In this episode of Future Finance, hosts Glenn Hopper and Paul Barnhurst discuss Anthropic's reported IPO plans, financial performance, and the future of AI profitability. They explore the company's reported S-1 filing, rapid revenue growth, infrastructure costs, and the risks of investing in AI companies.
Glenn and Paul also explain how different revenue accounting methods affect financial reporting, whether current AI valuations are sustainable, and what finance professionals can learn from the industry's growth.
Expect to Learn
What Anthropic's reported IPO filing reveals about AI risks.
How rising AI revenue and computing costs affect profitability.
Why different accounting methods change reported revenue.
What drives AI company valuations and investment risks.
How AI is changing financial analysis and decision-making.
The discussion highlights why finance professionals need to look beyond revenue growth and focus on profitability, cash flow, and long-term financial sustainability when evaluating AI companies.
Follow Glenn:
LinkedIn: https://www.linkedin.com/in/gbhopperiii
Follow Paul:
LinkedIn: https://www.linkedin.com/in/thefpandaguy
Disclosure: Portions of this episode (such as the introduction or promotional segments) use AI-generated voice narration produced under human editorial review.
Future Finance is sponsored by QFlow.ai, the strategic finance platform solving the toughest part of planning and analysis: B2B revenue. Align sales, marketing, and finance, speed up decision-making, and lock in accountability with QFlow.ai.
Stay tuned for a deeper understanding of how AI is shaping the future of finance and what it means for businesses and individuals alike.
In Today's Episode
[00:00] - Introduction
[04:26] - Anthropic's IPO & AI Industry Growth
[08:54] - Major Risks in Anthropic's S-1 Filing
[15:24] - Computing Costs & Infrastructure Risks
[20:44] - Revenue Growth & Profitability
[28:31] - Revenue Accounting: Principal vs. Agent
[34:31] - AI Market Growth & Company Valuations
[40:10] - Cash Flow & Long-Term Investment Risks
[44:54] - The Future of AI Economics
[49:52] - Closing Remarks & Thanks
Full Show Transcript:
Co-Host: Glenn Hopper (00:42):
Welcome to another episode of Future Finance. I am Glenn Hopper, and I am using my radio voice today to introduce my co-host, Mr. Paul Barnhurst, the FP&A guy himself. Paul, how are you doing?
Co-Host: Paul Barnhurst (00:54):
Well, you're using your radio voice. I'm going to use, I don't know if you saw it on LinkedIn, but I shared a picture of- you know who Bob Ross is?
Co-Host: Glenn Hopper (01:01):
Yeah. Oh yeah.
Co-Host: Paul Barnhurst (01:02):
So on one side I had Bob Ross. On one side I had me and I called it Happy Little Forecasts. Today I'm wanting to have some happy accidents, if you know what I'm talking about.
Co-Host: Glenn Hopper (01:15):
Yeah, I don't know about happy accidents. I think they work fine when those happy accidents are trees, but probably not fine when they are drivers in your forecast.
Co-Host: Paul Barnhurst (01:25):
What I call them, variances are just happy little accidents. So I'm going to share real quick the image. I think our audience will get a laugh out of it. And so here we go, Glenn, what's your take? Happy little forecasts. What do you think? My new logo, I could pull it off?
Co-Host: Glenn Hopper (01:47):
I think so. And I think you guys both have distinguishing hair-based characteristics that make each of you stand out.
Co-Host: Paul Barnhurst (01:54):
All right. Well, I have to get your thought on one comment before we get to our real topic for today in this post, because I think you'll laugh. So someone told me, Louis Kinsey, can you please do a full episode in Bob Ross mode? Could catapult you to a new level of internet celebrity. Let me paint you a picture of quarterly results. Don't worry about what's behind it. That all gets painted over anyway. We just need them up and to the right for the report. A little more green, less red. Yep, that ought to do it. What do you think? It's like old and I go, if you send me an entire script, I'll do an episode.
Co-Host: Glenn Hopper (02:35):
We know you have to do it painting. You have to actually be in front of the canvas.
Co-Host: Paul Barnhurst (02:39):
I think what I might do is I might do charts instead of paintings.
Co-Host: Glenn Hopper (02:43):
Yep, yep.
Co-Host: Paul Barnhurst (02:43):
So I'll have Excel open because that makes more sense for finance and I'll kind of be building as I'm talking and, "Oh, well that was a happy up-and-to-the-right accident. We'll just leave it and keep going." Well, that driver increased revenue. Well, I balanced a budget with that fake number.
Co-Host: Glenn Hopper (03:04):
Let's drop in a cute little plug right here.
Co-Host: Paul Barnhurst (03:08):
And so that's what I'm thinking would make for a fun episode. So anyway, sorry audience, I totally derailed this episode because Glenn and I like to have a little bit of fun. So back over to you, Glenn, you were kind of getting us going.
Co-Host: Glenn Hopper (03:22):
Man, so this week, I mean, I guess the thing we talk about is Anthropic's S-1. That's what everybody's talking about and it's pretty interesting. I guess first off, 2026 has been the year for Anthropic sort of displacing OpenAI in the workplace. When you see their S-1, when you look at their revenue growth and the financials and beating OpenAI to the IPO, we're talking now about IPO by Thanksgiving roadshow the week before. Pretty amazing how much the AI landscape has changed this year. I'm just back from, I was in New York, I did a client dinner with Board Planning. I did an event with Microsoft, then I was in San Francisco for Basecamp doing a session with Campfire. Then I was at SuiteWorld where I had a session as well. And so deeply immersed in talking to people about AI for the last two and a half weeks nonstop without coming home in between.
(04:24):
But as you talk to people, it's very different right now than it was when you and I first started talking about AI and finance. Back then when you had me as a guest on FP&A Today talking about -
Co-Host: Paul Barnhurst (04:39):
Now three and a half years ago?
Co-Host: Glenn Hopper (04:41):
I guess so because what was it? Was it November of 23 or 22 when ChatGPT 3.5 dropped? AI is completely skewed my whole concept of time. I don't know. I feel like two decades have passed since that first episode we did together.
Co-Host: Paul Barnhurst (05:00):
Because I was trying to remember, was it 22, was it 23? It was November 22. You said you needed some time to finish, you built your application, your kind of FP&A tool. And so I think it was February when I had you on because probably December you were building something and then we had you on February, March of 23. So about three and a half years ago. And then you and I started this show in what, 24? March?
Co-Host: Glenn Hopper (05:28):
Oh, has it been that long? Yeah, I guess so. Yeah.
Co-Host: Paul Barnhurst (05:31):
And so we've been doing this for about two and a half, I think, or maybe it was June. It was somewhere in 24 we started this. It's been a little over two years because we've done well over a hundred episodes now. Now, some of those are reruns, but we've released well over a hundred episodes. I think we've recorded 60 probably, 70, somewhere in that range. So yeah, it's been that long.
Co-Host: Glenn Hopper (05:53):
I'm going to say when I built my little app using ChatGPT to help me write code, nobody was saying vibe coding back then and there was no writing directly into the... I was having to copy and paste code over, do my own troubleshooting. The context windows were so small.
Co-Host: Paul Barnhurst (06:11):
It was that app in a word at that point.
Co-Host: Glenn Hopper (06:13):
Right. That app that took me a couple of months to build using AI to help me write the code for it, that's one prompt and it's done in five minutes today.
Co-Host: Paul Barnhurst (06:21):
That's what I was just going to say. It's what about a five-minute project. What other technology, and maybe it's recency bias, because computers have every 18 months kind of the Moore's Law double and cost goes down. I don't think that's near as big an advancement, even though it doubled, did it allow us to do the leaps and bounds of things we're able to do today? An example, before we get to the S-1, and I can even pull it up, have you seen the video that shows Will Smith over the last couple of years eating spaghetti, the timeline? It goes from my hand is halfway through my face to I look like a real person in three and a half years.
Co-Host: Glenn Hopper (07:03):
Yeah. Those original videos were nightmare fuel. They were so, not even uncanny valley. They were just, I don't know what's going on here, but make it stop.
Co-Host: Paul Barnhurst (07:11):
They were awful. You're watching it going, my job is safe. Remember when we started Future Finance, we had the episode for a while. You know your job is safe when... And Gemini or Bard, I think it might've been still Bard at the time, recommended you put glue as a way to thicken your pizza paste or eat rock.
Co-Host: Glenn Hopper (07:35):
Hold it to the crust.
Co-Host: Paul Barnhurst (07:36):
Yeah. Or even the one example where somebody gave the pros of human sacrifice was when we... And I'm like...
Co-Host: Glenn Hopper (07:46):
So okay, here's the thing. I'm doing air quotes here. AI is messing up today, but the difference in how it's messing up today, getting us to the Anthropic S-1, 94 pages of risk factors that involve existential risk. So we're thinking of the same technology -
Co-Host: Paul Barnhurst (08:07):
Is it 94 pages long for its risk factors?
Co-Host: Glenn Hopper (08:10):
Yeah. And they're not like what you would see in an... I mean, they have the standard S-1 kind of risk factors, but these are existential... And I'll pull them up in this one here in a minute.
Co-Host: Paul Barnhurst (08:19):
Pause for a second. It's kind of blowing my mind. Has there ever been an S-1 with 94 pages of risk factors?
Co-Host: Glenn Hopper (08:26):
I don't think so. I mean, first off, 340 pages, that's a pretty hefty S-1, but this is -
Co-Host: Paul Barnhurst (08:31):
I mean, you get 200 sometimes. I mean, I think 100, 200 is probably pretty normal, but three pluses.
Co-Host: Glenn Hopper (08:37):
But I think this is on track. I don't know. We don't know where it's going to come in valuation-wise, but this could be a $2 trillion IPO. I mean, that would be insane thinking about -
Co-Host: Paul Barnhurst (08:49):
Yeah, but what are they raising though? I mean, valuation for the whole company may be two, but if we look at SpaceX, which was the biggest at time, what did they raise? 80?
Co-Host: Glenn Hopper (09:00):
Maybe. Yeah.
Co-Host: Paul Barnhurst (09:01):
I think it was under 100, and that was the biggest raise for an IPO and the biggest valued IPO. So either way, we're dealing with huge numbers. It's only 100 billion. How much do you make, Glen? Are you anywhere close to 100 billion a year?
Co-Host: Glenn Hopper (09:20):
I mean, there's a slight rounding error there that... Yeah. I mean, these numbers are just astronomical. So there's the length of the S-1, the 94 pages of things that you wouldn't... These are not financial disclosures. When Microsoft went public, they weren't talking about risks of human extinction, but they -
Co-Host: Paul Barnhurst (09:41):
You're going. Human extinction was a risk in there, wasn't it?
Co-Host: Glenn Hopper (09:43):
Oh yeah, absolutely. And it's like pages 22 through 110, and I've got it open here and I probably should have had better notes on the specifics of it. But a couple of the things that they cover are in the existential risk part.Let's see, I jotted some of these down. Where did I put those?
Co-Host: Paul Barnhurst (10:01):
Well, you're looking at that, just so everybody knows. I'm also pulling up Anthropic or Claude and asking it to summarize the risk factors from the IPO, which is just kind of... It's so weird to think, all right, hey, tell me about you're using AI to understand their company.
Co-Host: Glenn Hopper (10:21):
Okay, here's what I had in my notes. They mentioned misaligned behavior where models resist, shut down, conceal information, and they've shown behavior resembling blackmail and oversight depends on models checking other models. They also talk about catastrophic misuse and they talk about, oh, this was a good one, uninsurable liability. No single insurer and the possibility that the whole industry wouldn't have capacity to cover a catastrophic AI incident. So think about that. They're saying AI could do something so terrible here that the entire insurance industry couldn't cover the losses.
Co-Host: Paul Barnhurst (11:02):
Let's take worst c case scenario. It gets access to a code, and you know what I'm talking about, and launches a weapon of mass destruction. Isn't that probably the worst existential crisis it could do? I can't think... I mean, something related to nuclear has to be one or two.
Co-Host: Glenn Hopper (11:22):
Yeah. And they talk about bad actors getting it, bioweapons, and they cover all of this, which is interesting if you think about the brokers going out with this in the road show, and it seems like the anti-sell strategy, but I don't know. And truthfully, if something like that happened, I feel like liability coverage for Anthropic will be the least of our concerns.
Co-Host: Paul Barnhurst (11:48):
Yeah, if we blow up the East Coast, whether Anthropic can cover the damage that's done is not the first thing I'm going to be thinking about. I'm going to be thinking about, well, I guess no more episodes with Glenn.
Co-Host: Glenn Hopper (12:03):
I mean, what's their motivation for that? It's either the most candid S-1 ever or the best legal armor ever written, or it's one thing... Sam Altman and Dario Amodei have been, and all the people who've left both of their companies and talking about the existential risk, but the founders of these companies talking about the risk in a way, I don't know, it's hard not to picture it as keeping them in the news and keeping them at the top of the news. It's only bad publicity if they spell your name wrong. So if they talk about these negative things, but I'm not completely dismissing that that's a possibility. It's very interesting that all this goes in an S-1 and I don't understand their motivation.
Co-Host: Paul Barnhurst (12:54):
So I'm going to take... There's a couple different ways we could look at it. At the end of the day, most business... I'd say businesses are self-interested. You could talk about capitalism, the whole idea of Adam Smith and greed and the invisible hand and everybody's going to do what's in their best self-interest. And so the skeptical side is, hey, are we putting it out there as a way to say, look at us. We're saying if we don't have regulation, these are all the type of risks and it's on the government. We're trying to be responsible here. We even called it out in our S-1 to say there's all these huge type of risks and we've been saying publicly we need better regulation. And so if you don't give it to us when this happens, who should you be blaming if there's something catastrophic? Uncle Sam, big brother, not us.
(13:51):
Is that a pessimistic way to think about it?
Co-Host: Glenn Hopper (13:56):
Yeah. I mean, I guess that's it. But you know what? One of the risks they did mention was if there's draconian government regulations that come in place and slow everything down too. So that one actually makes sense as kind of a standard risk, not it's going to be the end of humanity. So that's what, 94 pages out of 340? That's whatever, a quarter to a third of the whole length on that.
Co-Host: Paul Barnhurst (14:21):
And so I searched the web and brought up a few things on it. So I want to go a little farther and get your though on a couple. So you mentioned government conflict. They talk about customer and channel concentration. Top five customers were about 25% of the revenue first nine months. I'm not too worried about that. They're diversifying very quickly. Compute supply and cost, yes, that's huge, but as long as they can make money, people will figure that out. The market will figure out cost and compute supply. Agree?
Co-Host: Glenn Hopper (14:51):
Yeah. And so the interesting thing here, but we'll go to their top line and their revenue growth and even their loss. But on those compute commits, you can't back out of them and it's something like 550 -
Co-Host: Paul Barnhurst (15:04):
637 billion in named compute contracts is the number AI brought back. I didn't check it, but I'm sure that's roughly right.
Co-Host: Glenn Hopper (15:14):
Yeah, and that's roughly, somewhere between 500 and 750 billion is what I had. And I think we've talked about it here, is this circle of money that goes around between NVIDIA and OpenAI and Anthropic and all. Well, I do have a comment on the Shell accounting game. However, there are two acceptable GAAP approaches. OpenAI takes one and Anthropic takes another. But to close out on the circle of money, I mean, so Amazon invests and then they hold convertible notes and then they sell compute and Amazon also sells Anthropic's models through its cloud marketplace, and then Google invests and it sells compute too. And then the chip commitments and all that, that same money is going in a circle and everyone's kind of booking it as a win. So is that a flywheel or a bubble? I mean, yes, to some extent. Who's to say who's going to ultimately win this, but there is going to be a market correction at some point on all this.
(16:26):
I mean, being in the middle of this bubble and pretending you're not would be a dangerous way to approach investing in any of this.
Co-Host: Paul Barnhurst (16:36):
Agree. So I'm going to run through just their summary of a couple more if you want to talk anym and then I'll let you take it back over revenue, everything else. So we talk compute, they list physical infrastructure, can the grid keep up? Valid concern. We're already seeing some of those challenges. I mean, the energy needs are huge. Being here in Utah close to the Stratos data center that wants to use twice the entire power of the state of Utah and be all methane gas. That's a lot of pollution you're putting out in a state that already during the winter has the worst pollution in the country. There's some huge legal risk with all that, I think. So that's an example. Now that's government data center, but the idea is still valid. Customer and channel concentration. Okay, I'm not too worried there. I don't think that's a big one.
(17:25):
Single product line, I get it. Hey, they're saying if something happens, it could be a huge drop. We could be disproportionately impacted. All right, fine. But I don't think anyone right now is super worried about that. Government conflict rulings, I think that's a legit good concern. Copyright liability.
(17:44):
1.2 billion and they have several lawsuits pending. I think they should be on the hook probably for a lot more, but that's a big risk. Revenue quality, fine. Model behavior, which is what we talked about, right? It's AI models may act in unattended, misaligned ways, cybersecurity, governance, talent and culture. And then the big one, catastrophic risk.
Co-Host: Glenn Hopper (18:10):
Yeah, that's uninsurable.
Co-Host: Paul Barnhurst (18:12):
Yeah. Question. I mean, if you were insurer, so you're Warren Buffet, let's say still working doing insurance, would you insure Anthropic and how much would you insure them for and how much would you charge?
Co-Host: Glenn Hopper (18:23):
Yeah, that's the problem with this. The rate of change we're hearing about misalignment and all the things out there. I mean, there's a cap that's well below whatever the potential damage that might be.
Co-Host: Paul Barnhurst (18:37):
Well, it's going to be even well below probably I would imagine their market value because how much would you charge someone to insure for two trillion? What insurance company can fund two trillion of claims? I mean, let's think Katrina was what, a few hundred billion? I mean, the biggest have all been in the hundred billion ranges, right? Have we had one big catastrophic event that's hit a trillion? I can't think of one. Probably five, 600 billion, and that's spread across multiple companies and governments stepping in. I mean, does Anthropic get a few hundred billion and it's where it can all be aggregated together from multiple... Because what insurance company's going to underwrite a trillion dollar or two trillion? I can't imagine any... They can't, can they?
Co-Host: Glenn Hopper (19:27):
Yeah. No, and that's what they identify in the S-1, so that the whole industry may not be able to cover it.
Co-Host: Paul Barnhurst (19:36):
I mean, I know you could insure them, Glenn, but the rest of us.
Co-Host: Glenn Hopper (19:42):
Yeah. So we've got the risk factors, but I'll tell you what was genuinely surprising to me in this is we haven't seen the numbers from OpenAI or Anthropic until this. Some has leaked out and we've seen a little bit, but with the SEC filing, we do see their actual numbers. And the big headline probably before we even talk about revenue is, oh my God, they had $81 billion loss. And I would have thought, oh, of course, it's so expensive to build these in the compute and all that, but that $81 billion loss, only 693 million is an operating loss. The rest of it is converting debt to equity. It's the -
Co-Host: Paul Barnhurst (20:31):
Yeah, 72 billion of the 81 is strictly related to change in future value of convertible notes, derivative, stock and note conversion type stuff.
Co-Host: Glenn Hopper (20:41):
Yeah. And that works for the company. They convert the debt to stock and then all the safes and everything, but having that much of an accounting adjustment, that says they had $693 million of loss in the first half of 2026, which still, that's not nothing, but if you look at... I guess this is the part that was even better than I and probably a lot of other people anticipated. So Anthropic was the second place, and we haven't seen OpenAI's numbers yet, but I'll be curious when they come out. So Anthropic is here first and we see, okay, 2023 for the full year, they had 41 million in revenue. So what's that, I don't know, a decent, like a PE-backed mid-market company. 2025, 4.6 billion. So that's pretty nice jump there, 10x over 10x jump.
Co-Host: Paul Barnhurst (21:40):
Yeah, you're a Fortune 500. You're not a Fortune one, but you're probably a few billion. You're now in an enterprise space, clearly. You're an enterprise company in billions.
Co-Host: Glenn Hopper (21:52):
So first half of 2026, 16.3 billion. So Q1 of 26 alone matched what they did in 2025. So Q3 adds about five billion of quarterly revenue and compute cost over this time dropped from... Compute used to be 2X what their revenue is, and now in the past year it's gone down to compute cost is 55% of revenue. Suddenly we have a pathway to profitability here and revenue per employee, $6.6 million. So this is actually feeling more like a business model when you look at, and I know you talked about that nobody's really, I don't think, too worried about that client concentration issue. They have 12,000 customers that are doing over 100K.
Co-Host: Paul Barnhurst (22:45):
Correct. I don't think anyone's concerned that their biggest issue is a customer leaves them.
Co-Host: Glenn Hopper (22:50):
Yeah. Yeah. Oh, and so their net revenue retention basically, so it's about 900%, which if you think about... So say that's 9X what their customers spent a year ago. And these are not contracts. These are people can... I mean, I guess for enterprise, I guess they probably... I haven't seen the enterprise contract for them, but these are people who are deciding month to month how to use this tool, and Anthropic is basically winning them over every month and getting them to increase their spend. And we saw that earlier this year when we saw, oh, it went from, okay, we're spending X amount per user per month to the whole token maxing thing when people -
Co-Host: Paul Barnhurst (23:34):
Let's take me as an example. I went from $20 a month to $100 a month. I'm a 500% increase on the revenue they're earning, and that doesn't count the fact that some months before I went up to 100, I also bought some extra compute. And even in one of the months where I think I was at 100, I bought a little extra, so let's call me 525 compared to what my baseline was last year. So I mean, I'm not that far... I mean, yes, that's half the 900, but that's on the consumer side, and I'm a small business. How many people, small businesses like you? You probably, when you were on ChatGPT, let's take you, when you were on ChatGPT as your main, you're doing what, $20 a month on Claude? Just a basic -
Co-Host: Glenn Hopper (24:20):
Yeah, because I was really, I mean, for a long time, I wasn't using -
Co-Host: Paul Barnhurst (24:23):
And then didn't you go up to the 200 a month account on Claude?
Co-Host: Glenn Hopper (24:26):
Yep, yep. But I'm also now on the $200 a month plan on ChatGPT.
Co-Host: Paul Barnhurst (24:33):
Let's just take that example. In a year, you increased 1,000%.
Co-Host: Glenn Hopper (24:38):
Yeah.
Co-Host: Paul Barnhurst (24:38):
You were greater than the average increase. I mean, it's crazy to think, but so if we take the two of us together, we're not far off the average. We're about a 750. I mean, if we use dollars because you're weighed heavier, it's probably close to an 800% increase in dollar terms for the two of us. I mean, it's kind of funny. You don't realize it till you sit down and all of a sudden start doing the math on your own. You're like, "Oh yeah, it does fit."
Co-Host: Glenn Hopper (25:03):
Yeah. So I mean, I guess the thing there was we saw this investment for years and we're like, "How are they ever going to get profitable? The cost of compute is so high." But now we're starting to see that because the usage is going up and the cost for compute is coming down. And you got to think there's that trade-off though. The models, the compute part gets more efficient, but we're burning through more and more tokens. So kind of an interesting...
Co-Host: Paul Barnhurst (25:34):
Ever feel like you go to market teams and finance speak different languages? This misalignment is a breeding ground for failure in pairing the predictive power of forecasts and delaying decisions that drive efficient growth. It's not for lack of trying, but getting all the data in one place doesn't mean you've gotten everyone on the same page. Meet QFlow.ai, the strategic finance platform purpose-built to solve the toughest part of planning and analysis of B2B revenue. Q flow quickly integrates key data from your go-to-market stack and accounting platform, then handles all the data prep and normalization. Under the hood, it automatically assembles your go-to-market stats, makes segmented scenario planning a breeze, and closes the planning loop. Create air-tight alignment, improve decision latency, and ensure accountability across the team. Here it is. This is 23, 24, 25, first half of 2025, first half of 2026.Go-to-market stack, makes segmented scenario planning a breeze and closes the planning loop. Create a
Co-Host: Glenn Hopper (26:48):
Yep, yep.
Co-Host: Paul Barnhurst (26:49):
Look how quick that came down. That's like Moore's law of the doubling of computers and halving the... I mean, it's kind of holding.
Co-Host: Glenn Hopper (26:56):
Yeah.
Co-Host: Paul Barnhurst (26:57):
Not exactly, but the basic principle, the whole idea that every 18 months you double what you're able to do and you half your cost. It's holding so far, and then some it looks like. I mean, one company, one example, I mean, it doesn't perfectly hold, but you can kind of see that pattern, right?
Co-Host: Glenn Hopper (27:17):
Yep. Yep.
Co-Host: Paul Barnhurst (27:18):
All right. Anyway, anyone who wants can look at that. I just quickly put up on the screen going from a thousand percent compute cost of revenue to 55% in just a few years.
Co-Host: Glenn Hopper (27:29):
Wow, this time has gone fast. I do want to talk about something that I think our audience, the finance and accounting nerds will like. So they're kind of a battle between talking about their revenue and revenue growth. There's a difference between the way OpenAI books revenue and how Anthropic is doing it. So think of the two companies are selling the same thing to the same middle, and one counts the full price as revenue, and then the other one counts only its slice. So it's the same money, same profit, but it's a different headline revenue number.
Co-Host: Paul Barnhurst (28:08):
Sure. One's trying to take everything as revenue to show a bigger number, and the other is basically doing a net of sharing. It's like the processors who used to try to count everything that ran on their RELs as revenue, and you're like, "No, you're getting a 2% fee. That's not all revenue."
Co-Host: Glenn Hopper (28:26):
Yeah. But in accounting, that's principal-versus-agent. So if you're the principal, well, you're the one on the hook for delivering the product, so you record the full sale. And if you're an agent, then you're a go-between, so you record only your fee. So OpenAI is reporting as if they're an agent, which is understating the revenue compared to the way Anthropic does it. And the test, I guess, for principal-versus-agent would be who controls the product before the customer gets it and who carries the risk. I think I'm siding here with Anthropic on the way that they would put... And now that Anthropic has said this is the way we do it, I couldn't see OpenAI coming into IPO and not recording revenue the same way because both are allowed under US accounting rules. Anthropic says they're the principal. OpenAI says we're not, but whoever's right, an outside auditor has to sign off on it.
(29:23):
So obviously Anthropic is going through all this now. So the difference is we get these really huge revenue numbers and the stocks are going to be priced as a multiple of revenue, so that's important there. Margins, if you book... Hang on, I'm getting my debits and credits mixed up. If you book gross, you're going to have more revenue, but you're going to have a lower margin. And then I guess if they're booking it differently, you can't compare them. So I think now Anthropic ha said, "This is the way we're reporting it, so that's the way OpenAI's going to report... As well.
Co-Host: Paul Barnhurst (30:00):
Right. I mean, and it makes a huge difference. An example, I supported a business. It was a shuttle business for car dealership saying, "Hey, don't do your own shuttles to take people home when they leave their car. Use Uber and Lyft through us. We give you an application. You don't have to worry about the depreciation and having the car and having the driver and all that kind of stuff. We recognize the full revenue." 90% of that went to Uber to pay the driver and their share or 95%. So we had a million in revenue, 995,000 in expenses or whatever. And so a net of like 5,000. I mean, it's amazing how different that can look. And you really have to look at that net when you're thinking about growth because no investor's going to give you full credit for that revenue. They're going to discount it.
Co-Host: Glenn Hopper (30:49):
Well, and then I guess that would say if you were valuing this regardless of how they report it, then you don't value it on revenue, value it on margin and profit.
Co-Host: Paul Barnhurst (31:03):
Yeah, I think so. So I mean, I think there's some interesting just accounting finance things. Anyone who's listening, what's your take? Do you think it should be principal? Do you think it should be agent? I mean, ultimately, what do you think is the right way? The way Anthropic's doing it? Ignore which one ultimately wins. What do you think is probably the cleaner, more kind of appropriate way? Obviously there's nuance here. We wouldn't have two major companies choosing different methods. We all know accounting has a... There's judgment in accounting. Where do you side?
Co-Host: Glenn Hopper (31:39):
I side on the way Anthropic's doing it, honestly. I think that they're the ones responsible for delivering the service and they can... Yeah, I don't know. I'm sure there are technical accountants who could argue with me.
Co-Host: Paul Barnhurst (31:49):
A hundred percent, of course.
Co-Host: Glenn Hopper (31:51):
Which is why we somehow ended up with the two Frontier Labs reporting in different ways. But I'll tell you what's interesting is this is the first Frontier Lab to undergo the SEC and the audit scrutiny of all this. So it is going to be very interesting to see what shakes out of this. And for a company, regardless of what industry you're in, to have the meteoric growth that Anthropic has and go from that startup mode to we have to get IPO ready, I'll be very curious to hear about the audit results from this IPO and the SEC and how all this goes. And thinking about how quickly it's coming, just filed the S-1 this week and pre-Thanksgiving IPO, it'll be interesting to see what happens across the market through the end of the year from this massive IPO rolling out.
Co-Host: Paul Barnhurst (32:44):
Yeah, it'll be interesting. And I mean it helps explain partly why Anthropic has grown so quickly that they're recognizing the full amount. I mean, it'd be really interesting to see what is their growth percentage for OpenAI and Anthropic when we have both of them and you have the ability to do the accounting the exact same way.
Co-Host: Glenn Hopper (33:04):
Well, soon enough we will because OpenAI will come in 2027, I'm sure.
Co-Host: Paul Barnhurst (33:08):
It will be interesting to see because there are sources to estimate. If it was net reporting, it would cut revenue. One source estimated the low end is six to 10%. The high end you're saying could be as high as 27% of the revenue. Six to 10 isn't that big of a difference because some of all the consumer stuff isn't going to be impacted by this.
Co-Host: Glenn Hopper (33:30):
What it all comes down to me is, so if we take the Q3 pace right now, put Anthropic between 65 and 70 billion for 2026, they have commitments, 600 and whatever billion that you had. The growth rate, we have super hockey stick growth here that can't go on forever. But the TAM here, and I don't think they explicitly discuss the total addressable market, but you have to think the total addressable, every knowledge worker job in the world out there is part is...
Co-Host: Paul Barnhurst (34:07):
Someone tried to estimate or saw something that said something like $30 trillion or some crazy number is the TAM.
Co-Host: Glenn Hopper (34:13):
And think about, even on the business side, how many businesses are not, they don't have a business? I don't know. I think adoption is growing and it's growing beyond just, okay, you can use Copilot. And I think businesses are deciding if you can use Anthropic or OpenAI or both or sure use... I keep leaving Gemini out of this and I do know one Fortune 100 company located here near me, Gemini is their primary provider, but I don't know many in the mid-cap space who are using that. But anyway, that's all part of Google, Alphabet, all that. But to meet the commitments, the growth has been insane. It's hard to say we're going to continue on this rocket ship or even accelerate it. I don't know if I would -
Co-Host: Paul Barnhurst (34:58):
Let's do a little math exercise. So 2027 revenue at the rate they're growing is what? 64 billion? Is that kind of the forecast or is that full year 2026?
Co-Host: Glenn Hopper (35:09):
Say 65 billion. Yep.
Co-Host: Paul Barnhurst (35:13):
65 billion and we're talking what? There's rumors that it could be hit as high as a two trillion IPO. So that's what, four times revenue? Or sorry, 65 billion, that's 40 times revenue. Am I thinking about that right? So I mean what company... But their margins, even if they can get to the point where free cash flow on 65, I guess if I can get up to a trillion, that's really not a bad in revenue. I mean, let's see, the biggest company in the world is what? Two, maybe 200 billion in revenue?
Co-Host: Glenn Hopper (35:51):
So NVIDIA has the biggest market cap right now, right?
Co-Host: Paul Barnhurst (35:55):
I believe so. They're right up there. But what is the biggest revenue company in the world for a second? And then what's their... So the biggest market caps are in that two trillion range. Amazon revenue tops 700 billion, ended 12 years run at number one last year. So I mean, if you think 700 billion in revenue, it's only three times revenue. And I would think the margins, if compute costs continues to drop, they're not margins of like an Amazon. They're a little higher.
Co-Host: Glenn Hopper (36:31):
The question would be... Yeah.
Co-Host: Paul Barnhurst (36:32):
That's
Co-Host: Glenn Hopper (36:33):
The question would be the -
Co-Host: Paul Barnhurst (36:34):
Trying to think, how do you support a two trillion valuation?
Co-Host: Glenn Hopper (36:38):
Yeah. And what are their capital requirements? What's the lifespan on these chips and how often? And when you finally just... Well, we're talking about compute. We haven't even talked... Well, you mentioned it earlier with the power demands as well. So there's significant, significant capital build out that's going to have to keep happening. But capital versus what the business spins off ahead of those capital allocation decisions. Yeah, I don't know. I'm not buying the IPO on this one. We'll let the market sort of settle down around it. I don't know what kind of... I mean, you've got to think it's going to have a lot of consumer bump on the day of.
Co-Host: Paul Barnhurst (37:22):
I would imagine it's going to have an 80% run up, if not more, on day one. Now the question is, what is it three months out? Six months?
Co-Host: Glenn Hopper (37:31):
And once all those employee lockups start expiring and...
Co-Host: Paul Barnhurst (37:34):
Let's talk about that. Did you see how they're doing the employee lockup? They're not doing a typical one.
Co-Host: Glenn Hopper (37:38):
No, I didn't see that.
Co-Host: Paul Barnhurst (37:39):
They're not doing it on a 180. I was reading, and I wish I had the article in front of me. I might look up and see what it says, but they're allowing on day 24, 48, 86, every couple weeks they're allowing certain groups and certain people to sell out. It's not your traditional lockup. It's very different.
Co-Host: Glenn Hopper (37:57):
And man, thinking about that, regardless of what happens just with the valuation being so high, there's going to be what? A couple hundred, a couple thousand new multimillionaires rolling around the Bay Area now?
Co-Host: Paul Barnhurst (38:08):
There'll be billionaires. So here's how it works. I just looked it up according to S-1. Day one, some shareholders who are not founders can sell up to 10% of their locked-up shares once trading starts. The filing measures that 10% against holdings at the end of the month before the prospectus date. Okay, fine. Then they can do it again at day 27, day 48, day 62, and 76 all allow further partial releases. The filing shows an additional 7% of holdings releasing at each of the day 62 and day 76 steps for at least one group. And it couldn't find on the web what the percentages are on 27 and 48. Main lockup end. For at least one group, the rest days locked until whichever comes first, two full trading days after the Q1 earnings release or the close of day 180. And then there are also some that are year-long lockup, so it's all over the place.
Co-Host: Glenn Hopper (39:07):
Well, going to be interesting to watch.
Co-Host: Paul Barnhurst (39:09):
All right. So I think this kind of would be a masterclass to sit down and do the valuation and see how people are valuing this company. Because if we just come back to finance, we got finance nerds on here. We all know valuation is a matter of what do you think the discounted cash flow is going to be? And this valuation is 100% dependent on the terminal growth and the terminal growth, not just in revenue, but really in capital expenses. I mean, personnel, fine, but the compute costs, it's not the electricity. Yes, the grid has to hold up. That's a risk thing. But the DCF is how much are you spending on capital? How much is that impacting long-term, your margins? How quickly can you get profitable? And how much can you continue to grow? This is very much a growth play. When do they start paying dividends?
(40:03):
Not anytime soon, I would think.
Co-Host: Glenn Hopper (40:05):
Yeah, no, no.
Co-Host: Paul Barnhurst (40:06):
10 years maybe.
Co-Host: Glenn Hopper (40:09):
On the people side also, their revenue per employee was over $6 million. So that's the thing with AI, and it's not just for Anthropic, it's businesses now across the board are with AI, the expectation that you don't have to... Think about how much their revenue grew over the last two years and relative to that, how few employees they had, and they didn't have to add new employees to get that revenue. I mean, this is kind of the model for the future and it is... Yeah, I don't know. Actually, they did have a statement in there about their own universal basic income that they're setting aside a billion dollars or something for. But anyway, that's a whole other show and we're just about at time here, but it is... Yeah, you know what? That would be a great Financial Modeler's Corner episode for you, wouldn't it, or at least a webinar to model out the valuation for Anthropic?
Co-Host: Paul Barnhurst (41:09):
I think it would. I think I'm going to get somebody on to do that. I just pulled it real quick. Revenue, first half of 2025, personnel cost was 46% of revenue. So ignore the per person revenue.That's pretty high. That's not unheard of for a software company. We often hear 70% of operating expenses, especially for hyper growth, but how quickly they brought it down. What do you think it was for the first half of 2026? Personnel was a percentage of revenue.
Co-Host: Glenn Hopper (41:37):
Gosh, so I need to look at their revenue growth again.
Co-Host: Paul Barnhurst (41:42):
I'll give it to you. They went from... First half was 1.1 billion and they went to 16 billion. Expense, so if 46%, their personnel was about 500 million, if that's 1.2 billion. So roughly 1.2 billion, we'll call it for revenue in the first half of 2025, call it roughly 5.50 of personnel, roughly 46%.
Co-Host: Glenn Hopper (42:07):
And so with that growth, man, I couldn't imagine.
Co-Host: Paul Barnhurst (42:10):
They increased from first half, they increased it 600% personnel cost, yet it went down on a percentage basis of revenue more than in half. It went from 46 to 19%.
Co-Host: Glenn Hopper (42:27):
Wow.
Co-Host: Paul Barnhurst (42:28):
So a 27% decrease, and their compute cost as a percentage of revenue went from 200% to 50%. So if you look at compute and personnel, which are their main costs, in first half of 2025, they were three billion, let's call it three on one, two eight on 1.2. So two and a half times as high, roughly. This year, the two combined were 12 on 16, so about 66%. So they were able to completely flip the model in less than 12 months in a hyper growth business. Well, hyper growth business does that.
Co-Host: Glenn Hopper (43:14):
I mean, and that's why I'm more optimistic seeing in this because all you would hear is the amount that was being spent and that early on how far they were from being profitable. And now you can start to visualize the path to profitability on this. So yeah, I think it's, again, still a bubble and I can't wait to see OpenAI's numbers behind this, but it does -
Co-Host: Paul Barnhurst (43:42):
Yeah, I'm still not encouraging anyone to run out and invest in the IPO.
Co-Host: Glenn Hopper (43:45):
Yeah.
Co-Host: Paul Barnhurst (43:47):
And the other thing I want to say, well, we've spent almost 50 minutes on this. Guess we bring it back home. What's our takeaway for the audience? But it's a fascinating time.
Co-Host: Glenn Hopper (43:57):
I mean, the takeaway to me is how quickly, like we said at the beginning of the show, how quickly we got here relative to adoption of the smartphone, of the internet, of broadband, of personal computers, and how it's just been a whipsaw trying to keep up with all of this and trying to train on it where it's hit a point now where it's in the air. AI is everywhere, whether you're using your favorite social media app or doing a Google search or any kind of work you're trying to do. So we're all using it one way or another. We don't know what the economics of it are going to be, but it's... Because think about the dot-com bubble and all the ask Jeeves and the other pets.com and what went away from it. And at the time, I don't know, I can't imagine what the crystal ball would show coming out of this.
(45:05):
Is one of these, is OpenAI or Anthropic going to end up being the Yahoo of the future or are they both going to coexist some way? Not to mention, oh, by the way, Facebook or Meta has their own platform, what Google's doing and all that.
Co-Host: Paul Barnhurst (45:19):
Google. Yeah. Well, the other thing I was going to say is if you go back and look at those companies, you go back to 2000, we were both around then, almost none of them... I mean, the amount they were, they were all paper companies. Their valuations, there was just no path there. It was just 100% exuberance. Looking at this right now, thinking of many of these finance companies, again, public where they're just nowhere close to profitable, its operating losses aren't bad. Those operating losses were what, on 16 billion, 600 million in the first half of the year? So at 4% or something? Yeah, 700. So not even a billion on 16 billion. Yeah, a couple percent. I mean, most companies that are going public, their operating losses, if they're a hyper growth company, are much higher than that. Now, obviously their total loss is the 80 billion and there's some noise in there and what's the...
(46:21):
But that's not cash. All that convertible notes and stuff, it's not like Anthropic has to pay 80 billion in cash.
Co-Host: Glenn Hopper (46:29):
Yeah. So the straight... Yeah, good point. The straight on paper operating loss is more
Co-Host: Paul Barnhurst (46:36):
Cash. I'm sure there's things below the line that you're going to have to factor in there, the capital expenditures and all those things.
Co-Host: Glenn Hopper (46:43):
So that 700 million is down from 2.1 billion a year earlier. So again, you're seeing the path towards an actual profitable business.
Co-Host: Paul Barnhurst (46:54):
Yeah. I have to say, and I think you would agree much better than you thought it would be. Is that a fair statement?
Co-Host: Glenn Hopper (47:00):
Yeah. Yeah. And it made more sense.
Co-Host: Paul Barnhurst (47:02):
The operating loss would probably be more like 10 billion or something crazy. And I think a lot of people did because the numbers we were hearing now, but here's what's... The reported capital spending is false because it rents almost all of it's in compute instead of buying it.
Co-Host: Glenn Hopper (47:17):
Right. It goes back to that's back to the...
Co-Host: Paul Barnhurst (47:21):
So I mean it's operating cash flow in the first half of 2026 was $3.9 billion. I was like, so free cash flow was about 4.3. As we close here, all numbers were generated by Anthropic. Remember, you should always validate AI generated numbers, but don't take anything I say as investing advice or as accurate.
Co-Host: Glenn Hopper (47:45):
Ever. Yeah.
Co-Host: Paul Barnhurst (47:47):
Oh, thanks. Ever. I'm just saying the AI. Wow. All right, but we've gone for 50 minutes. Should we call it here?
Co-Host: Glenn Hopper (47:56):
Let's call it. Let's call it. This is exciting times in the AI world and yeah, it was super revealing to go through and look at that S-1. I didn't read all 340 pages myself. That's my confession, but I did spend a lot of time like you did between Claude and my Copilot plugins going through and looking at the numbers. So otherwise I'd have maybe read... I don't know. How many pages of a 340 page S-1 would you read? And think about the lives of analysts now, how much easier their job is when they don't have to read a 340 page S-1 end to end.
Co-Host: Paul Barnhurst (48:33):
I would do exactly what I did during this call, pull up my AI and have it summarize things for me. I would look at the financials myself. I would look at the risk section, management discussion, depending on how long they are. I might read some of them, but a lot of it, I'm just going to let something summarize it for me.
Co-Host: Glenn Hopper (48:51):
Yep.
Co-Host: Paul Barnhurst (48:51):
All right. Well, thank you for joining us. We know that was a long episode. We'd love to hear from you. What are your thoughts on the S-1? You investing, overvalued. Do you care as a finance professional? Does it mean anything to you? Let us know, but it's an exciting time for the AI world. All right, radio voice to sign us off comes from Glenn.
Co-Host: Glenn Hopper (49:10):
Goodnight, everyone.
Co-Host: Paul Barnhurst (49:13):
All righty. Good night. Thanks for listening to the Future Finance Show. And thanks to our sponsor, QFlow.ai. If you enjoyed this episode, please leave a rating and review on your podcast platform of choice, and may your robot overlords be with you.