# AI CapEx Boom and Valuation Risks

**Podcast:** FT Tech Tonic
**Published:** 2026-09-16

## Transcript

Hello, I'm Tim Bradshaw, the FT's global tech correspondent.
Tectonic will be back soon with brand new episodes.
But in the meantime, here's a highlight from the recent FT Weekend Festival.
It's me debating what might be the most important question in the financial world right now.
Is there an AI bubble and when will it burst?
I spoke to Richard Kramer, a former top-rated analyst at Goldman Sachs and founder of Arete Research, and to Ophelia Brown, founder of Blossom Capital.
one of London's top early-stage investors.
We talked about the massive amounts of money pouring into AI right now, the real value of companies like OpenAI and Anthropic, and whether there's ever been a more exciting and confusing time to be a tech investor.
I started by asking Richard Kramer to give me a sense of the importance of tech and AI in particular to the wider financial markets.
I mean, the S&P right now is about $65 trillion.
And about 40% of that, about $27 trillion, is tech.
You know, we have a vast, an unprecedented aggregation of resources at the top of the market.
Seven of the most, 10 to 10 most profitable companies in the world are tech companies.
And next year, that will likely be nine of the 10 most profitable companies in the world are tech companies.
Three years ago, I was here with John Thornhill and mentioned that the tech companies had about a trillion and a half dollars in sales.
If you throw NVIDIA into the addition of Google Meta, Amazon, Apple, and Microsoft, that's three trillion dollars of sales.
They generated 400 billion dollars of cash flow.
So we're talking real money, I think, at some point.
And that's actually going down this year to $175 billion, but $200 billion of that is showing up at NVIDIA.
So this AI investment is kind of single-handedly propping up the U.S.
economy right now.
$700 billion on CapEx, that'll be a trillion next year.
And spending $300 billion...
A trillion on CapEx next year.
A trillion on CapEx next year, spending about $380 billion on R&D.
So that's kind of where we are in terms of scale.
Okay.
And on the private markets, what does the Arabium look like from your perspective?
It's the end.
From a private markets perspective, I think every participant considers this to be the biggest technology revolution.
of our lifetime and our generation.
And commensurate to the size of investment on the public market side, a trillion into CapEx, 500 billion went into two companies in the last year, OpenAI and Anthropic.
And that makes up about 50% of total investment globally into AI companies.
So in terms of where we stand, I think most would say that we're still relatively early in this boom period of technological innovation.
I think what's...
fascinating is when we do a show of hands of who's got access to gbt and claude is that to put it in perspective i think there are 1.5 to 1.7 billion iphone users but open ai alone has a billion of monthly actives each month so the distribution and the spread already amongst consumers is very high but on the enterprise side we're still relatively early yeah and still early in in crystallizing some of that uh VC investment in the public markets as well, which we will kind of certainly get to.
But so, I mean, you spend a lot of time in San Francisco, in the Valley.
Is there talk of this or fear of this being a bubble there?
Or is that just something that those of us who are outside looking in worry about?
I think it's more of an outsider's perspective.
I think I came back this week off my 20th trip.
to San Francisco this year, and I can certainly say that no participant there is thinking of this as a bubble.
I think they really are thinking of it as a technological innovation that's going to unlock so many pieces of our life and so many sectors that have, you know, especially when you come to think about the physical economy that hasn't been touched by even the software upgrade.
And I don't think it wouldn't be socially acceptable to call this a bubble.
Okay, interesting.
Richard?
What do you think on that?
I mean, you've seen a few of these cycles in your career in various guises.
Does this feel like we're reaching a crescendo?
I think in the markets, to paraphrase West Ham United, we're forever blowing bubbles.
We're always going to be having these explosions of enthusiasm.
And I think you have to distinguish between the investment that's happening and the valuations that are happening.
On the investment side, I said you'll see big tech spend a trillion dollars in CapEx.
They're hoarding resources.
They are doing this because they can.
Amazon, Google, and Microsoft have a billion and a half dollar, or trillion and a half dollar, sorry, order book for their cloud businesses.
And this is going to create a huge value transfer from the internet industry to the semiconductor industry.
So that's all great on the valuation, on the sort of investment side.
On the valuation side, there is right now a degree of consensual hallucination.
of what I call an extrapolationist fantasy.
As an example, when SpaceX went public, it said it had a $26 trillion total addressable market of all the software applications in the world.
Now, Anthropic is going to go public mentioning a $30 trillion total addressable market, coincidentally also all the software applications in the world.
But of course, you have the Microsofts and the Oracles and the Salesforces and all those other companies that are sort of embedded there already.
So it can't be that everyone prospers in a world of 2% GDP growth.
And I think, you know, one of the issues that's feeding this bubble talk right now is that, as we've talked about before in this panel's...
You have the sycophants and stenographers, the analysts, who are basically cheerleading these companies.
It's all about the growth rate.
These are the Wall Street bank analysts.
Congratulations on a great quarter.
And how should we think about AI?
And they're allowing the companies to fill in the narrative.
And it won't work that all companies will win.
There won't be a one rising tide that lifts all boats.
And I don't think we've started the competitive phase of what we're going to see in AI.
Richard, I mean, it's very easy to kind of look at...
SpaceX valuation or the kind of mooted multi-trillion dollar valuations for an Anthropic and an OpenAI when they eventually go public or maybe sooner rather than later in Anthropics' case.
But is that, like, to a Philly's point, like, you can say these things are a collective hallucination, but that's not the same as calling the top.
Like, it feels as though we at the FT have been having this debate with our readers for some time, and there has always been, you know, there was skepticism in 2024 that this couldn't...
possibly go any higher, and yet here we are, trillions of dollars later.
Yeah, and I think what I would have to say is that equity markets are deeply unmoored from traditional intrinsic valuation right now.
We are not back in the days of Buffett and Munger suggesting that a company's cash flows should be valued irrespective of what the company does.
It's all vibes.
And, well, it's not all vibes, but, you know, we've never seen a company like SpaceX, which is worth, worth, two trillion dollars.
It's getting at 100 times sales.
That market cap is...
Now, there's some caveats here.
One is you actually floated only 5% or 6% of the company, so it isn't actually worth...
That doesn't reflect what investors have put into the company.
And this is a stock that's been up at 200, it went down to 100, it's back at 150.
This is a speculative investment vehicle.
And...
The thing that's enabled this, when we step back from the public markets, is that 75% or 80% of the trading is algorithms and market-neutral funds.
And they say, hey, the company beat the quarter.
What do they do?
They're not interested in the company itself.
They're interested in using that market cap as a way to make money in what are fairly speculative investments.
And when you discount 20, 30, how many data centers are going to be in space, it is fairly speculative compared to your traditional retailer or chemical company or bank.
Sure.
But there are so few opportunities to buy into that vision of the future if you believe it.
I guess that's why that's sort of pushing up those kind of prices.
I don't know.
You're an investor.
I mean, I don't want to jump straight to Orbital Day centers, but it is one of my favorite topics because it's just so brilliantly sci-fi.
You're an investor in a company called Cowboy Space founded by...
one of the Robin Hood co-founders that's trying to do this.
Why do we need to put data centers in space?
Is that something that the AI boom really can enable?
I think that the bull case for this and the investment case for putting data centers in space is that we don't have enough data centers on Earth that are going to be able to power the CapEx cycle for what we want to achieve with AI.
And I think if you think that AI is an inevitability and it's going to pervade every aspect of our life, then we need to find somewhere else to put these data centers.
The science behind it, that we can use robotics and get things into space, that's totally proven.
That's fine.
So then the question is, how do we get the electricity back and power all these things?
And that's what Cowboy Space is working on at a cost point, and this is the most important point, at a cost point per GPU that we can do on Earth.
That's the point that...
is unproven at this point, and that's where the investment is going.
But is there a need for it?
Yes.
Even with a trillion dollars going into the ground already to time build?
Even with a trillion dollars going into the ground.
And that's seven trillion by 2030.
Okay.
And so where does that leave?
I mean, given this somewhat febrile environment, where do you think that leaves Anthropics prospective valuation?
Or if you don't want to speak to kind of specific...
let's kind of swap in Anthropic or OpenAI.
The market has still not seen a pure play AI model company listing.
We've got Google, which is an advertising business with a model on the side.
You've got NVIDIA, which is just kind of soaking up all of this capex.
I mean, saying these are trillion dollar companies is lowballing it, isn't it?
Yeah, I mean, I completely accept that there is a speculative frenzy right now, and that's being fueled by...
endless discussion.
I mean, it used to be three years ago, it just started as cocktail party chat, and now it's impossible to ignore.
The AI boom, and these are the companies that are the pure plays.
What's going to be fascinating to see is when these companies list their filings, just how much money not only do they lose, but how much money are they planning to lose, which is, again, a big change from what we saw in public equity markets for the last 100 years.
You used to have to have three years of profitable trading history before you could go public.
Now it's a question of who has the largest losses.
So I think there's going to be a huge interest in these IPOs and especially huge interest in picking through the numbers.
in a world of off-balance sheet debt and circular financing and where does that money come from to see if these companies can execute on that vision.
I think it's fair to say that we also haven't seen the scale.
in this cycle that we've ever seen before.
So to the point that I made earlier about iPhone penetration, it took Apple 13 years to get to 1.5, 1.7.
It took OpenAI less than three years to get to 1 billion users.
So when you're already at that scale and you need to path the way forward in terms of growth assumptions that you're making, it's not the biggest leap that you have to make.
That's true, but what I would say is right now OpenAI can claim a billion users, but what portion are they monetizing?
How many are subscribing?
And what Apple in the billion and a half or two billion or two and a half billion iOS device user base, they are clearly, they built a hundred billion dollar services business out of that.
So they're monetizing all of those users.
Philia, you're not an investor in Anthropic or OpenAI.
I mean, would you buy stock in an IPO at a $2 trillion valuation for Anthropic?
Don't hold me to investment recommendations.
All right, well, okay.
I can tell you something else.
Polymarket, which is one of the biggest prediction markets.
Sure.
over 75% of their users thinks the IPO ends up in a day and would be buyers into the IPO.
So I think that tells you something about retail demand.
Yeah.
The stock.
One of the things that, as Richard said, is going to be very interesting when we actually see these filings, which I think in Anthropik's case is hopefully going to be this month.
Do you think they're still kind of growing at the same pace that they were?
Because there was the kind of sense that...
Anthropic, I think, came out with 60 billion ARR, was the latest kind of word, or 64, which is an extraordinary number for a company that was barely making any money two or three years ago.
But the sense was that around some of the kind of VC water coolers, they were actually looking for more like 70 or 80, and that's, if you're trying to draw that curve, it's actually already slightly slowing down.
I don't know, like, are we still...
on that same kind of exponential trajectory?
I mean, with the IPO upcoming, it literally is the trillion dollar question.
But I think there's also so many unknowns around this sector.
I mean, where does regulation come into play?
What's allowed in terms of model development?
What do we need in terms of model development?
So I think it's not just a question of the financial performance today and projections 12 months out.
I mean, again, when we say we're early in the industry, there's a lot more other factors that we're going to have to consider.
When you look at, I mean, the regulation question is very live with all of these instances of OpenAI's agents kind of running over, hugging face, and others, it is emerging.
Like, is that, you know, a great proof of the capability of these things?
And you kind of see the, you see these kind of stories of agents kind of creating their own message boards and coordinating with each other.
Is that like, you know, for...
for maybe more of Sam Oldman than a Dario Amodi, like, is this a great proof that these things are capable?
Or do you think this is actually something that invites regulation and could kind of stop the party?
I think certainly regulation is a big theme, a big topic here.
And in terms of the industry participants, I think most would say that there needs to be some kind of regulation.
I mean, already, US government is having access to models earlier to be able to say what they can and cannot release.
And that's critical.
I think many...
industry participants fully believe in AGI by 2030 in terms of how they see development going.
And it's a question again of what CapEx is going to enable that investment and that activity.
You know, the biggest kind of ones that are concerned wouldn't hesitate to compare this to the world and development of nuclear.
So that's kind of how extreme you could consider this.
Yeah.
Richard, do you feel like there's a kind of sense of...
AI's capabilities topping out.
I mean, that was the big kind of discussion around NVIDIA's stock price a year or so or two ago was that, you know, actually the kind of scaling law that, you know, more is more intelligent, you know, more chips is more intelligence was breaking down and we had the kind of deep seek panic that, you know, you could do this on tough and tapenny from Huawei even.
That conversation seems to have got a lot quieter recently.
Yeah, look, I think there is a framing of AI inevitability, which has now become gospel.
That said, I think you just pointed out one of the many quite brutal unintended consequences that if you're not being served Kool-Aid at breakfast, lunch, and dinner, you might be concerned about.
Hugging face, as an example.
OpenAI claims it didn't know that it was hacking this company.
I mean, imagine the staff at 700 Tesco's walked into Sainsbury's and started chucking jars off the floor.
I mean, this is a corporate governance failing, first and foremost.
And it also explains why two stocks, Palo Alto Networks and CrowdStrike, have half a trillion dollars in cybersecurity market cap.
Because we're going to have to protect against this stuff.
I personally was mortified by the US administration weighing in this week saying that OpenAI doesn't need or require licensing to use copyrighted materials to train its models.
This is when they intervened in the New York Times copyright.
Yeah, and hats off to Baroness Kidron, who's speaking later, that she's fought the fight to preserve copyright and create her IP.
But if that's akin to walking into a newsagent and just stealing the FT and reprinting it...
We licensed to open air, actually, so, yeah.
I feel that's a necessary disclaimer.
But, you know, for all the other people that don't have your heft to win those licensing deals.
So I think this AI inevitability argument needs to get challenged by a few of those...
those unintended consequences.
And one further one, you can see in multiple states in the US, there is a big backlash against building AI data centers.
Right.
Either for the fear of threat of jobs, for the concern that it's going to drain the water re-table.
And finally, you know why Elon and OpenXAI can build its data centers faster?
They just put up gas turbines.
And they pretend that they don't.
So that's a bit of an environmental disaster.
And that's one of the reasons why, alongside these big tech companies, Mag7, NVIDIA, etc., you see tremendous investment interest in power generation and transmission that was moribund for 20 years.
You see, you know, my father was a nuclear engineer at Westinghouse.
He retired in 1994.
There was a 30-year moratorium on R&D and nuclear.
And that's now been revived.
And there's tons of money going into fusion of companies of all sorts.
So you're seeing some ancillary segments that are going to have to be in place to realize this $7 trillion CapEx boom when already there's a question of whether they'll be able to spend a trillion and find the power and the water and the resources to do it.
I do think this is a very interesting way that AI is changing other things sort of around it.
I mean, for 20 years...
venture capitalists like yourself could put a reliable amount of money into a software service company that would have just very predictable growth in any given category.
That seems like it's been blown up in the last couple of years, but I don't know.
You have a couple of those in your portfolio.
Do you think software is still a good investment in the right hands, or is it all just going to be soaked up by Anthropic and OpenAI and Google?
I mean, I think the great thing about the venture capital industry is that if you ask...
participants that question six months ago, you'd get a very different answer to what you get today.
So things change rapidly.
Yeah, the SaaSpocalypse hit hard early in the year.
The SaaSpocalypse, which meant the death of all software companies, was a theme that pervaded for the last 18 months until we had Salesforce earnings and another number of software earnings come in the end of Q2 and people have now decided that actually...
Software plus AI is perhaps where a lot of value still lies.
And if you unleash agents on the software and on the data, then you can extract a lot more value.
So I think that the future looks a bit brighter than it did six months ago.
We're out of time, but I just want to finish on one kind of question at the risk of undermining the previous four to five minute conversation.
What keeps coming up in sort of private conversations that I'm having?
with my colleagues as much as with investors, is like, has there ever been a time when there is so much uncertainty in investing right now?
Like, the range of outcomes are just enormous.
Like, from where we are right now, has your old job ever been harder?
I can certainly say, I mean, I've only been doing this 15 years, so I still feel like a baby of my industry, but it feels like I've started a year ago.
Everything you knew is suddenly unknown.
As you say, the potential outcomes, boom, bust, more likely boom.
But, I mean, anyone who says that they know with certainty where we're going from here, it's just fallacy.
And I would say, you know, when you look at the wider market, the willingness to look at things which would otherwise make you scream and run and hide and run for the hills, 40 trillion of US debt.
random tariff wars being started here, there, and everywhere.
All of those levels of uncertainty have just been waved aside and shrugged off by the markets.
But that is a function of just how much money there is sloshing around in the world global financial system, and it needs to find a home.
This has been great.
Thank you, Richard.
Ophelia, this is a pretty good conversation.
Thank you.
That was me, Tim Bradshaw, speaking to Richard Kramer and Ophelia Brown at the FT Weekend Festival.
Thanks to both of them.
Tectonic is produced by Edwin Lane.
The executive producer is Topher Forges.
Sound design by Breen Turner and Sam Giovinco.
And the FT's head of audio is Flo Phillips.