AI Valuations from Aventis

Live from Schenectady, New York, it's SaaS Talk with the Metrics Brothers, Growth and CAC.

And I'm Growth, better known as Ray Reich, founder and CEO of Benchmarkit.

And I'm CAC, better known as Dave Kellogg, EIR at Balderton Capital Independent Consultants. And the author of Kellogg.

And together we are the Metrics Brothers.

And we go together like froth and bubbles.

Bubbles? I love bubbles. When I was a kid, I loved Mr. Bubbles bubbles bath, but then I discovered champagne, Dave.

Not that— not those kinds of bubbles, Ray. I was thinking more about financial bubbles, as in, you know, the book This Time It's Different.

Oh, you're talking about stocks? Did you see that the price peak-to-earnings ratio of the S&P 500 is now around 25.9, and that's well above the historic median of 17.2 and the highest we've seen since 2000.

I did see that, Ray, but while the stock market does look pretty priced to perfection for me, uh, we're not an investor podcast, and listen to the trailer, we don't give either business or investment advice.

So why are we talking about bubbles then, Dave?

Well, I was thinking more about Sand Hill Road bubbles than Wall Street bubbles, Ray, and I was talking specifically about AI valuations. and fundraising, which will be the topic of this episode.

Ooh, double double, toil and trouble, fire burn and cauldron bubble.

Eye of newt and toe of frog, wool of bat and tongue of dog. Ray!

Hey, hey, we're the Metrix Brothers, Cat, not the Macbeth Brothers.

Oh no, so, so no more evil potions then, Shakespeare. Uh, and before we dive in, let's hear a word from our sponsor.

The Metrics Brothers is proudly sponsored by Benchmarkit, the B2B AI and software industry's largest, broadest, and most contextualized benchmarking index. Dave Kellogg and Ray Groth Wright have put the Metrics Brothers podcast on hiatus for remainder of 2026. All episodes we have ever recorded are still available on your favorite podcasting app. You can reach Dave @Kellogg and Ray on LinkedIn @RayReich. Thank you so much for listening to the Metrics Brothers podcast.

So today's episode is based on a report from Aventus Advisors called AI Valuation Multiples. Ray, we covered another report that Aventus produced a while back. I can't remember which one. Do you?

I don't remember. I apologize.

I can't remember. It was a good one though. I would say that these people produce some good research. Don't know them well, but they've put out several good reports. This is one of them. Uh, the The title of this report is Funding Rounds, Valuations, and Investor Universe from 2010 to 1H25. So that's the subtitle anymore, but that is the time period covered. So it's 15 years of data.

Hey, I was just shocked when I first saw this. Who's been tracking AI valuations for 15 years?

Yeah. Now, to be fair to Adventus, first, to be fair to Ray, it's a really good question. I think Ray and I are both old enough to remember the reality that in the year 2000, when the internet bubble was in full bloom, people would say, I need Java programmers with 10 years of experience, which, Ray, wasn't possible because—

It had only been around for a couple years.

Yeah, 5 years exactly. Java was invented in 1995. And this kind of thing does happen in bubbles where, like, literally HR people are saying you have to have 10 years of experience in a technology that's only lasted So, when Ray had this objection to the report, I was like, hey, wait a minute. Now, of course, generative AI is quite new, at least in reaching the mainstream in any way, but it is fair to talk about financing data back to 2010 for AI. In fact, even when I was— Jesus, Ray, but I was still living in France, I think. I had friends doing AI companies. They were neural network kind of machine learning companies.

Yeah.

Right? But the technology's been around for a long time, natural language processing, and that's what Aventus defines here. I went and found their definition. It's actually not in the report. But I found it online, and they're covering natural language processing, machine learning. There's a couple other subcategories in there. I don't have enough windows open to tell you, but yeah, this is 15 years of data, and it is going to raise some questions about what's in and what's out.

Yeah, well, I kind of thought about it, and it's like Marvin Minsky, who was, you know, kind of one of the godfathers of AI at the MIT labs, and then I looked, yeah, there's companies like SentinelOne, Dataiku, Databricks, CrowdStrike, even C3 AI and Tom Siebel. And they existed way before this whole generative AI thing become our language of the day.

Certainly. I think the— by the way, 2 things. One, I found the 6 industry subsegments. It's generative AI, machine learning, predictive analytics, natural language processing, robotic process automation, and intelligence systems. So those are the 6 tags they're using. to analyze this data whose source was Crunchbase. The other thing I'd say there, Ray, is particularly I would say in like finance and security, like credit card fraud and security intrusion detection, those are all very much kind of, I believe it was machine learning or predictive analytics apps where they're looking for things that are unusual, and they've been around for a really long time.

Yeah, for a really long time. In fact, you mentioned this is going back 15 years. And 57,000 financing rounds is covered in this report.

Yes, correct, Ray. So the analysis is based on— let's just do this more formally— Crunchbase for the privates, Capital IQ for the publics. It's global scope. The 6 subsegments are those I just referenced, 57,000 rounds. And to be clear, this is a data analysis, not a survey. Ready to go in, Ray?

I'm ready. Let's dive in.

Okay, so I'm going to give you slide numbers for those following along with the, you know, the PDF version of this report. I am now on slide 4. So number of rounds done, 237 in 2010, up 10x to 2,424 by 2015, doubling again between 2015, or more than doubling between 2015 and 2020, going from 2,424 to 5,582, so 5,582 rounds, peaked in 2021 at 6,798 rounds, and in the first half of 2025 there were 2,700 rounds, which if you annualize that by doubling it, it would be 5,500. So still below peak in terms of number of rounds. What do you think, Ray?

Yeah, I was really surprised when I saw this because it was specific to AI funding. Not overall funding. And then I see the tailing off in '22, '23, and '24. And then I thought, well, the only way that could be true is if the actual value of each of these funding rounds went up because there's so much investing going on in AI right now.

Yeah, which it certainly has. And that's the key here. So number of rounds might be the wrong metric if you're trying to understand this particular bubble, because it's not a bubble of lots and lots and lots of companies raising the same amount of money. It's a bubble where some Very large companies are raising virtually unprecedented amounts of money. So let's look now at total capital raised on slide 6. Uh, $1 billion in 2010, $10 billion in 2015, $37 billion in 2020, a peak of $86 billion in 2021, and we are on track to smash through that peak in 2025, with $110 billion raised in the first half of 2025, annualizing that to $220 billion this year. Thoughts, Ray?

Well, I'm just doing some back-of-the-numbers math, right? So if we do $5,500 this year and $220 billion, does that say it's like $40 to $50 million a round? So that is very high. But I think more importantly, when you see some of the corporate investors and the capital raises, like Microsoft invested, what, $10 billion in OpenAI in 2023. OpenAI recently did a $40 billion venture round. This tells me that the hit rate is going to have to be higher, or the IRR for a lot of these investments are going to be much lower than expected. Does that math kind of make sense?

Yeah. Kind of, kind of-ish. Let's go through it. I think the main point you're raising here, Ray, is correct, which is the amount of money these companies are raising is unprecedented. And it's not because there's a lot of companies all kind of raising similar funding, as software really was, I'd argue, in the past. Oracle didn't raise 100 or 1,000 times more capital than its competitors, right? Maybe 2 or 3. But here we have this kind of arms race And you are seeing rounds that are just unheard of by the past. Microsoft, you said $10 billion in OpenAI. That's a good— by the way, about half of the capital invested in AI in Series A through D came from corporate investors, which is a point you were also making.

Wow.

So, Microsoft dropping $10 billion into OpenAI, Meta dropping $14 billion into Scale AI, kind of a reverse acquihire, I think. I can't remember the details on that one. And then OpenAI's $40 billion round that was, in fact, a venture round, but I guess some corporate investors were in there. Oh, yeah. So, so yes, Ray, the amount of capital being raised is spectacular for reasons that we talked about, that you need to go, you know, buy a bunch of hardware from NVIDIA and, you know, build a bunch of data centers or rent a bunch of data centers to play in this game. In fact, since NVIDIA just announced their financials today, I think you probably saw this, Ray, but I'm going to mess up the numbers. I think their top 3 customers spent Was it $20 billion last quarter? I can't— was it either $8 billion or $20 billion on the top 3 customers? It was some spectacular amount of money that people are spending on NVIDIA.

Well, you know, the other thing this reminded me of, and it really wasn't just how much people are spending on NVIDIA, but when I saw Microsoft invest this much in OpenAI and then Meta investing that much in Scale AI, and you're right, it was kind of a reverse acquihire. It also made me think of the Kind of early 2000s when companies like Oracle invested a lot of money in NetSuite and SAP invested a lot in SuccessFactors. So enterprise on-prem software investing in SaaS companies, it kind of feels like that, but in a much larger bet type of way.

Yeah, it does. And let me just, I'm going to correct myself on the NVIDIA thing. So, so it was just today, NVIDIA reported $46 billion in revenue for fiscal 2Q, of which $41 billion came from their data center business. And of that $41 billion in data center business, just 2 customers contributed an estimated $18 billion. So you had 2 customers who on average contributed $9 billion each to one company in one quarter in purchasing. of hardware to do AI, right? The reason I belabor these numbers is they're just mind-blowing when you slow down, take a breath, and look at them. They're incredible.

Yeah, and like we always say, and the disclaimer says, we don't provide investment advice. But when I see that type of concentration and the build-out we're having here in 2025, I'm not too excited about 2027 and '28.

But yeah, I don't know. I don't wanna make a prediction there, Ray. All I wanna say is it's a very good business to be selling picks and shovels, right? That's the old internet saying. that if you're in a gold rush, you want to sell picks and shovels, um, because everyone buys them whether they strike gold or not. And that is certainly what's going on here with Nvidia. So we're talking about the rounds being bigger. Why don't we go look at median round sizes on slides 7 and 8? And I'm going to arbitrarily compare 2 data points because we're going to throw a lot of numbers at you this episode, so we have to simplify in some ways. And I'm going to simplify by just contrasting 2015, 10 years ago, to today, 2025. Pre-seed median was $130K, now $300K, double-ish plus. Seed was $767K to $2 million, kind of triple-ish. A, $6 million to $12 million. Surprised how low that median is for As, but again, they're including a lot more than just the highly hyped generative AI companies. Series B, $15 to $28 million, doubling. Series C, $20 to $56 million. Again, those are medians by round size. On average, just eyeballing it, I'd say they're all up by double. And obviously medians are going to be far less affected by the size of the monsters. Ray, thoughts here?

You beat me to it, Dave, because I was thinking about some of these really big rounds that we talked about earlier. If we looked at the mean or the average, I bet it's Much higher. We don't have the data to do that calculation, but I'm thinking it's probably more in that $70 to $100 million range.

Yes, it is, Ray. I want to say, well, first, I don't know how much it pulls it up, but second, they certainly would pull it up. And I think the fact they're only— I think the fact that medians have only gone up relatively small, like 2 to 3x, means there's a relatively small number of companies doing truly monster rounds. So it's having a modest effect on the median. And if you had the mean, I think it would be a lot, but I don't have any mean data.

Got you. And I think you also have a slide, I think you call it your money slide, and that's slide 9, right?

Yeah, slide is the money slide, is basically valuations. If there's only one slide to look at the deck, I'd imagine most people would want to look at slide 9, which is AI valuations. And I'm going to run the same comparison as I did last time, comparing 2015 to today. Uh, pre-seed valuations were before and after, $1.3 going to $3.6, seed $3.7 to $10. Series A, $14 to $46 million. Series B, $130 to $367 million. Series C, $250 to $795 million. So again, this kind of double or triple, um, in the trailing 10-year period. Thoughts here, Ray?

I think it's also important to note that that slide 9 was pre-money valuation. And because the size of the rounds went higher, I just think it's important to note that it was— it's pre versus post-money.

Fair enough. Fair enough. Okay, so let's go to slide 11 now and talk about multiples, because sometimes valuations can be misleading, right? Because you need to normalize them in some sense. They're not misleading, but let's just say incomplete. Because to say a company is worth, you know, $1 trillion, is that a lot or a little? I don't know. I need to compare it to something. The thing most people compare it to is revenue, or maybe something like a Rule of 40 score. But what I think we're going to do right now is just look at EV-to-revenue multiples on these people. And that range has gone from 3.6— the range of EV-to-revenue multiples today is from 3.6x to 225x. So I'm not doing the 10 years ago comparison anymore. We're just looking at slide 11 on the range of EV-to-revenue multiples. The top decile is 82.8x, right?

Wow.

The top quartile is 57.5x. The median is 24.2x. And I'm just going to throw some examples at you to make this real. I believe these are all as of their last round. Perplexity, 140x EV to revenue. Hugging Face, 90x. Glean, 72x. Anthropic, 60x. Ciera, cybersecurity company, 60x. OpenAI, 30x. Windsurf, 30x. Six Sense 26, Databricks 24, Jasper AI 22, DataIQ 12, Replit 12. So, Ray, what are your thoughts on this one?

Well, 2 thoughts. Number one, if you just look at the current kind of SaaS and cloud industry and you look at the median EV-to-revenue multiples for public, it's around 5.7. So it's like, wow, way out of context. But then I remember just a few years ago during the Grow at any cost days, 2020, 2021, 20x multiples day for a private funding round were definitely top quartile. And now median is 24.2x in AI. It's just the contrast is stark.

Yeah. So you can almost think of the multiples almost switching over a quartile, right?

Yeah.

What used to be the top quartile is now the median in AI. You got all kind of everybody take one step left as we go the AI world. And yeah, these are very, very high valuations. I even think, just frankly speaking, a typical software company in typical times is valued somewhere between 5 and 10x revenues, right? If you're growing super fast, maybe more than 10x. If you're growing super slow or unprofitable, less than 5x. But that's always kind of my rule of thumb. And obviously, in AI with these first Some of these growth rates are incredible, right? So there's a huge difference between the forward growth rate and the backward growth rate, which accounts for some of this. But some of it is just the bubble, that there's a FOMO, whatever they call it, fear of missing out, and people are willing to pay a lot of money to get into what they perceive as the winners. And I think the other thing that's going on is that if this thing is going to truly be a monster company in the future, whether you bought in at a valuation of $200 million or $300 million really won't change much. to your multiple on your investment. And I think a lot of VCs feel that way right now. But, you know, their job is to invest money. Things are expensive. They can either sit on the sidelines and not invest, or they can pay the— you know, selling to willing— what is it? We're selling to willing buyers at the fair market price, the quote from Margin Call. Yeah.

Well, what I also found interesting is on slide 14 was the foundation model valuations, right, versus like coding assistants. So They seem to be— they really skew that top decile, right? The foundation model.

Yeah. Why don't you go ahead and walk us— or I'll walk through them real quick. So I don't know if they skew anything because I guess to me they're segmented separately, I think. But I don't have the slide in front of me, right? So Anthropic, $60 billion. xAI, $80 billion. OpenAI, $300 billion. So these are truly spectacular valuations. But these people are playing the race to be the platform of the future. And these markets, you know, at very low levels of the tech stack, it typically ends up an oligopoly. There are only 2 or 3 winners and they're worth a ton more money than everybody else. Everybody else kind of gets chewed up and spit out. So it doesn't surprise me that's going on here. The only one that's a little surprising is OpenAI, just because, you know, a lot of the other people are paying, you know, specifically Google and Meta are paying for AI investments out of free cash flow, which is very different than these people, Anthropic and OpenAI in particular, who are doing it using venture capital. X, I assume, is funded partly through Musk directly, but I do think it's also done venture rounds. So these are people raising venture capital to go fight a game where the other people are just basically using free cash flow. Okay, so the conclusions here to make from all this are AI is white hot. It is certainly capital intensive, particularly for the platform companies, the foundation model companies. Training these models costs billions, literally tens of billions. Valuations are high. 2 companies gave— I'm going to say it again. 2 companies gave Nvidia $18 billion last quarter. Last quarter. Just 2 customers, right? Get that through your head. That's why we, we harp on the numbers here, folks, because they're so far outside of anything we've talked about in the past that I just like to get them into my system and go, wow, I need to internalize that. Um, so yeah, this capital-intensive game, valuations are high. The median is 24x revenue. Ray, when I was a kid, they said to me that 15x was a good price-to-earnings ratio, not a price-to-revenue ratio, right? And earnings are, say, let's just say on average 20% if you're lucky. I think it's actually 12% for the S&P 500 today, but to make the math simple, you know, if your PDE is 15 and your profitability is 20%, then that means your price-to-revenue is 5, you know, right? So, uh, if I did the math correct, right? So, um, these are spectacularly high valuations because they look— not only do they look— these price-to-sales ratios, which is what they are, not only look like price-to-earnings ratios, they look like spectacularly high price-to-earnings ratios. And that kind of omits all the COGS and all the OpEx of the business, right? That's the difference between earnings and revenue. So, so these multiples are really, really high. Um, if you look at general now, I mean, public companies, we talked about the S&P 500 is 26 X price-to-peak earnings, I think it was, or price-to-earnings. S&P 500 is 3X price-to-sales. Our price-to-sales ratios sound like price-to-earnings ratios. And that, to me, is always one of these bubble indicators, frankly. But if you want to exploit this as a startup, if you're like, there's always 2 sides of this, like, oh, wow, stocks are expensive. But the other side is, wow, what a great time to sell stock, right, if you're a founder. And if you want to exploit this, the answer is, A little different from the past, which, and I think some founders have not internalized this yet, but it's not just good enough. It's not good enough to have an AI story. You need to have an AI story and some AI results if you want to get an AI valuation, because these companies are building and shipping products and getting to scores, if not $100 million in ARR, and they're doing it in quarters, not years. So, you can't say, I've got a great AI story, And expect to raise money in an AI valuation. You need to have AI results. So that's the biggest takeaway for me. And it's the thing I find myself telling the founders more and more. Other thoughts from you, Ray?

No, I just, if you don't have an AI story and AI numbers, you know, like Jason Lemkin said, 80% of VCs aren't shopping. And if you don't have those types of numbers right now and you don't have a great AI story, it's going to be hard to raise.

Yeah, basically, I think Jason's point was that you're excluding yourself from about 80%. If you don't have an AI story, period, you're excluded from about 80% of the market. It does mean there's 20% of VCs out there and private equity people looking to make such investments, but the vast majority of the market will just not be interested in you if you don't have an AI story. So you have to have an AI story, but if you want to get an AI valuation to go with that AI story, you need to have driven some results. And that's the part that I think is tricky for some founders. But by all means, get working on your AI story. But, but unlike the past, for those of us who've been around for a while, that alone won't be enough.

It won't. And we got to wrap up. Goodbye, everyone.

See ya. The Metrics Brothers AI, SaaS, and More is a production of The Metrics Brothers Growth and CAC and a member of the Benchmarkit Podcast Network. By accessing this podcast, you acknowledge that The Metrics Brothers make no warranty, guarantee, or representation as to the accuracy or sufficiency of the information provided Or the humor content of the jokes told. Ray, information, opinions, and recommendations presented are, according to our spouses, probably wrong and provided for information purposes only. This podcast should not be considered professional or, for that matter, unprofessional advice. We disclaim any and all liability for any direct, indirect, undirect, misdirect, incidental, special, ordinary, consequential, inconsequential, or other damages arising out of any use of or God help you reliance upon. Upon the information presented here. Ray Grothreich is based in New York City and available on X at Ray Reich. Dave Kellogg is based in Silicon Valley and available on X at Kellogg. Schenectady, which is French for unspellable, is not our actual production location. Thanks for listening.