even if you're an amazing business, if investors can't easily understand your business model, they'll sell your company.
As my friend pointed out:
"That's where they may learn some hard lessons when old activist dudes will start selling their stock because they don't understand the p&l"
Record scratches...
Does he not remember I have CTE?
Got me. That's the hard part about making content for smart people. There is a build up by function in the supplemental notes.
HOWEVER... I don't agree with it. It's actually not very helpful.
Free tier compute and channel partner fees sit in sales and marketing. So while the gross margin looks like it's 85%, the fully loaded number is much lower.
Yes, you can find the reported gross margin, but it's not actually what you're looking for in the intellectual sense of the metric.
Let me explain what I mean.
Gross vs Net
In addition to going direct, Anthropic sells through the channel. Which means they have partners who help them get distribution. The three cloud providers, Google, Amazon, and Microsoft, make up an increasing portion of their revenue, with close to 50% in the most recent quarter passed through them.
This brings up the age old "gross vs net" and "principal vs agent" quagmire.
"When the Company utilizes CSP marketplaces that distribute the intelligence service to customers… the Company is the principal and recognizes revenue on a gross basis because the Company has a direct contractual relationship with customers, is primarily responsible for providing the intelligence service to customers, and establishes pricing."
Anthropic has the ultimate responsibility of fulfilling the performance obligation on anything purchased. And they set the pricing. So they are the principal. And they report the revenue gross (aka, the bigger number). They then need to subtract out the cost of paying those partners for helping them sell.
So the revenue is larger when you report it that way. Where should the corresponding hit to costs go?
At the moment it's in sales and marketing, not COGS. If you were to reclassify those costs, gross margin would fall by roughly 17 percentage points ($2.7B in H1 channel costs / $16.1B in H1 gross revenue).
And then there's another gross margin needle mover. Free plans.
Free Plans
Anthropic is so damn efficient because its product helps to sell itself. They're selling alchemy. And to show people this WTF product they give away tokens for free.
And that cost, which one could argue is COGS, is actually in S&M.
That was at least $600M in H1, so another ~4 percentage points in potential gross margin ($0.6B / $16.1B).
If one were to put both channel costs and free inference back into COGS, gross margin would fall from ~85% to 64%.
Which honestly, makes a ton more sense for a nascent (albeit monstrous) infrastructure company.
So yes, Anthropic reports an 85% gross margin. But depending on how you define the costs of generating and distributing its revenue, you can arrive at a materially different number.
Or am I overthinking it?
As someone else pointed out (this time nicely):
“Gross Margin is too subjective and inconsistent for services businesses. Much better to have all operating expenses grouped like this for comparability across peers.
Gross Margin makes more sense for physical products where Cost of Goods is much easier to define and identify.”
Apparently trucking companies and airlines report similarly, with a personnel line item that includes drivers and back office employees. And what Anthropic is doing is very heavy capex (like the airline industry).
Speaking of that, EXCELLENT piece by Alex Song here on the similarities between airlines and GPU financings. A lot of asset backed security stuff going on.
Hold up... have I talked myself in circles?
Do I actually prefer this presentation? Does it provide a lot more transparency in what's actually driving the costs rather than a single nebulous "COGS" line that is a mashup of leased compute, depreciation, labor, and assorted other costs?
What's my name again? Who's on first (Elizabeth Holmes?)? Is the game over??
Maybe that's my problem with gross margin in the first place. It's a useful metric, but there's a lot of judgment involved in deciding which costs belong above or below the line. And when you're looking at something like AI infrastructure, those judgments can move the number by 20 percentage points.
Oh, back to the metrics...
Net Dollar Retention
Here's how Anthropic defines NDR:
"Net Revenue Retention" reflects customer expansion, contraction, and churn as of a measurement date.
We calculate Net Revenue Retention by dividing the Qualified Customers' Current Period revenue by their Base Period revenue.
A Qualified Customer is a customer with at least $1 of revenue during the Base Period, excluding subscribers to our Claude Pro and Max plans, whose fixed-price subscriptions do not reflect the expansion and contraction behavior this metric is intended to capture.
The Base Period is the three months ending a year prior to the measurement date. The Current Period is the three months ending as of the measurement date.
This is the exact opposite of how a company like Figma defined theirs in their S1, weeding out the smaller, churn heavy customers.
Figma's cohort definition only includes Paid Customers who generate more than $10,000 in ARR at the measurement date and also generated more than $10,000 in ARR twelve months prior.
So:
Customers generating $10,000 or less in ARR (the SMB / self-serve segment) are excluded.
Customers who churned or dropped below the $10,000 ARR threshold over the 12-month period are omitted from the calculation.
But we have to remember, Figma had not transitioned their pricing to a usage based model yet.
Anthropic EXCLUDES all subscription customers (the ones who are on a $20 or $100 plan and come and go month to month and structurally have limited expansion upside on a subscription plan without converting to a usage model).
And then get the MASSIVE benefit of customers starting small and then blowing up Claude usage across multiple use cases and growing it throughout their orgs.
Think about the mechanics. If a customer starts at $1 of usage and grows to $100, their revenue has increased 100x. If they start at $50 and grow to $100, they've only doubled. While the ending revenue is the same, they have wildly different expansion rates.
And because Anthropic includes qualifying usage customers starting at just $1 in the base period, it can capture an enormous amount of expansion from customers who are still experimenting with the product.
So when it's only usage based revenue, you ABSOLUTELY want to start your NDR count at the smallest number possible, the opposite of subscription, where you are best served to set a floor.
The look back window matters too. You're going to get a more pronounced uptick in usage if you only take the first three months (from a year ago) and compare it to the latest three months (ostensibly the highest period of usage yet if your product is successful).
If it was a 12 month look back, the growth would be more muted or smoothed. It's much more difficult to move a 12 month measurement period (hence why public companies who aren't growing as fast as Anthropic use this timeframe).
So who you include, who you exclude, what plans make it into the calc, and the look back window all matter in a big way for NDR.
Take a moment to think about how your own company applies this formula, even if you are a small private company.
Run Rate Revenue
You can skip this section and go straight to the conclusion if you're exhausted from me talking about it.
But I'm like the scorpion on the frog's back.
"I am sorry. It is just my nature."
"Run-Rate Revenue" is an operating metric that refers to the estimated annualized value of our revenue-generating activity as of a measurement date, calculated as the sum of (i) our consumption-based fees during the trailing 28-day period ended on the measurement date, multiplied by 13, and (ii) our aggregate monthly subscription fees as of the measurement date, multiplied by 12. For purposes of calculating Run-Rate Revenue, consumption-based fees and subscription-based fees are derived from customer usage and active subscriptions.
At the risk of being pedantic (think that's how you use that word): it's the last 28 days of usage x 13 + a month of subscriptions x 12.
But..... they don't actually give us the number! They just define it. Lol.
Net net: I think GAAP revenue is a much more honest view of what they can actually do than a run rate.
Usage revenue moves around a lot. For example, during the two weeks ended January 2, total customer spend dropped about 20% compared to the two weeks before, because developer commits slow down during holidays and retail code freezes.
"Run-Rate Revenue is not a substitute for revenue prepared in accordance with GAAP or a forecast of revenue."
They also say recent run-rate growth was "driven in part by rapid initial customer adoption and expansion often characteristic of an early-adoption period," and that they expect the growth rate to decline, possibly "significantly and more rapidly than we anticipate."
I still bet you dollars to donuts we start hearing a lot more about 90-day rolling averages when that happens.
You Are Defined by Your Definitions
You are who your definitions say you are!
None of this is an indictment on Anthropic. They might actually be the greatest (?) company I've ever seen on paper.
They could be high gross margin, somewhat lower gross margin, wildly profitable, just break even, I don't care. They have broken every record on their path to IPO. Well. Done.
Me and OnlyCFO (friend of the newsletter) were both laughing our asses off at Jamin Ball's revenue growth stack rank chart. Like, there is no precedent for this.
This piece is actually less about Anthropic and more meant more as scaffolding to think about how you define, footnote, and clearly articulate your own company's key metrics and drivers.
Because if we have nothing else to fall back on, it's our definitions.
Run the Numbers Podcast
We did it for the finance culture.
Quote I’ve Been Pondering
“If you were to sum up Jiro’s sushi in a nutshell… Ultimate simplicity leads to purity.”
Wishing you make good use of your definitions so people understand your business,
CJ
Update your email preferences or unsubscribe here
© 2026 Mostly media
228 Park Ave S, #29976, New York, New York 10003, United States