We're talking about both OpenAI and Anthropic.

Do you think the nature of those partnerships could be a differentiator in terms of the future success of one of these companies? So we started with GPUs. Now we've got TPUs.

Just think an even more specialized AI compute engine.

Microsoft, Amazon, and Google dominate cloud services, but not any one of them winning.

The latest I'm seeing on OpenAI in terms of their run rate revenue is $25 billion as of the end of February.

Welcome to all our listeners, and welcome to this week's episode of "What's The Big Deal?" where we take a look under the hood of major deals in the public and private markets, and we also explore finance industry developments. Now, I'm Deborah Taylor, and I'm going to use my experience from my career in investment banking to bring insights from a public markets perspective.

And I'm Graham Smith, and I'm going to use my experience in private credit to bring the private market perspective here.

Thanks, Graham. And please can you introduce us to what the big deal is for this week? So this week we're talking about actually a couple deals that are going to come to market and something that I suspect we're going to revisit quite a few times this year. We're talking about both OpenAI and Anthropic, the two behemoths in the AI space that are both getting ready to launch IPOs. We'll see which one's going to come first.

But today we want to spend a bit of time and talk about some of the similarities and some of the important differences between both of these companies.

Absolutely. We're returning to the topic of AI, which is quickly becoming our favorite theme. And in terms of what we'll cover in this episode, we will indeed start with discussing their business models.

They do sound like similar businesses, but in reality they're quite different, and we'll try to see which one has the competitive edge.

We're also going to look at recent fundraising and their current financials just to get some steer on what the businesses might be worth.

And finally, with both these companies likely to IPO soon, we'll discuss whether this is actually a winner-takes-all battle for market share of the AI market. As we said, Graham, right at the start there, on the surface, these sound like quite similar businesses.

We've got OpenAI and also Anthropic.

They both have their AI businesses with large language models. Now, OpenAI, they have ChatGPT, now on its fifth version, which I think we're all pretty familiar with, and then Anthropic with their model Claude, which is maybe less known in the consumer market, and they're now on their fourth version. So both have these models that they're running at the moment. But in terms of their businesses, they are quite different. And Graham, you've got a bit of experience looking at software companies. So can you just talk us through, walk us through how the businesses are actually different? Yeah. Let's talk about, first of all, how these businesses, you might think of as software companies, but in a lot of ways are, and they aren't. There are some pretty important differences, mainly related to the infrastructure for both of these businesses. So in a traditional software model, you have some CapEx and R&D spend. You're developing a software platform. But in general, thereafter, your CapEx spend tends to be fairly light. We all like to think of software as fairly high cash conversion businesses because you develop the product, and then once you have it developed, you sign these long-term customer contracts, get cash in the door, and generally don't have to spend a ton to maintain your platform. The difference between traditional software and AI is because of the compute capacity required for these LLMs, actually, the compute capacity and the infrastructure becomes a really important factor and kind of a limiting factor to growth for businesses like this.

So I'm sure we've all heard the terms GPU, TPU thrown around. So Debs, do we know what these terms are? Graham, actually, I have to admit, I don't. And I love a TLA myself.

That's a three-letter acronym. So please, can you enlighten us? What do you mean by GPU? Okay. So I think everyone who's followed this space has probably seen companies like NVIDIA do incredibly well because GPUs are really the compute engine for AI. So the way a typical, say, computer processor works is, this is really, really high level, kind of processes information on a sequential basis. The type of math that is required for LLMs really likes a compute engine that processes information sequentially. So traditionally, the type of math that GPUs run has been used for computer graphics.

This is just kind of what they've been designed for historically, and it just turns out that the processors in GPUs are just much more efficient and much more suited to the type of math required by LLMs. So we've also seen terms called TPUs as well, tensor processing unit.

Tensor is really just the core mathematical item, I suppose, for lack of a better word, that LLMs are computing through.

So we started with GPUs. Now we've got TPUs.

Just think an even more specialized AI compute engine.

And the thing about both of these products is they require a huge amount of power, and these LLMs just require a huge amount of this compute capacity in order to operate efficiently. So unlike traditional software, AI, we see both OpenAI and Anthropic making investments in different ways in compute capacity.

OpenAI is spending a ton building its own and owning its own compute capacity. Anthropic's also spending a ton of money, but more in the leasing or the rental of compute capacity.

We can talk about the differences between the two, but in both cases, we're having to spend a ton of cash just to build out the compute platform to run these models. And that makes this investment, I think, quite different from a traditional software business.

Okay. So you say that the infrastructure is really important, but the way that they're kind of setting themselves up around that infrastructure is actually quite different for these businesses. Now, you mentioned Anthropic effectively leasing their infrastructure. I believe that's something to do with how they're partnering with the cloud providers. Can you elaborate on that? So Anthropic has been partnering with some cloud providers to provide their compute capacity, namely Amazon and Google.

And instead of owning the capacity like OpenAI is doing, they are effectively spending money to lease compute capacity. So think about this in terms of, say, the old retail comparison, where we're comparing businesses that rent their stores versus businesses that own their stores.

So we got metrics like EBITDA, so EBITDA but before rent, that help us compare businesses with the same fundamental business model, but a different financing and ownership structure.

And I actually really wonder if we're going to come up with some similar metrics to evaluate businesses like this, where you've got Anthropic that's spending a bunch of money to lease capacity with partners like Google and Amazon, and then you've got OpenAI that's touting huge infrastructure investments in terms of owning its compute capacity.

They're also, of course, partnering with some of the big cloud providers.

Of course, Microsoft, Nvidia have both made big investments into OpenAI as well. But what OpenAI has been telling investors is that actually us spending, I want to say it's $500 billion, or committing to spend half a trillion dollars on compute capacity is a really big barrier to entry, just because they want to own the capacity and the infrastructure. Which one turns out to be the winning strategy? I think it's probably too early to tell yet, but I can certainly see benefits and drawbacks on both sides of that story.

For sure, because as you mentioned, the analogy with retailers that own or lease their buildings.

If you own your own infrastructure, although it's costly up front, that gives you a huge amount of control, particularly if there are constraints around capacity, well, that's yours to manage. Whereas if you're renting, which is Anthropic's sort of angle on this, then sure, they don't have to pay for everything up front, and also maybe it gives them agility.

They can tap into other sources whilst they're growing.

But then the downside is that if there are constraints around capacity, they're just in the queue. They are kind of downstream, if you like, of those providing the capacity. So in the short term, it sounds like it gives Anthropic an advantage around agility and the cash flow side. But longer term, maybe it skews the benefits to OpenAI on the upside.

Quite possibly, and again, I'm certainly no expert in this space, but everything that I understand about the AI space, about large language models, really suggests that compute capacity is the barrier to not just entry, but barrier to progress, in that the more compute capacity we have coming online, the smarter these models are going to get, the better they're going to perform. So right now, one may have an advantage over the other in terms of the current product offering, but there is an argument to say that, say, OpenAI spending this huge amount of money on compute capacity, even though maybe their product offering for certain customers isn't as interesting as Anthropic's today, in the long run might actually turn out to be a really important part of the strategy.

Mm-hmm.

Again, I think it's wait and see a little bit.

I know there's a bunch of compute capacity that has been committed, that's being built right now and is going to start to come online soon. And I think we're going to see some really interesting step changes in the capability of some of these models in the not too distant future.

And I think at that point, we might start to get an idea of just how important owning that infrastructure is versus leasing it.

It feels a little bit too early to make that distinction just now.

Absolutely. And to be honest, I think it's worthy of a whole separate episode, talking about the infrastructure side of things.

The amount of money flowing into- Agreed ... these hyperscalers. So maybe let's just park that for now.

But let's talk about another area where they're quite different, in terms of their route to market. Because actually, OpenAI and Anthropic, they've kind of tried to sort of gain market share in quite different ways, haven't they? Can you talk about that, Graham? Yeah. I think everyone listening to this, I think, will have heard of, if not used both.

I don't know about you, I actually use both on a regular basis.

I find them better suited for different tasks.

OpenAI has got a fantastic brand and consumer awareness just in terms of their reach.

They have spent a lot more focus on the retail consumer, right? And the benefit to that is the visibility, it's the brand. The drawback is, think about how little most consumers want to spend on technology like this.

I can't remember the stat off the top of my head, but I think in terms of the percentage of customers, the vast percentage or the vast majority of customers in terms of percentages for OpenAI are not paying. They're on the free tier. Right? So if you compare that to Anthropic, who's really gone after enterprise customers with Claude, Claude Code is so popular because it's such an interesting tool.

But also, if you think about the way they have marketed themselves and built their brand, selling to enterprises, they've got more revenue visibility and longer term customer contracts.

They're selling to people who actually have money to spend on this kind of technology.

Personally, I wind up paying for both every month, but I'm a drop in the ocean compared to the big corporates that spend millions and millions of dollars per month, per year, whatever you're looking at, on tools like Claude. So I do think right now Anthropic has the slight advantage in terms of business model.

Just great customer base, better revenue visibility.

These companies both burn a ton of cash right now.

But if we think about how do you value businesses, what kind of revenue is more interesting, it's the predictable recurring revenue that investors find a lot more interesting.

So I think despite the fact we've got different types of infrastructure right now, it feels to me likeAs we sit here today, which company is doing better right now? It feels like Anthropic. Do they continue to keep doing better in the next, say, three, five years? Who knows? We'll see what OpenAI's compute capacity does for them, but right now, I think Anthropic is in a really interesting and exciting place.

Absolutely. And I think one thing that we could just sort of touch on is the fact that we have very limited data about their numbers.

But what we do know is the splits of their revenues in terms of sources of their revenue streams.

So the latest figures that I've got for OpenAI, about three-quarters of their revenues are subscriptions, particularly from members of the general public, and the rest, APIs.

That's basically paying for sort of automated access to the models. Whereas for Anthropic, about 85% of their revenue comes from APIs versus the rest from subscriptions.

Now, my understanding of APIs, I don't do coding myself, but I do work with people that do, is they spend a lot of time setting up code that then automatically pulls the data in from these APIs.

So that presumably gives quite a lot of stickiness to those API revenues versus subscription revenues, which they are generally viewed as quite good quality revenues, but you can just turn them off on a whim and switch provider, which maybe in a way is not so easy with APIs.

Yeah, I think that's absolutely right.

So API just simply stands for application program interface, I believe.

Oh, good acronym there. And all it means is, this is...

I should check and make sure that's right.

Yeah.

But all this really means is if you're a software company, if you're a developer, and you want to integrate Anthropic's product into your platform, either Anthropic or GPT, doesn't matter, you're going to use some kind of interface to interface with that product directly, vis-a-vis an API. That differs massively from the consumer, where you're sitting in front of a chatbot effectively, and to your point, if you want to change that to another one, you can do that really easily.

So if you're building a program of any kind that really relies on a specific type of AI and an API call into either Anthropic or OpenAI, you're a lot more likely to be sticky. And it's not to say, by the way, that you can't rip those out and change it with something else, because I can also see a world in which these models get so good that you can quite easily take that same program, plug it into the competitor API, and the model is just so good it understands exactly what you need to do and takes everything from where you started and just keeps running with it.

Are we at that point yet? Probably not.

So I can really see at least a near-term benefit for Anthropic's Claude in terms of their stickiness, just through developers building Claude specifically into their product.

So we've talked a bit about the infrastructure differences for these companies.

We've also talked about the differences in their route to market and their focus on different types of customer. What about their strategic partnerships? I know that OpenAI as well, that's well-known for having a very close partnership and actually quite a lot of funding from Microsoft.

On the flip side, Anthropic, a bit closer to Google and Amazon Web Services, obviously because it leverages their cloud services.

I think also they've been involved in some funding as well for Anthropic.

Do you think the nature of those partnerships could be a differentiator in terms of the future success of one of these companies? Maybe. I think it's too early to tell right now, and the reason I say that is that it does feel like there's actually quite a bit of overlap between the two. We've seen Microsoft, NVIDIA, Amazon make big investments to both of these platforms. So you can also see this as a way for some of the big guys to be hedging their bets a little bit.

Mm.

I mean, obviously Anthropic's got a huge commitment from Microsoft.

That's the main partnership. But we've seen these big players make investments into both platforms, so I don't know yet. I don't know. I mean, what do you think? Do you think it's a real differentiator just yet? It's interesting, and we might come back to this, but my view is that it increases the chance that this will continue to be a fragmented market as it evolves. Because we've seen Microsoft, Amazon, and Google really sort of dominate cloud services, but not any one of them winning. And I just wonder to what extent the fact that they are all partnering with both OpenAI and Anthropic, that continues to result in fragmentation of the market rather than just consolidation around just OpenAI or Anthropic.

But I think that's a really interesting sort of discussion point, and I'd love to come back to that.

But yeah, that's my view. I don't think it's necessarily that if you align with one of them, you're going to win.

Yeah, I agree with that. I mean, the thing that I think about with this kind of thing is this technology is so quickly becoming so good.

If one of them gets to the point where-- Maybe we used to talk a lot about developing real artificial general intelligence, or AGI.

I feel like we've pulled back from that a little bit in recent months.

But if one of these models just gets so good that the other one is almost rendered useless, not useless, but so good you just don't need the other one- Mm ... do you wind up in this winner-take-all situation? And is this some of these cloud computing platforms hedging their bets a little bit? Yeah.

I don't know. I want to keep up to date on this and continue the chat, for sure.

Absolutely. Well, let's talk about one other area where there is a difference. We have some information about the revenues and maybe even a little bit of information about the cost of these businesses based on recent news articles.

Could you just kind ofProvide us a little bit of insight into the numbers for these two companies, how they compare, and maybe what that tells us about which one is winning at the moment.

Okay. So right now we've got both Anthropic and OpenAI, kind of in a similar ballpark. So I just want to look at my stats here to make sure I've got this right.

So the latest I'm seeing on OpenAI in terms of their run rate revenue, I want to come back to the run rate discussion in a second, is $25 billion as of the end of February. It's crazy to think that in today's world, the end of February for a business like this is actually quite dated. And then Anthropic, we're talking about $30 billion run rate as of early April, so a little bit more recent.

So both very much in the same ballpark in terms of revenue. Now, run rate revenue, you can really make an argument for why that's important for a business like this.

For anyone who doesn't know, when we think about run rate revenue, we're trying to form a view of effectively what is the revenue of the business today kind of going forward. Because if you look at a high growth business like this, I think I read a stat saying Anthropic doubled their revenue monthly, from one month to the next, not that long ago.

So usually when we're putting our conservative investor hats on, we're looking, say, last 12 months to get a view of underlying real performance. But for companies that are growing so quickly, you generally tend to look at revenue on some kind of run rate basis.

So I don't know how run rate here is being calculated, and I think it'll be really interesting once we start to get more detail through both these companies' S1s to find out how they're thinking about some of these run rate calculations. Because you can take a pretty simplistic approach and take the last quarter and multiply it by four.

You can take the last month and multiply it by 12.

You can kind of form a view that that's a semi-reasonable way to look at a business like this. But there are ways that these run rate calculations can get really muddy. Let's say one, if not both of these companies, are including some kind of unsigned, but maybe expected contract. Let's say they are assuming more revenue comes online from compute capacity that's about to come online but hasn't been turned on, or we haven't received contracts that underpin it yet.

I don't know what's in these numbers yet, and I don't know how much confidence we can place in them. And that's something I'm really interested to get under the hood of once we start to get a little bit more detail through the S1s.

Ooh, so it does sound a little bit murky, but nonetheless, quite surprising that Anthropic, having been very much the late starter, has at least matched OpenAI's position in terms of its revenues and possibly is overtaking.

That's quite significant, isn't it, Graham? Yeah, agreed. And also I'm trying to find the last stats that I've seen on the purported valuation.

I want to say OpenAI was around a trillion, and I'm just kind of surprised that the target valuation for each of those is so different right now. Right? So if we've got Anthropic that's really caught up to OpenAI in terms of revenue, I would argue today probably has a higher quality of that revenue in terms of visibility through enterprise customers.

I'm just kind of surprised that we're not talking very similar valuation targets for both of these businesses, unless there's just a huge amount of value ascribed to OpenAI's compute platform in their valuation. But no one really knows yet, right? Because all we've seen are some headlines that are being talked about in the market. We don't have any real detail to evaluate yet, and I'm really interested to start getting some of this and forming a real view.

Absolutely. And I think even compounding what you were just saying, something that I was reading recently was about efficiency of training the models. So OpenAI, I think they're reported to have training costs, which are about four times Anthropic's recently, which given the comparable performance, you say that you use both these models. I do too as well. There's no real discernible difference in terms of the quality. It suggests that Anthropic has been much more efficient about its approach to training its model.

And maybe, and this is just my interpretation, the fact that they're focusing on enterprise customers, maybe they're being much more focused about how they train their model for a specific use case.

Whereas OpenAI, they're having to train ChatGPT for all applications, and maybe that is making them less efficient about how they deploy their training costs. So we're in a situation- Yeah, I don't know. I'm interested in learning more.

Mm.

It's a really fascinating topic to dive into. But just think about the amount of cash that needs to get spent to bring these models to life. It's the investment in the actual infrastructure.

It's the ongoing energy cost. A huge amount of that cost is spent on model training. And then you finally get to spend all the money on the CapEx and spend all the money training the model.

Now I finally get to the point where I can start selling the model and actually earning money on the actual, I think inference is the term that tends to be used for that.

Absolutely. And actually, that's a great point, and it brings us very nicely onto their fundraising because both of these companies have been raising extraordinary amounts of cash. They've had to.

They are burning through cash at a huge rate, not expected to be cash flow positive for a number of years yet. So let's talk about their fundraising, because it does give us some insights into what these businesses are worth. So, Graham, what do we know about the recent fundraising rounds in particular and what it implies for their valuations? Let's see. So they're both in similar kinds of states in terms of their development as companies.

So I've just got my notes here in front of me.

Anthropic raised their Series G in February of this year.

They raised $30 billion through that fundraise at a $380 billion valuation.OpenAI has been raising more money, admittedly, recently.

Their last fundraising was their series F, 122 billion at an $852 billion valuation. And to be fair, they raised also, I guess we've got two series F dates, one in February, one in March.

In February, they raised 110 billion.

In March, 122 billion. So, $230 billion raised just earlier this year.

And again, we know where that money's going.

This is going to building out that compute capacity.

So that might explain part of the difference in terms of the expected IPO valuation, is just that asset-heavy infrastructure that OpenAI is building for itself.

Okay, that's really interesting.

Because I think we have struggled so far to say why, if they're both comparable in terms of their revenues, they're both doing really well at gaining market share, why would they have such a difference in terms of their latest valuation? As you said, the latest fundraising gives OpenAI a valuation about $850 billion versus Anthropic's at closer to, what, 380 billion.

That's still quite a big difference.

But you suggest maybe that's because of the infrastructure differences.

Potentially. I don't know. I'm literally making this up on the fly.

Mm.

And you look at the difference between those two numbers, and you say, "Okay, OpenAI is investing a half a trillion dollars in infrastructure." That's about the difference in terms of valuation between the two.

And you say, "Okay, maybe that's it." You can also see OpenAI pitching investors with-- I can see people being worried about overspending and overcapacity, but if you think about there just being an unlimited demand for compute capacity for all these models, you could also see OpenAI saying, "Hey, if things don't work out with some of our models and we don't need it all- Mm ... what's stopping them from leasing that to Claude or whoever else, right? It could be an interesting secondary revenue stream for them.

Oh, that's a really good point, Graham. Completely see that.

That basically you've got that ability to pivot with that infrastructure in a way that Anthropic doesn't.

So that control point, really, about leasing versus owning your assets really comes to the fore there.

I think what's interesting, just looking at the numbers, and the information about the recent fundraising, that we do have slightly different types of investors participating.

In particular, Anthropic has much more involvement from institutional investors. We've got Fidelity, D. E.

Shaw, well-known institutional investors quite heavily involved in the more recent fundraising, compared with OpenAI. I do wonder whether that will give Anthropic a little bit of an edge when it comes to the actual IPO, because these are the sorts of investors you're really looking to, to be big participants in that actual IPO event.

So, I think that could flatter slightly their valuation when it actually comes to the IPO.

But the numbers I'm hearing rumored, and it is very much rumors at the moment, that Anthropic is aiming for a valuation of about $600 billion versus OpenAI looking for a trillion dollars in terms of their IPO valuation.

What do you think about that compared to the recent fundraising? It seems relatively in line with what we might expect. I guess if we're saying 600 billion for Anthropic, you say, okay, last funding round done at just about $400 billion.

On the basis their run rate revenue, however that's being calculated, has doubled in the last couple of months.

Do you see a world in which their valuation at IPO is quite a bit higher than the last fundraising round? Yeah, you probably do.

Mm.

So that doesn't sound too crazy. And again, on the OpenAI side, the last private investment they had valued them at 850 billion.

By the way, compared to the SpaceX discussion we had the other week, where they were going from, I don't know, 800 billion to $5 trillion, going from 850 to a trillion doesn't sound too crazy. So, I don't know what I don't know yet, because we don't have all the information.

But based on the trajectory that we're seeing here, those don't sound like insane numbers to me.

Yeah, absolutely. And actually, the other thing that we're hearing a lot of, both in relation to SpaceX and Anthropic and OpenAI, is that there's expected to be really large retail allocations.

Typically, in an IPO, you get 5 to 10% allocation to retail investors. And I think SpaceX, we're expecting maybe a 30% allocation and possibly similar for these companies.

And that just reflects the fact there is so much appetite from the public to invest in these companies.

It's really caught the imagination, I think, of the public, hasn't it? Yeah. Just think about the conversations you have day to day. We're both in financial training, so obviously this comes up all the time in terms of what AI tools can we use.

Are there any of these tools that we think are going to replace our jobs or someone else's jobs? Just think about the number of conversations you have about this kind of platform on a day-to-day basis with whomever.

So on that basis, with everyone kind of invested and excited, can you see there being a lot of retail demands? Yeah, probably.

Absolutely. I think we should probably challenge slightly what we're hearing. I don't want to get too swept up in the excitement of AI.So I did a little bit of analysis on the valuation side, and I think if OpenAI are targeting a $1 trillion valuation, we know that there is a lot of investment for them at the moment that really drags down their free cash flow in the short term.

There's good expectations longer term about the fact they have control over their infrastructure, but the valuation of $1 trillion does require really aggressive growth.

We're talking probably revenue growth, CAGR of about 30% over the next decade, and it would give them an implied revenue multiple. Remember, revenue multiples are really important for IPOs. A multiple of about 40 times, which is not dissimilar to what we had with Cloudflare when they did their IPO, but they were very much positioned as the market leader in network security. That's the market that they operate in.

Whereas OpenAI obviously wouldn't be necessarily the market leader.

They kind of share that position at the moment with Anthropic.

In terms of Anthropic's rumored IPO valuation, about $600 billion.

They do have much lower CapEx needs at the moment.

Longer term, as we've mentioned numerous times so far today, sort of longer term concerns about ability to kind of control their infrastructure and maybe capacity constraints, but much less stretching valuation.

We're probably talking a revenue CAGR over the next decade of around 25%, and that valuation would put them on a multiple of only 20 times. So slightly less ambitious, I think, when we're talking about Anthropic's IPO valuation.

Now, I think the big question, the question du jour, shall we say, is really whether the valuations that we're hearing rumored, are they supportable for both companies, or is this a winner-takes-all situation where really it's only feasible for one of them to dominate in, let's say, 10 years' time? Now, I'm hearing numbers for the global AI market in 10 years' time of about $5 trillion. So we're seeing a huge sell-off in the public markets for companies disrupted by AI.

So there is going to be presumably some transfer to the businesses that benefit from AI. But in addition to that, there's going to be hopefully productivity improvements. So all of that should result in a really vast market. So $5 trillion is what's being talked about. Now, if we use the numbers that I've just talked about in terms of the revenue CAGRs for OpenAI and Anthropic, that would put them on revenues of about 300 billion or slightly more than that in 10 years' time. That's only about 6% or 7% of a $5 trillion AI market. And that actually sounds quite achievable, doesn't it? Yeah.

It doesn't sound- Yeah ... too ambitious. And in my mind, that points to a much more fragmented AI market, not just this duopoly or even monopoly by one AI company. It suggests that there's going to be spoils for lots of different participants in the market.

Probably Amazon, Google, Microsoft with their cloud services, possibly some of the end users in terms of companies like healthcare companies, robotics companies, financial services, who then need to apply the AI in their business.

They'll want a piece of the pie as well.

So I think in my mind, the numbers suggest that we're not talking about a winner-takes-all situation.

But Graham, what do you think? I think it really depends on how these models improve in coming months and coming years. In terms of the overall landscape, in terms of the potential size of the market, I very much agree with you, and I also hope that's the case, by the way. Because I think the worst thing that we could do really as a society, if you think about the potential for this technology to put people out of work, like really take over huge pieces of the economy, I think the worst thing we could do is really give that to one player. The thing I think about sometimes is if you think about the compute capacity that's coming online, think about, say, OpenAI building all this compute capacity, do we reach a point where one of these models reaches, maybe not AGI, but a level that's close enough to that where it's just so good that it kind of renders the other ones, not useless, but certainly not as useful as they used to be.

And in that scenario, do we end up in more of a winner-takes-all market? Maybe, but I would hope but I'm not holding my breath that we have some kind of regulation that prevents that from being the case. I think that's kind of a big ask, just given, one, the political environment in the US right now, and two, just how much faster this technology is progressing than anyone in government probably even can keep up with.

But we'll see. I don't know if we have an answer to that question right now, but I certainly hope you're right.

Hmm.

Because it's a really weird time in history right now. You think about just the history of computing, where everything used to be governed by this thing called Moore's law, which basically said every, what, one to two years, the number of transistors on a computer processor would double, and that was kind of the thing that limited progress.

Now it's how much money can you spend on infrastructure, how many GPUs can you buy, how much power do you have access to, and do we wind up in a situation where the winner is just the one who builds the most compute capacity and therefore trains the best model? Maybe.

I don't know.

Yeah, it does. It's starting to feel a bit sci-fi, isn't it? I still remember "Star Trek." Right. I know. Or having those flip phones for being able to teleport around the universe. Maybe we're nearly there.

But no, it does feel very sort of sci-fi, but also, I think very exciting from an investor perspective.

And it's important we don't get swept up in just all the rhetoric around the IPO companies, that you challenge the valuations. But there's definitely opportunities, and the skill hereIs identifying where the opportunity is.

Is it with these newcomers, with their large language models? Is it going to be with some of the incumbents in terms of cloud services providers who could end up taking some of this AI market, or even, as you say, disruption from someone else completely? Really interesting to see how this plays out, I think.

Yeah. I think this is going to be the first discussion of many, and we'll get a lot more insight and a lot more detail as we get closer to listing. And I guess the big question is going to be, who's first? Mm.

I do feel like there's going to be a big first-mover advantage in terms of investor appetite valuation, and I'm sure they're both, as in Anthropic and OpenAI, are both rushing to be first to the starting line here.

Which one's your money on, Graham? I'm going to go Claude Anthropic for being first, and I guess I say that just because of the enterprise tie-in. Do they have more potential-- and I don't know what their customer list looks like, but do they have more entrenched relationships with, say, some of the banks, some of the stakeholders that actually bring the IPO together? Mm.

I don't know. It's just a shot in the dark. What do you think? I'm going to actually go the other way. I'm going to go with OpenAI.

I think the pressure's on with them.

If they don't move first, it's very much a risk that the IPO just doesn't take off in the way they would want.

I think also, just some of the messaging, some of the public statements that they've been making maybe suggests that they're really pushing to IPO soon. But hey, let's see.

We'll see how it plays out. We'll watch this space definitely, and hopefully, many more episodes to come on both AI and also definitely the infrastructure side of things.

So I hope that all of our listeners have enjoyed listening to this week's episode, into our deep dive into OpenAI and Anthropic.

If you have enjoyed listening, please do like and subscribe to our podcast. And if you're listening on YouTube or watching on YouTube, please do subscribe and also leave us a comment.

We do love to hear your comments, and we will respond to them, won't we, Graham? Definitely. No, 100% we will. Yeah, please engage.

Let us know what you think. If there are any topics you want to hear about next as well, let us know down in the comments, and then we'll get them in our next few episodes.

Yeah, absolutely. So until next week, same time, a new deal and some fresh insights from myself and Graham.