80% of OpenAI and Anthropic's Revenue Comes From 1% of Customers

Growth Newsletter: https://levelingup.beehiiv.com/subscribe Need marketing help? Visit: https://www.singlegrain.com/ and https://npdigital.com/Want to recruit great marketers? Find them here: https://marketingschool.io/hire New Ramp data says 80% of OpenAI and Anthropic's enterprise revenue comes from 1% of their customers, a concentration risk unseen in any other software category. Eric and Neil dig into what that means for the AI trade and for every company betting on it. Then Tomasz Tunguz's five-year writing data: AI didn't reduce his editing, it raised the quality floor, and the weakest posts gained twice as much as the strongest. Plus ICONIQ's headcount report (the 100%+ growers added 133% more headcount while the 50-100% band cut hiring almost in half), YC's product-market-fit data on user conversations, and a detour into foldable phones. Key takeaways◾1% of customers drive 80% of OpenAI and Anthropic's enterprise revenue. That's the risk◾AI raises the floor of your work, it doesn't shrink the effort◾10+ user conversations a week is the most predictive variable for finding PMF Chapters00:00 80% of AI revenue from 1% of customers (Ramp)03:29 AI productivity raises the floor (Tunguz)04:55 ICONIQ: who's still hiring in 202609:07 YC data: the one variable that predicts PMF13:03 Foldable phones and the China gap

• 0:00Speaker 00

  Finding customer information shouldn't mean digging through emails, call notes, and spreadsheets. HubSpot Starter Plan brings all of those details together, so it's easier to track what's happening with every deal and what to do next. It pulls information from calls and emails. Then it gets to work, logging and summarizing them. Right after your call ends, HubSpot pulls next steps from your call notes and drafts suggested follow-ups. And it stays up to date for you, so you can build demand, win more deals, and spend less time on repetitive work. Your records, insights, and next steps are ready for you to act on. And you don't have to miss a beat searching for any of it. Try the Starter Plan for $10 a month. Learn more at hubspot.com. Now, did you know that 80% of Anthropic and OpenAI's revenues come from 1% of the customers?

• 0:39Speaker 01

  It's way less than 1%, but yeah.

• 0:41Speaker 00

  Yeah, so this is actually new data from Ramp, right? So they have a lot of credit card data. So you can see here, top 1% of businesses, right? And then you have everyone else. This is the majority of their revenue. And then you have the next 9% and then everyone else. This is the standard Pareto principle.

• 0:56Speaker 01

  Yeah, but for them, it's even skewed because there's really so many large corporations. And there's a lot of people who'll give them, I don't know what their fees are, 20 bucks a month? What's the lowest entry? Is it 20?

• 1:05Speaker 00

  20 bucks, yeah.

• 1:06Speaker 01

  Yeah, so there's, there's a lot of people that'll pay $20. There's a fraction of 1% that can spend millions of dollars with them. And that fraction of a percent is where they're making all their revenue.

• 1:17Speaker 00

  And this is really risky if you're one of these companies, right? Because it's like this, this concentration risk, I've never seen this before, right? The companies in the top 1% skew heavily towards the tech sector. So you have like, like a Microsoft, for example, right? You have a Meta, for example, you have Amazon, which they're, they're kind of building their own models. Meta's built a model that is already like a, like 5.6 equivalent for, from like a chat GPT standpoint. And so here's the thing. If you look at Nvidia, they just bought Hugging Face for $13 billion. If they're in on the Hugging Face, which is an open source company, right?

• 1:51Speaker 01

  They were also one of the investors too. I don't know if you knew that.

• 1:53Speaker 00

  In what?

• 1:54Speaker 01

  In Hugging Face. Nvidia's one of the earlier investors.

• 1:57Speaker 00

  Nvidia also has money in OpenAI and I'm sure in Anthropic too. But they're, they're gonna just go for everything, right? And so you have Meta going open source, you have Nvidia going open source as well. So then what happens to these companies, right? And I still think they're gonna be fine, but the question is, is Anthropic a multi-trillion dollar company right now?

• 2:13Speaker 01

  Uh, they're supposedly trying to go public at over 2 trillion.

• 2:16Speaker 00

  2 trillion, yeah, bigger than SpaceX's.

• 2:17Speaker 01

  Yeah.

• 2:18Speaker 00

  Yeah, so I don't know, like it's, it's risky, but hey, I still love using these products. Don't get me wrong. I'm just like, this is kind of a little scary, so.

• 2:24Speaker 01

  It's, I, I agree with you. It's extremely scary, but their client concentration isn't bad. Think of a services business. So flip it around, don't look at the majority of their revenue coming from 1% of the customers. The boy bankers will look at this, is that 1% isn't five companies or 10 companies? So when you look at how much of the revenue comes from the major, uh, the majority of the companies that actually they generate revenue from, the concentration isn't as bad.

• 2:51Speaker 00

  So what Neil's saying is, it's not that bad. Also, these companies are fairly stable. My thing is, as the other models, like if in a year Nvidia's models are pretty much up there, right? I think there's always, again, there's always gonna be room for frontier models like Anthropic and OpenAI. But as Meta is basically catching up, right? All the money they're spending, like first, maybe a year ago, everyone's like, what's Mark Zuckerberg doing, right? Now it's like, oh damn, they're actually catching up, so.

• 3:14Speaker 01

  Yeah, because the way people look at it, like let's say someone in finance. Yes, uh, 99% of the revenue comes from people who pay small amounts. All right, uh, or 1% of the revenue comes from people who, what was it?

• 3:30Speaker 01

  80% of the revenue.

• 3:30Speaker 00

  80% comes from 1%.

• 3:31Speaker 01

  Look at, 20% of the revenue comes from 99% to be like, okay, here's that 99%. We don't have to worry about that. They'll look at the 80% and then within that 80%, it's not like one company makes up 20%. It's still a large number because they have millions of customers. And that large number, it's spread out. Yes, you can have some outliers in there, but I bet you it'd be very rare if Anthropic had more than, call it 10% of the revenue coming from one customer.

• 3:58Speaker 00

  Yep.

• 3:58Speaker 01

  If I had a guess.

• 3:59Speaker 00

  Yep, we'll see how it plays out, but I want to call something, this piece out. So, so this is AI productivity as it relates to writing, so content generation, for example, okay? So Tomasz Tunguz, the, the VC that is still at Redpoint Ventures or Theory VC now, okay? So he wrote a post, "AI productivity doesn't mean what I thought it means." So if you look at this, this chart over here, Neil, what this is explaining is the post that he's written, he has a blog, he's been writing for a while, right? And so you can see what AI is actually doing is it's helping the weakest posts gain, right? The strongest posts aren't getting that much better, but the weakest posts out there, they're becoming significantly better. So it means it's just raising the floor a lot higher.

• 4:39Speaker 00

  And what he's learning from this is that it actually means that we're doing, it raised the ceiling of the same amount of work, which means it actually still increases human effort. It doesn't shrink human effort.

• 4:49Speaker 01

  Yes.

• 4:50Speaker 00

  So we're gonna be doing more still.

• 4:52Speaker 01

  Yeah, I, I think we're entering into a new era where people are less of the narrative of AI is going to kill human labor. There's gonna be no need for human labor. I don't know what's gonna happen 10 years from now, but people are making this bet in the short run, which is like, oh, we're gonna have mass unemployment over the next few years. And you have really smart people on both sides, and I don't know who's right. I just know I haven't seen it with the companies we work with, and we work with a lot of large enterprises. I can't name one off the top of my head that really is laying off because of AI. Even in engineering, most of the companies I know are actually hiring, they're not laying off.

• 5:28Speaker 00

  Did you see the, um, the post from Iconic on headcount? So Iconic's a VC firm, right? And so, uh, they were talking about here, let me just pull this up for you. So, so basically, if you're a hyperscaler right now, you're adding a lot more headcount, right? So, um, you're adding 133% more headcount in, uh, the first half of 2026. But if you're in the 50 to 100% of growers, right, in terms of your growth rate, you're, you've cut your hiring by almost in half. And so it just depends on kind of where you're at. If you're a hyperscaler, you're gonna be adding a lot more headcount, but.

• 6:00Speaker 01

  Cut your hiring or cut your staff by half?

• 6:03Speaker 00

  Your staff.

• 6:04Speaker 01

  So they're firing.

• 6:05Speaker 00

  Yeah, okay, so no, sorry, sorry, this is headcount growth, this is headcount.

• 6:08Speaker 01

  Headcount growth, so they're, they're cutting how many new jobs they're looking for. So instead of looking for 100 jobs, if they're only growing at 40% a year, they're now only looking for 50 jobs. So they've slowed down hiring.

• 6:18Speaker 00

  Can you see?

• 6:19Speaker 01

  Or open roles.

• 6:20Speaker 00

  So I'll explain, because it's probably a little hard to see. So I'm showing a, a chart over here. And basically, the hyperscalers are at the very top in the red. So you can see in 2023, 2024, it actually went down, what the, the growth rate went from, uh, 119% to 65%. Like people are like, oh my God, you know, AI is gonna destroy a bunch of jobs, right? 2025, it shoots back up to 112%. And then 2026 first half is 133%. This is the hyperscalers up top. Now, if you're in the 50 to 100% uh, growth rate band, you're in the yellow over here. Am I making sense so far?

• 6:51Speaker 01

  Yes.

• 6:51Speaker 00

  Okay, 2023, 2020, uh, 2022, 2023, so 47%. Then in 2024, it goes to 35%. And then 2025, it goes to 46%. And the first half of 2026, 25%. So you can see these lower bands over here are actually hiring less. And it kind of maps to some of these conversations they've been having with other entrepreneurs, where they're talking about, the language they're using is like, not hiring as quickly, right? But when you look at the, the news from all the big companies out there, they're hiring a lot more.

• 7:18Speaker 01

  As Benioff put it on TV, he was with Dario, the Anthropic CEO. And I, I think it was on Kramer, he was talking about, hey, AI is crushing all this stuff and staff and all that. He's like, everyone, they were talking about how they're hiring like crazy. All of them are on Slack, all of them are using Salesforce. And he's just like, they're all a few blocks and they're adding more seats and increasing usage year over year, not decreasing it.

• 7:43Speaker 00

  Yeah, and Salesforce has, has good free cash flow. By the way, the, the SaaS apocalypse, man, is going up. Yeah, see, it's a good bet, but anyway, so.

• 7:53Speaker 01

  I still think it's, it's not over. I think it'll take like another six months to, six months for things to adjust to a year.

• 7:59Speaker 00

  Well, we had said this, this is a year trade.

• 8:01Speaker 01

  Yeah.

• 8:01Speaker 00

  So, um, okay, but I want to go a little deeper into this and we can, we can continue on. So if you look at, if you're growing at less than 25%, your headcount growth is 3%. So minus one point, okay? If you're 25 to 50%, your headcount.

• 8:14Speaker 01

  Sorry to interrupt, what do you mean minus one point? Because you're at three, you're still growing at 3%. I don't understand the minus one.

• 8:19Speaker 00

  Because the year before, the growth rate was 4%. And then this year so far, year to date is 3%.

• 8:24Speaker 01

  Okay, so it shrunk down, you're still growing, but you're only growing at 3% instead of 4%.

• 8:29Speaker 00

  Yes, correct. And so like mostly, they're, they're not adding as much, right? And so this one over here, by the way, you look at, it went from 16%.

• 8:35Speaker 01

  What's the percentages?

• 8:36Speaker 00

  This one, sorry, this one's, if you're growing at 25 to 50% as a company, you went from 16% last year, and then this year to date so far, you're at 10%. So you dropped six points, which is pretty significant.

• 8:47Speaker 01

  So you're growing your headcount by 10%.

• 8:48Speaker 00

  Correct.

• 8:49Speaker 01

  But it's 10% on a bigger number as well.

• 8:51Speaker 00

  That's correct. Well, I mean, we don't know what the whole sample size looks like, but yeah, yeah.

• 8:55Speaker 01

  Okay, read the rest, if you don't mind.

• 8:57Speaker 00

  And then if you're growing at 50 to 100%, you were 46% last year, and now it's 25%, so down 21 points.

• 9:04Speaker 01

  Were they growing at 25% the year before? Because I'm assuming their growth is slowing down, that's also causing.

• 9:09Speaker 00

  We don't know that, they don't, yeah. But we have to pull the Iconic report, this is like Saster did a, did this. But like, you can see that the, the hyperscalers are more or less continuing to, to push more. But I just think it's, it's interesting that when you, when we talk to friends that have, uh, you know, maybe companies that are growing at a decent rate, they're, the language is still very much, we're not looking to add as much. But they are looking to add more AI-pilled talent, I'll tell you that much, everybody is, so.

• 9:32Speaker 01

  Yeah, yeah.

• 9:33Speaker 00

  Anyway, that's that, so let's move back over here, so. Frictionless Growth Marketing, hosted by Sonia Thompson, is brought to you by the HubSpot Podcast Network, the audio destination for business professionals. Frictionless Growth Marketing, hosted by Sonia Thompson, shows how modern brands grow by removing the friction in their marketing and customer experience that pushes customers away every single day, often without leaders even realizing it. Each episode delivers practical insights, real brand examples, and conversations with marketing leaders, showing brands how to win with the new rules of growth, identity-driven marketing, cultural relevance, and frictionless customer experiences. Listen to Frictionless Growth Marketing wherever you get your podcasts.

• 10:14Speaker 00

  Real quick, if you want to acquire customers faster and more efficiently this year with the latest strategies and tactics, then check out singlegrain.com. That is my ad agency, again, www.singlegrain.com. Check it out, and if it seems like a fit, we'll get in touch and help you with a free marketing plan.

• 10:31Speaker 00

  Let me show you this, Neil. What do you think is the one X factor to getting to product market fit? And do you want to define what product market fit is? And then I'm going to show you some data.

• 10:40Speaker 01

  So Eric Reese and the Lean Startup and Sean Ellis back in the day, they had this thing where it came to product market fit. You would look at how many people, how many of your customers were extremely satisfied and couldn't live without your product. And they would run a survey. It's probably a terrible way of defining it. But based on the data, depending on how many of your people really love it and can't live without your product, you tend to have something called product market fit. And in essence, what they're really looking at is if a product has really high usage and people keep needing to use it, you're more likely to have product market fit than if you don't have that.

• 11:18Speaker 00

  In essence, really, product market fit is just people find your product useful.

• 11:23Speaker 01

  Yes.

• 11:23Speaker 00

  And the majority of people find your product useful. And Y Combinator has this batch outcome data. Okay, so Y Combinator, they backed Airbnb, they backed Dropbox, other companies.

• 11:33Speaker 01

  A lot of them.

• 11:33Speaker 00

  Yeah. So when you think about this, companies that entered a batch, so a batch is like a class, okay, for startups joining, right? So companies that entered a batch with 10 plus user conversations per week found product market fit within six months on average. Okay. Now, on the flip side, companies with fewer than two user conversations per week took 18 plus months or never found it. So it's the frequency of user conversations. It's the most predictable variable for PMF, product market fit timeline. And this is interesting because there's a company that I backed called Levels. They do the continuous glucose monitor. When I talked to the CEO, I can tell he was doing a customer call with me, right? Just the way he was asking me questions. I was like, how many of these have he done? He said 4,000, right?

• 12:14Speaker 00

  Now, did they necessarily, you know, hit rocket ship growth? Not necessarily, but I do think they found product market fit. But that's the amount it takes. And this is one thing I'm training our engineers on right now. It's like, hey, guys, it's not just ship a bunch of stuff. You actually have to talk to our team internally. You have to talk to customers as well. And that's not something you're used to doing. But now that you can do more, you're expected to know how to talk to customers. You're expected to know how to do marketing moments as well. But most people don't know that you actually need to talk to you. I remember Neil, back when I first met him, when I would call him, his voicemail would go to, this is Neil from Kiss Metrics, right? So he was actually talking to customers all the time. You want to talk about that?

• 12:51Speaker 01

  Yeah, I always talk, I still talk to him right now. And the reason being is if you talk to customers versus don't talk to them, you're only going to learn so much. Your team can tell you one thing. A customer is going to tell you something else. It doesn't mean the customer is right. Actually, I genuinely believe in most cases, not all the customers right. And in some cases, you know, your team's going to be right. But people have different viewpoints. And if you don't talk to people, you won't understand what is the real truth and what needs to be changed and what needs to be fixed. And that's why I love talking to people. But, you know, you talk, if you go back to when Steve Jobs was around, he would talk about how if you ask your customer what kind of phone they may want or device, they'll tell you something, but it's not really what they want

• 13:35Speaker 01

  because they don't always know what they want. So you have to be careful when you get feedback as well. Because sometimes the feedback is based on their limited understanding of that subject. And it could be skewed. Like, back when the iPhone first came out, if someone would ask me what kind of phone would I would want, I'd be like, oh, the BlackBerry? It's really cool. That is what I would want. Like, uh, like, uh, before the iPhone came out, right? I would be like, BlackBerry, you don't need anything more than that. You have a mouse, you have apps, you have a keyboard. This is great.

• 14:17Speaker 01

  If you look actually at iPhone, I don't want a BlackBerry anymore. iPhone is great. But I couldn't envision that until I actually saw the product. And by the way, you know, Apple has a new phone coming out on Wednesday.

• 14:30Speaker 00

  The flip phone, are you getting it?

• 14:31Speaker 01

  The, the foldable.

• 14:33Speaker 00

  Yeah, I know. Not the flip phone, the foldable, you're right.

• 14:35Speaker 01

  Yeah, I don't know if I'm going to get it.

• 14:38Speaker 00

  Well, I'm Chinese, so I might get it. Because Chinese people get it, I'm just kidding. I would, I'll probably still, I'm used to, here's the thing, that's, this is interesting, right, Neil? Because back in the day, when the iPhone first came out, I had a BlackBerry. And I was looking, when the iPhone came out, I'm like, why would I need this? This is so stupid, right? So what might happen is when this foldable phone comes out, we might look at it as like, it's really stupid. And we'll probably still end up getting it anyway.

• 14:58Speaker 01

  I may get it. I'm curious, because a lot of these phones are made in China. And whoever the manufacturers are, for somehow they always have leaks. And you can actually see like what it looks like. I don't care for the design as much. I hope it looks better than what people are showing online. So we're going to have to wait and see. But they already have the molding. I'm curious to see how it works. And I'm curious to see if you can watch like a movie better. I'm curious to see the size when it folds and fits in your pocket. I'm more interested in what Huawei has, the foldable.

• 15:29Speaker 00

  Huawei?

• 15:30Speaker 01

  Huawei. I'm more, wow, my bad. I'm more interested in their foldable tablet. And if Apple comes out with the foldable tablet, which they're saying is a few years from now, that I would buy.

• 15:44Speaker 00

  It doesn't look like this, does it? It looks.

• 15:47Speaker 01

  That's what it supposedly looks like.

• 15:48Speaker 00

  I'm looking, because your dimension is, and I saw it trending on X. And so I'm like, oh, that's what it looks like.

• 15:53Speaker 01

  So to me, I don't know what it is. That's not that appealing from a design perspective. I don't know why.

• 15:58Speaker 00

  You know what? I mean, this looks like a Chinese phone to me. I mean, it is built in China.

• 16:03Speaker 01

  Yeah, like, it's like they're trying to create an iPad out of a phone.

• 16:08Speaker 00

  It does look like an iPad. It's an iPad phone.

• 16:11Speaker 01

  But I don't, do you really want that?

• 16:12Speaker 00

  No, I have an iPad right here.

• 16:14Speaker 01

  Yeah.

• 16:14Speaker 00

  I'm good on this.

• 16:15Speaker 01

  So it's like, I don't know if I'm going to buy it.

• 16:18Speaker 00

  Well.

• 16:19Speaker 01

  Because I don't really use my phone for anything other than emails.

• 16:24Speaker 00

  Like that? Yeah, I don't know about that. This is my first time seeing it.

• 16:28Speaker 01

  Yeah, so because I only use my phone for phone calls. Uh, sometimes doom scrolling on Instagram, but not that often. Uh, funny enough, a lot of the people who log in from my Instagram, you know, like my employees doom scroll on my account. And I can tell because.

• 16:45Speaker 00

  Because the algorithm changes.

• 16:46Speaker 01

  Yeah, and I was like, what are all these car videos? They're like, what are all these fitness videos? Like, you know me, dude, am I really that into fitness?

• 16:54Speaker 00

  No, no.

• 16:55Speaker 01

  So then I get the most residency. And I can tell who did it because sometimes like there'll be, um, uh, videos of women. I'm like, oh, I'm like, I know who it is because they're single, right?

• 17:06Speaker 00

  That's funny.

• 17:07Speaker 01

  And then some are like cooking and I'm like, oh, I know who did this as well.

• 17:10Speaker 00

  Yeah, that's funny. So you can tell who last used it, basically.

• 17:13Speaker 01

  Because they accidentally forget to switch back to their account.

• 17:17Speaker 00

  Yeah, that's funny.

• 17:17Speaker 01

  And then the best is when some of my videos start turning into Portuguese. And I'm like, I know who's doing this.

• 17:23Speaker 00

  Yeah, that's funny. So that's the lesson. Keep being consistent. That's how you went at marketing.

• 17:28Speaker 01

  Yep.

• 17:28Speaker 00

  All right, goodbye.