On June 9, 2026, we hosted a live webinar titled “AI M&A, Valuations, and Hot Trends.” The webinar was presented by Marcin Majewski, Managing Director, and Filip Drazdou, M&A Director at Aventis Advisors.
You can now watch the full webinar replay below. If you would like to download the presentation material used during the session, you can easily do so by clicking the download report button on the left (if you are using a computer) or by scrolling at the very end (if you’re using a phone).
Marcin Majewski:
Hello, everybody, and welcome to our AI M&A, Valuations, and Hot Trends webinar. I’m here with Filip, and we’re going to guide you today through our thinking on what’s happening in AI. A lot is happening. Like everyone else, we’re trying to keep up. There’s so much going on that we might even need to adjust the webinar mid-session, because there could be a sudden release of something, or a new IPO filing, while we’re talking.
We try to stay grounded in reality rather than fall into a narrative, whether that’s the gloomy one or the hyper-enthusiastic one. We want to be realistic and balanced, so that you can make up your own mind. So we’ll discuss AI in the public markets, M&A in the private market, AI funding trends, and then we’ll wrap up with a few words on what we think the future will bring, and leave time for Q&A.
Filip Drazdou:
No notes from me, it’s a good plan. Let’s jump right into it.
Is AI a Bubble? What the Shiller PE Ratio Tells Us
Marcin Majewski:
The first thing that comes to mind when you think about AI and the financial markets is the B-word: bubble. So we wanted to look at the historical trends, and here we present what I think is the longest time series available for financial markets, which is the Shiller cyclically adjusted PE ratio.
A very quick explanation. PE is the price-to-earnings ratio, here for the broad equities market in the US. It is cyclically adjusted because the earnings figure is a 10-year, inflation-adjusted average.
So it doesn’t show you the current sentiment of the market; it shows you how the market trades on a longer-term basis. Does it recognize value fairly, or does it overpay or underpay relative to those readjusted earnings?
I like to start here because it gives you a very long-term perspective, one that goes beyond any single person’s lifespan. It shows that gyrations are common, and that there have always been up cycles and down cycles. We shouldn’t expect that this time is different.
These upswings tend to run long. There was the cycle that ended with the crash of 1929. Then the long run from the Great Depression to the end of the 1960s, roughly 30 years. Then the climb through the 1980s and 1990s until the dot-com peak in 2000 and 2001.
The current cycle started around 2009, and we’ve apparently been riding this wave for 17 or 18 years now. So how long can it go on? It’s definitely a bubble; on that I have no doubt, because bubbles always occur. But honestly, does it last 10 more years, or two? That’s very hard to predict. We can have our bull-bear discussion here.
Filip Drazdou:
Not 10 more years, for me. When I look at how long this ratio stayed above 30, it wasn’t for long. It went above 35 at the peak of the dot-com bubble, but only for a couple of years. So as we enter that zone now, I’d say the maximum is two years or so, just looking at the historical patterns.
That said, no one says it shouldn’t go to 50 or 60. It peaked around 32 in 1929, but the dot-com top went much higher. It might go to 50 or 60 easily.
Marcin Majewski:
Until we started preparing this, I thought the same. But now I think a technology this revolutionary can drive the narrative for many years to come. It’s a very good narrative. So my view is that the market is overvalued, but it might keep going for the next 10 years. It’s so hard to predict.
Filip Drazdou:
And if you add a couple of trillion dollars of market cap from SpaceX, OpenAI, and others coming to the public markets, the picture shifts again.
Marcin Majewski:
It gets even more overvalued, because their earnings are low, so they automatically push this chart up. We’ll be moving into record territory, probably in the near future.
Filip Drazdou:
The other way some people look at it is by asking what your next 10-year return is if you buy at a given price-to-earnings multiple, or a given Shiller PE. Over the next 10 years, the return from here is not going to be great. So if someone is in the market for the long term, the historical data says now is not the best time to buy.
Marcin Majewski:
Definitely. But then again, it’s very hard not to buy into the narrative. It’s such a good narrative to resist. Call it the fear of missing out.
What’s Driving the AI Rally: Inside the AI Supply Chain
Filip Drazdou:
Let’s zoom in on what actually drives this.
Marcin Majewski:
Here we have two lines. One is the S&P 500, the broad market index. The other is our proprietary, equal-weighted AI index. We’ve put both on a log scale so it’s easier to compare relative moves. What you see is a pattern similar to the early 2000s.
We tracked these AI companies back to 1995, by which I mean what we now understand as the AI supply chain engine, and they also boomed back then on an infrastructure build-out. At that time it was telecom and fiber in particular, and networking infrastructure. A similar entangling started recently, around 2022 and 2023, when our AI index began to move faster than the S&P 500. I don’t think this is complete; I don’t think we’ve seen all the movement we will see. I expect even more exponential growth in the AI stocks. Filip, your thoughts?
Filip Drazdou:
Looking at this, I was surprised by how many of the same companies we added to our AI index were listed back in 2000 and 2001, and went through the same cycles, networking, servers, electronics, manufacturing services, construction. A lot of them are still around. After that bubble, they’re now having a revival on the current AI capex. The story is the same: there’s just so much capex that the whole supply chain benefits.
The way we built our index is that we looked at one big event when the markets dropped a lot, the release of DeepSeek. That was when everyone suddenly thought AI was going to be 10 times cheaper and you wouldn’t need all those billions of dollars of investment. A lot of companies across the value chain dropped that day, so we collected and analyzed them, sorted them into categories, and looked at how they performed over the past couple of years.
The best-performing subsector was data center EMS, the electronics manufacturing services companies you can outsource your electronics assembly to. Those went up by more than 800%.
There was a big multiple expansion: before, these were boring companies trading at 5 times EBITDA; now they’re seen as picks-and-shovels suppliers, so they trade at around 25 times EBITDA, and their earnings also increased, which is where the rest of the price growth came from. Then you have all kinds of suppliers of equipment and hardware for data centers, from HVAC to electrical equipment to chips, plus the semiconductor supply chain and the companies that construct all of it.
Looking at this, basically any sector related to AI benefited over the past three years. The difference versus the broad market was massive: our overall AI index returned around 550%, the S&P 500 around 190%.
Marcin Majewski:
Amazing. And these used to be boring businesses.
Filip Drazdou:
They’re high-capex, low-margin businesses, now priced quite high. We’ll see how that goes. If this is a bubble, then those have to come down, and it becomes a cycle. A lot of these are very cyclical sectors, where you can have big earnings for a while and then a sudden drop. You over-invest in capex, you overbuild the factories to produce HVAC, and if something bad happens to that capex, you’re left with overcapacity.
That’s also why the Shiller PE ratio is so useful: it looks at 10 years of earnings, so it removes the effect of the cycle. Even if you triple your earnings in one particular year, the assumption is that what went up must come down, and every industry runs on its own cycle. So if you’re an electrical equipment or HVAC manufacturer whose order book triples now, it won’t be forever, and eventually someone is left with all the capacity the company invested in.
Marcin Majewski:
Yes, the build-out won’t last forever; it’s largely a one-off. You do have to replace the GPUs, but I don’t think we’ll build out all of the Earth’s data centers. I hope not.
Filip Drazdou:
You have to replace the equipment every 5 or 10 years, so there’s some useful life there, but it’s not the same form factor as recurring revenue.
Marcin Majewski:
It’s not recurring revenue, no.
Ten Signs of a Market Bubble, and Where AI Stands
Filip Drazdou:
On the bubble question, we also looked at the elements that Jeremy Grantham and Edward Chancellor identified in one of their recent papers. They list 10 markers of a bubble, and you can decide for yourself whether AI ticks them:
- Important technological innovation: AI now, and telecoms and the internet before. Check.
- A simple, compelling story: this time, that we’ll never have to work again.
- High-profile champions: Thomas Edison during the electricity boom, the dot-com founders like Bezos and Bill Gates. Now we have Sam Altman and Dario Amodei. Similar pattern.
- A wave of new publications promoting the merits: there are AI-focused magazines and media, and the coverage is extensive.
- A growing supply of new companies: the data shows business starts on the rise in the US, not just in tech but across different spaces.
- Suspension of normal valuation criteria: SpaceX is a great example.
- Application to even mature technology: this is a tough one, but you can argue we’re still early.
- A huge overcommitment of capital: the capex is in the hundreds of billions of dollars.
- The revenue you’d need to cover that capex and earn a return is very high.
- Frauds and fraudulent behavior: we haven’t really seen that one yet. That’s usually the last to appear; last time it was Enron and WorldCom, and during the financial crisis it was the banks. The shakeout is when it all starts going down, and that’s when a bubble is confirmed.
Marcin Majewski:
On the simple, compelling story, the version this time is that robots, or agents, will do everything. On the “even mature technology” point, I’d push back a little: I feel like we work with an MVP constantly. We use a lot of AI, but it’s definitely not the end state. For regular people doing mission-critical work, it’s just not there yet, and it will take a while.
So I can track maybe six or seven of these boxes, not all of them. That’s why I’m not concerned the bursting is coming anytime soon. I think it still has to grow a little more.
I’m bullish short term. Long term it’s going to burst, but I don’t know, maybe in two or three years. I think the music will keep playing for a while.
Filip Drazdou:
The big one to watch will be the IPOs of SpaceX, and then OpenAI and Anthropic, to see how they perform. My guess is the first one pops 50% on day one. That tends to leave a lot of room to run and invites a bunch of other companies to go public. But if it starts deteriorating, or the stock drops over the two weeks after the IPO, then we might be closer to the top.
What we don’t have here, which wasn’t on the checklist, is the froth we had in 2020 and 2021: meme stocks, SPACs, NFTs, crypto, the metaverse.
Marcin Majewski:
There isn’t that much of it now. We have SpaceX, and Tesla at 300 times earnings, but there isn’t much truly crazy stuff. Even Bitcoin is fairly down to earth. So I think we still have quite a bit to go.
Filip Drazdou:
To put the supply chain together: it starts from energy supply, electrical generation, nuclear, oil and gas. Then it runs through semiconductors and all the equipment suppliers to semis, the software they need, test and inspection, yield equipment, and assembly.
Then you have what’s actually inside a data center: electrical equipment, HVAC, networking, chips. And then the companies that operate all of it and construct it. The next step up is LLMs and AI apps.
Marcin Majewski:
Those I’ll add to the index soon.
Filip Drazdou:
We’ll add them as soon as they list, and that will be a very good validation proxy. We’ll finally see how the leading, top AI companies are actually valued.
Marcin Majewski:
It’s fascinating that the biggest winners of the AI race so far are not AI companies.
Filip Drazdou:
We could actually calculate how much market cap has been created across the chain. It may be more than the OpenAI and Anthropic valuations combined. NVIDIA alone is a $5 trillion company. That’s a big one.
AI M&A Activity: Deal Volume and Buyers Since 2015
Marcin Majewski:
M&A has actually been great in the AI world. We crunched the numbers on deal volume since 2015, and we were quite surprised: it seems AI M&A activity peaked back in 2020, and it probably won’t hit that record this year. There was a huge acceleration of dealmaking in 2020, which I think was the culmination of the first cohort of AI companies maturing.
That was before we talked about LLMs and generative AI; it was more machine learning in the traditional sense. That wave culminated in 2020, then decelerated through the COVID years and the subsequent collapse in valuations.
It has been on the rise since 2023, but the year-to-date figures don’t look that strong. It doesn’t seem like it will be a massive volume, and I think that’s driven mainly by less interest from financial buyers. There is, however, a new wave we’ll touch on, of the frontier labs themselves going the M&A route in their go-to-market motion.
On the structure of M&A activity in AI, the biggest group by far is developer tools, which are the foundation for AI. Then there’s a huge diversity of AI applications getting acquired. Machine learning peaked at one point, that’s the purple bar at the bottom, and was then supplanted by other types of applications.
Honestly, the diversity is huge; AI has a lot of use cases, and it would be hard to say which is in more demand from buyers and which isn’t. My sense is that if you make it in AI, the business is quite sellable. Anything to add, Filip?
Filip Drazdou:
In 2020, a big part of it was the macro environment and extremely low interest rates. The financial acquirers’ bar was very high, and almost all of the decline since then has come from financial investors, while strategics kept investing and increased gradually.
If you take the 2026 year-to-date figure and double it, it’s still going to be fewer deals, and again, that’s mainly financial buyers slowing down, being very skeptical toward acquisitions in general.
AI Valuations in the Private Market
Marcin Majewski:
Here’s a small sample on AI valuations. Unfortunately, most of these deals don’t get disclosed, and when they do, we don’t get reliable underlying data. The headline number gets disclosed, but not the underlying revenue or EBITDA.
From the sample we analyzed, though, we were not shocked. The valuations of AI companies have been more or less in line with what we’ve seen in other software sectors, software and IT services, maybe a slight premium, but nothing eye-popping. The private markets are staying mostly sane. We don’t see the eye-watering valuations from the funding rounds. So it’s quite grounded: around 5.4 times EV/Revenue and 15 times EV/EBITDA. That’s a premium valuation, but it isn’t shocking.
It’s a small sample, and we don’t capture all the loss-making, extremely high-growth companies that are obviously off this scale. We’re researching it, and with more data we’ll have more conclusive findings, but right now this is just a small sample and a few observations. For the broader picture of how these businesses are priced, our AI valuation multiples analysis goes deeper.
Filip Drazdou:
There’s also a very big diversity in the multiples and valuations. You get hundreds of millions for a company of just a couple of people, like we’ve seen with Wix, where it was a couple of guys, or even one person. And then there are a lot of AI wrappers, companies that just built a quick product around a model, and those aren’t sellable at all.
Marcin Majewski:
Or they sell early on at a low multiple, because they don’t see a future in the business.
Filip Drazdou:
So it depends.
Marcin Majewski:
Every business is different, so there’s no one-size-fits-all multiple. As you saw in the previous chart, the use cases for AI are so diverse that it really depends on each individual business.
Filip Drazdou:
It was a bit easier in SaaS, where you just have the KPIs: rule of 40, retention, and so on. You know where to put the multiple. Here it’s much more nuanced in terms of the technology. The question you always have to ask is whether a frontier lab could release the same product tomorrow. (For the contrast, our SaaS valuation multiples work shows how much cleaner that exercise is.)
Marcin Majewski:
And it’s not really a question of whether they will; that’s almost a certainty. The question is whether you can pivot fast enough. We’ve seen good pivots, like Jasper, who did a great job. But not everyone can handle it. In this business you have to be able to switch very easily.
How LLMs Are Moving Into IT Services M&A
Filip Drazdou:
The newest development in AI M&A is that the LLMs are getting into IT services. Within the span of a couple of months, OpenAI and Anthropic both announced they’re developing their own implementation arms. They’ve taken funding from private equity funds, struck partnerships with some of the leading consulting firms, and then made acquisitions.
OpenAI acquired a London-based company, which we estimate is around £20 million in revenue, has grown very fast, and has about 150 people. Anthropic acquired a similar company out of San Francisco, also growing more than 100%, with around 80 to 100 people.
What’s in it for founders and investors? First, I think they’ll continue this acquisition spree and buy more companies to build out their implementation arms. They’ve understood that it’s not enough just to have the models; you need people who go to the customers and show them how to use them. I also think some private equity funds will start creating roll-ups, or roll-up vehicles, to buy companies in that space and eventually sell to a bigger player like OpenAI or Anthropic.
So this is a great opportunity for IT services founders thinking about a good exit over the next two to five years. AI services is the thing to develop, both vendors are starting partner programs, and that’s where the growth is.
On top of that, traditional consultancies like Accenture are buying a lot of AI-native services companies. Faculty AI, for example, was acquired for around a billion dollars a couple of months ago, with only 400 people and a reported revenue multiple of about 15 times, which is unheard of in the IT services space. But here you’re paying to get the capability and the founder, who is now CTO of Accenture, and I think that was one of the major reasons for the deal.
Previously we talked more about M&A in the software industry and what’s happening for software companies doing AI. These are some examples from IT services M&A, which is also starting to grow. For where the underlying multiples sit, our IT services valuation multiples page tracks that in detail. Should we talk about funding for a bit?
AI Funding Trends and the Overinvestment Problem
Marcin Majewski:
Quickly, yes. I like the comparison to IPO volume. We’ve had five big IPOs and secondary offerings, including from Google and Meta, and the numbers are staggering when you look at the amount of capital raised through those issuances. If you then look at funding year-to-date for AI companies, it’s around $300 billion, more than all of last year’s AI funding combined.
I think it’s a sign that we’re in a bubble, and it confirms that this build-out isn’t sustainable. The money has to come from somewhere. These companies don’t produce a profit, and they’ll need to invest for years to come. It makes me wonder whether the moment it pops comes sooner, but it’s genuinely hard to say how long it can go on. You can see the full landscape of who is writing these checks in our overview of the top AI investors.
Filip Drazdou:
It really adds to the point about overinvestment in infrastructure. Everyone is investing, and they’re in a bit of a prisoner’s dilemma: they can’t afford not to invest. If you don’t invest and your rivals do, you lose. So eventually everyone is worse off, everyone is over-investing, and no one can escape it.
Who Wins After the AI Bubble: A Contrarian View on Where the Value Goes
Marcin Majewski:
We also thought about where the world is heading, and who benefits after the dust settles once the bubble collapses. We agree AI will bring productivity boosts, for sure. There will be less labor needed to get the same things done, so people will get more done and become more productive, and that productivity will create new things in manufacturing, in biotechnology, in services. Everything should improve thanks to the new technology.
In general, what you might call the San Francisco consensus is that labor will be replaced by AI. We wanted to ask how that translates into who actually makes the money. Our interpretation of the San Francisco consensus is that the share of GDP that goes to labor, to wage earners, shrinks: less money flows to humans, more to corporates (hence the bump in valuations), and a little more to the owners of physical assets.
Then we asked ourselves how it would look in reality, and this is our contrarian view. There will be a large surplus to share in the economy thanks to the productivity gains, but the distribution will not necessarily be what’s expected.
Take companies and corporates first. The big rise in valuations rests on the predominant narrative that profits will go up across the board; otherwise, why pay more for shares? But I’m not so sure. Businesses that used to be asset-light, because they ran on ideas and IP, will suddenly turn into capex-heavy businesses. Barriers to entry will actually drop, so industries that used to be monopolistic may turn hyper-competitive. I would be very afraid of corporate profits in the long run.
At the same time, the GDP that accrues to the owners of physical assets, land, data centers, manufacturing facilities, producers of minerals including rare earths, is where I think the value from AI ends up. We’re already seeing it: NVIDIA is doing great, but the companies doing the physical things that underlie this whole build-out are the ones making a ton of money.
So this is just to illustrate that the current consensus might not be right, and it’s worth looking at the assets people aren’t currently appreciating. On a horizon beyond 20 years, those may be the ones that actually appreciate, and the most value might not accrue to the winners of the AI race. That’s a bit of a digression, some food for thought, but I hope it triggers some thinking.
Filip Drazdou:
As we showed at the beginning, the benefits go to all the companies doing the construction, the hardware, the equipment, and so on. They’re the choke points right now; the bottlenecks are there. The money is, or will be, available for all of those companies, and they’ll overbid each other and drive prices up all the way down the chain.
We also have some example valuations of AI companies here, but those are funding rounds, so most of the time that’s a different picture.
Marcin Majewski:
It’s interesting to see how they’re compressing. The companies have grown so much that they don’t look so crazy here, to be honest. Or maybe we’ve just gotten used to very high multiples. But they’re definitely down, because the companies have grown since last year.
So that’s it. We arrived more or less on time.
Audience Q&A
On M&A of vertical AI solutions:
Marcin Majewski:
I have mixed feelings. On one hand, vertical software has been considered a safe bet so far, so by extension vertical AI should be the same. But Claude is getting so smart, and you can train it to do vertical tasks so quickly and easily, that I’m not sure how much sustainability vertical AI solutions have, unless they’re tied to a data source that’s very specific to that vertical.
If you have something for the maritime industry and you come up with a proprietary sensor you can bundle in to provide intelligence, then perhaps. But in general, I’m not that optimistic.
Filip Drazdou:
It also depends on the vertical, because every vertical is different. The maritime software example, or something for science, is very niche and specific, and that may be less subject to disruption by AI. But if the vertical is something simpler, like software for schools or education, where there are hundreds of players across different countries and AI already has the code base and the knowledge, then it’s much more exposed.
On whether productivity gains require people, not just technology:
Marcin Majewski:
I fully agree that productivity as an AI benefit requires real changes in people’s behaviors and processes, and that’s something only people can drive. We talk to a lot of companies and see very different behaviors. Right now it’s a competitive edge for the companies that embrace AI: they grow faster and deliver quicker. We see that mainly in software engineering, because engineers can move much faster with AI tools. It’s not 10x, but it’s real.
It requires people to change, and it requires sponsorship from leadership to drive that change. It won’t happen by itself. Governments and local governments will also need to help facilitate it. We’re from Europe, and on something like self-driving cars Europe is around five years behind San Francisco. Without that support it doesn’t happen. It has to be a concerted effort.
On the AI equivalent of the physical winners in telecoms (the tower companies, Nokia, Ericsson):
Marcin Majewski:
Brilliant question. The AI tower company is going to be the data center company, or the energy company, and it will probably be bundled: energy generation together with data centers, because you need to build them close to cheap energy sources. Those will be the ones that benefit, plus maybe some mining, because you need a lot of rare earths to drive the robotization of the economy. In general, metals, copper, steel, and aluminum are all up and coming.
Filip Drazdou:
For me, it’s also the grid, the distribution. That’s where the bottleneck is, and where the monopolies actually are. In most countries you can’t easily invest in those; they’re regulated. Anywhere there’s monopoly power or some kind of natural limitation.
Marcin Majewski:
And we should mention China. China is going to be a massive beneficiary, because for all the infrastructure build-out they’ll be the ones manufacturing the components. That’s what happened in telecoms too: Nokia and Ericsson were replaced by Chinese vendors, and I think it plays out very similarly here. Europe is actually strong in these old-school electrical engineering technologies, which is encouraging.
On the patterns we see in the buyer universe (private equity, Constellation, corporates):
Marcin Majewski:
Overall, we see a lot of skepticism. The Constellation-type companies aren’t buying high-flying companies, and all in all we don’t see that much adjustment in acquirers’ strategies. Very few people are making bold moves. Accenture is a good example of one that has. We scan AI companies constantly, and not that many get acquired. A lot of buyers are taking it slow and trying to figure out where the market is heading.
Filip Drazdou:
Even in IT services, it’s not as if everyone is snapping up AI companies left and right. There aren’t that many AI-native IT services companies to begin with. On the buyer side, OpenAI, Anthropic, and Accenture are active as of now, while the others are in wait-and-see mode, protecting their core business.
On whether there’s a people-oriented opportunity space, given that billions of people will still buy something even as many struggle:
Marcin Majewski:
We asked ourselves this today and looked at the structure of GDP. Honestly, the whole tech sector isn’t that big as a share of GDP. The backstop is the government, because they can tax everyone and run public projects and hire people, the way they did in the Great Depression. That’s the last resort if people really have nothing to do. But there are so many industries that will still need people, like hospitality.
Filip Drazdou:
Healthcare has grown like crazy.
Marcin Majewski:
Care services, yes. You want to interact with people. You don’t want to buy your coffee from, or talk about wine with, a robot. You want someone with the human experience. So I’m not that concerned. There will be a transformation. I’m from Poland, and I saw this when the Polish economy transformed from communism to capitalism.
All of a sudden, 25% of people found themselves unemployed, because the skills they’d been trained in, heavy industry, managing state-owned businesses, administration, were completely inapplicable to a modern, globalized economy. So there might be a gap. We struggle with hiring juniors too, because it’s so much easier to have AI do something than to hand it to a junior person.
There will be a struggle, but I think we’ll emerge with a new economic structure that’s even more service-oriented than today, and the type of services will differ.
I think a lot more of it will be empathy-based services, where you want emotional value out of the interaction, not only functional value, because the functional value will be commoditized and gradually go to zero. So maybe we’ll seek higher things in our interactions. That’s the positive view.
To close, we put together two scenarios, a utopian one and a dystopian one. We don’t know where it all lands, and it will probably oscillate between the two in the coming years. I’d love it to be the utopian version, though I’m sometimes afraid it goes the dystopian route. Let me finish on the utopian note. Markets and valuations do well, and we make a lot of money, because the capex benefits the broader economy, not only the owners of Anthropic and OpenAI. We work less and find new jobs.
I keep asking my team to fire me and replace me with agents; it’s not happening yet, but maybe one day we’ll work two days a week and enjoy the work we do. Civilization gets to focus more on creativity and connection with others. And maybe AI even brings governments and international organizations together toward a new world order that benefits everyone and prevents further conflict. Probably not all of it happens, but I hope a lot of it comes true.
If you are considering any type of M&A transaction in AI or the broader software space, whether as a buyer, seller, or investor, we would encourage you to get in touch and speak with us directly. You can also explore our latest data on AI valuation multiples for more context on how these businesses are being priced.
