Steve Hsu: And you can see already by twenty twenty-seven, twenty twenty-eight, you're talking about revenues that have to be in the trillions of dollars. Okay? So this is an incredible amount of money. A good comparator is total software revenues. So all the software companies in the world together, you're talking about one point five trillion per year in revenue today.

And what you're saying is that the AI companies or revenue really driven by AI itself has to more or less eclipse what we currently spend in total on all software by roughly twenty twenty-seven, twenty twenty-eight. That's the implied requirement if you want a decent return on existing and near, near future CapEx.

So these are, these are the numbers that have everybody concerned to some degree, and it basically quantifies the hurdle that the AI industry has to hit. The AI industry has to hit this hurdle, these numbers, in the next few years, or the CapEx that's, that's already been invested or currently planned for the next year or two, that will end up having a poor rate of return.

So if we fall far short of these numbers for future AI revenues, then this investment will turn out, when you look back, say it's twenty thirty and you're looking back at money that was invested in twenty twenty-six or twenty twenty-seven. You look back and say, "Wow, those guys got a terrible rate of return," maybe even a negative real rate of return from their investments.

And then people will look back and say, "This was a bubble."

Steve Hsu: Welcome to Manifold. This is a special episode. It will not be very long because I'm on the road. I was just at a big conference. Not big in terms of size or number of attendees, but big in terms of the amount of capital capital allocators represented here. As you can imagine, there was a huge amount of discussion about the AI CapEx build-out the Anthropic IPO, whether we're in an AI boom or an AI bubble or in a sense perhaps both could be true.

And so I thought I would share my views on this. In the slides I'm about to show you, I'm using figures that I myself have tweeted out or posted on X in the last, say, month or so, and you can go look at my timeline. So one of the recurring themes I've been discussing on X is whether or not I believe the that we're in an AI boom or bust.

What is the evolution in capabilities? How is this likely to translate into real productivity in the economy, revenues for AI companies, revenues for hyperscalers? So those are all the questions that I'm gonna discuss in today's episode.

Before I get into the financial analysis, I just wanna comment that the models themselves are improving very rapidly in capability.

And I think average people who are not themselves software developers or mathematicians or, say, theoretical physicists wouldn't necessarily be aware of just how good they're getting. And in my own case because I use the models heavily for research and also work on how to improve the model capabilities in areas like math and physics I'm very aware of what's happened.

And I would say that for me, I'm living in a kind of science fiction world now where I can talk to an AI or work with an AI that's actually better than, you know, most of the top level theoretical physicists in the sense that it has very, very broad understanding of our whole field literature that for example, is outside of my own specialty.

And so I haven't had time to dig deeply into that literature and master all the concepts and results, but, but it has. And then secondly, to do very, very powerful and complicated calculations very quickly. It's never been easier to do research in math and theoretical physics with the assistance of these AIs.

And so that has been a huge change just from a year ago. So if you remember a year ago, I wrote my paper in which,I had a result in quantum mechanics in which the original idea actually came from GPT. And so I, I worked with GPT to produce a publishable result which appeared in Physics Letters, and it was controversial at the time.

A lot of people just were very negative about it because they thought that AIs were only capable of producing slop research. I think people's perceptions concerning that have changed within the last year. By now, probably everybody knows that some very important open problems have been resolved in mathematics using what I would call AI super intelligence.

So I think it's pretty clear that they are superhuman in a bunch of different senses, even though their capabilities are quite jagged right now. I won't focus on this aspect in the current episode, but the math community is actually, I think, in a, in a kind of, trauma and adjustment phase where they, they really have to adjust to the fact that they're in the presence now of things which are they're not quite oracles. They can't solve every problem, but they can often make important advances on problems where humans have been stuck for a long time. I think the reason I just went through that discussion is so that you know that even though what I'm gonna present here is a kind of bear case about the AI CapEx boom and where the financial markets are likely to go with respect to AI, it doesn't mean I'm pessimistic about the capabilities build-out.

I actually think model capability will continue to improve, at least for the next couple of years but that it will take longer than that for humans to adapt to using AIs in a productive way in a broad sense throughout the whole economy. And that scaling into the broad economy has to be accomplished in order for current investments in CapEx or in equity in the frontier labs to pay off for that to have substantially positive ROI.

And there's a good chance it'll have substantially negative ROI. So in that sense, I am calling a kind of AI bubble right now. That is my best guess. Of course, again everything in finance, it, it's difficult to predict the future. This is what I feel right now. I could easily be incorrect, but I'll give you my analysis for why I feel this way.

And I think increasingly more and more people are coming to this perspective after digging into the numbers more carefully.

All right, so let me start by showing you some slides. The title for these slides is Both Things Can Be True at Once, and the two things are, number one, AI capabilities are increasing dramatically, and in the long run, I think they are going to be incredibly impactful in the economy.

However, the other thing that's true is that I think there's a mismatch between how much is being invested in AI, how AI companies are being priced today, and the level of revenues that those companies are going to be able to drive in the next few years. Beyond that, it's very, very hard to forecast because I think eventually we'll get into things like RSI.

The models will increase, increase in capability even faster. Butalso just as important and often overlooked, human gatekeepers at companies, institutions of higher education, the government, the military, to human gatekeepers and leaders have to adjust to using the AI in an efficient way so that it can actually have broad impact in the economy.

And I just think that will take longer than the timescales that are discussed here, like the next few years.

Okay.

So this graph is looking at AI CapEx And the main point I want you to take from this graph is that, you know, at the beginning, the hyperscalers could fund it mainly out of their own operating positive cash flows.

That's the light blue part of each bar. But now increasingly, the CapEx is being by CapEx here, I mean, data center build-out, purchase of GPUs et cetera. Now it's being funded increasingly by debt. And so these companies are having to borrow money in order to finance the next stages of CapEx.

And because it involves debt, bond investors have a very different psychology than equity investors or venture capital investors. Venture and equity people generally are more driven by narrative. They're willing to take a kind of long-term bet on something, you know, very impactful happening in the economy, some kind of story about some great new technology that's gonna change the whole world.

Bond investors are more conservative, and they will look more carefully at the net present value of future cash flows, whether the company can actually pay the interest rate for the money that they borrowed. And so now what's happening is that both because of the overall size of the AI build-out and also just because a different kind of investor is getting pulled in, you're starting to see much deeper analyses of these numbers, which I, I think paint a kind of negative picture of, whether or not a current CapEx is, is gonna turn out to be justified.

Okay.

So this is from, I think Goldman and also The Economist. I think the two curves, red is from The Economist and blue is from or the other way around. So blue is The Economist and red is Goldman. And these are annual AI revenues required to return a modest, say, ten percentcap-return on CapEx, that's already been expended or, or to a particular point in time.

And you can see already by twenty twenty-seven, twenty twenty-eight, you're talking about revenues that have to be in the trillions of dollars. Okay? So this is an incredible amount of money. A good comparator is total software revenues. So all the software companies in the world together, you're talking about one point five trillion per year in revenue today.

And what you're saying is that the AI companies or revenue really driven by AI itself has to more or less eclipse what we currently spend in total on all software by roughly twenty twenty-seven, twenty twenty-eight. That's the implied requirement if you want a decent return on existing and near, near future CapEx.

So these are, these are the numbers that have everybody concerned to some degree, and it basically quantifies the hurdle that the AI industry has to hit. The AI industry has to hit this hurdle, these numbers, in the next few years, or the CapEx that's, that's already been invested or currently planned for the next year or two, that will end up having a poor rate of return.

So if we fall far short of these numbers for future AI revenues, then this investment will turn out, when you look back, say it's twenty thirty and you're looking back at money that was invested in twenty twenty-six or twenty twenty-seven. You look back and say, "Wow, those guys got a terrible rate of return," maybe even a negative real rate of return from their investments.

And then people will look back and say, "This was a bubble." Okay? So I think these numbers are not controversial. They're very simple numbers where you just look at depreciation and the already planned CapEx that's announced by these public companies that they're gonna be deploying, and you can just back out, well, how much money has to be made in the future in order to justify these investments.

So this is not the controversial part. I think everybody in finance who's looked at these numbers finds them reasonable. So everyone agrees now that in order for this era to not be declared a bubble, we have to hit multi-trillions in AI revenue within the next few years. Okay? So that's the hurdle. The question is how plausible you find that level of revenue growth.

What is the level of revenues today? And so here I have on screen several different estimates from different entities of how much people are currently spending on AI. And the number's typically something like a hundred and fifty to two hundred billion dollars a year.

So that's about ten X less than the hurdle that we need to hit. So on the previous slide, we talked about the revenue hurdle that we need to hit in the next few years. That's in the trillions. Current expenditures here, shown here on this slide, are roughly a hundred and fifty to two hundred billion. So there's at least a kind of ten X that has to happen in the next few years.

So 10X growth in AI revenue has to happen in the next few years. Now, these numbers, a hundred and fifty billion to two hundred billion, even these numbers could be optimistic. For people who have lived through a speculative bubble, especially in tech, those people are familiar with the fact that even these AI revenues could be driven by speculative capital.

So if venture capitalists fund my startup and they value it at, you know, ten billion dollars even though we haven't done anything yet and I'm using the billion dollars that I raised from the venture capitalist to do a bunch of stuff with, say, AI applications, I am spending speculative investor money on a business model that hasn't proved out yet and may ultimately not turn out to be part of the real economy or a big part of the real economy.

But for a short period of time where capital is chasing AI ideas, okay, that's the, in a way, is in a sense the, definition of a speculative bubble. During that period, there's a ton of money around to be spent on consumption of tokens. And so it could be that this hundred and fifty to two hundred billion that's on this slide is not, quote, "real organic AI spend."

Real organic AI spend would be tokens purchased which improve some real-world product or real-world service for, say, a profitable company. Not a company that's running on speculative investor capital, but one that has products that consumers buy or other businesses buy, generates profits for them, and, and contributes to the actual productive economy.

Okay. So it could be that even this number maybe half or a third of it is that kind of, quote, "real organic AI revenue," and the rest is speculative. So it could be that it's not a 10X hurdle that we have to hit in the next few years in growth in total AI revenue. It could be 20X. It could even be 30X.

Okay. But it's at least 10X. I think everyone would agree that we have to have a 10X in the total amount of AI revenue to justify the CapEx that has already been done and will take place in the foreseeable future, say, just in the next couple of years. Okay. So that is, I think, in a nutshell, the situation that everyone is facing, and you either believe or you don't believe in that 10X growth.

And if you don't believe it, you think we're in a speculative AI bubble, at least for now. It doesn't mean you don't believe that in the long run, long run being five years or 10 years, that AI will deliver in some sense. It's just that the people that are doing this unprecedented level of investment, so trillion dollar a year investment in CapEx and also buying shares in AI labs like OpenAI and Anthropic, those people will turn out to lose money, that, that these will turn out not to be good investments.

And that, that's a definition of a bubble taking place during this few year period. Okay? And for example, NVIDIA might be way overvalued as a company.

Okay? Now, of course, we can't know how AI revenue will behave in the next three years, right? So that's the multi-trillion-dollar question. That's what everybody's interested in. That's why I spent a lot of time at this meeting talking to other startup people who are themselves building AI applications and what they're seeing in terms of rate of uptake for their products. I talked to people who gather statistics, like the ones that I'm showing here on the screen, on AI expenditures, to what extent they're going to US labs versus Chinese labs.

What is the actual total rate of growth of spending on AI? The- these are the key numbers that, of course, we can only take a snapshot of what's happening today in late twenty twenty-six, but we're trying to figure out what's plausible to happen in the next few years. Okay?

So here's a figure which came from Ramp.

If you're not in the startup space, you're not that familiar with Ramp maybe, but Ramp is a company that tries to handle your HR and accounting, and they have a, an ability for one of the features that they ship on their platform is they allow your company to track its AI spend. And so there's a pretty good cross-section of companies that are Ramp customers and many of them are startups, so they're the companies that you'd most expect to be heavy users of AI.

And Ramp publishes an AI their AI statistics. So this is the most recent one. I think this just came out, like, today or the day before. And what's shown on the left is token volumes. Those, those greenish bars are token consumption by AI users, and you can see there's this there's a very rapid growth in token consumption.

The red bars are token spend, so how much money has been spent in a particular interval, I think weekly that is monitored by companies that use Ramp as a, as a kind of internalstatistical tracking platform for their business. Okay? And I think these numbers are not perfect, but they are a representative or at least quasi-representative sample of a large number of companies that use AI.

And you can see a plateau. So you can see very rapid growth, but you can see a plateau since roughly summer in token spend, and that is the actual source. People are buying tokens from which are produced by models that are built by the AI labs, and they're running at data centers, which are built by hyperscalers.

Okay? So the money flowing in that's eventually meant to comprise those trillions of dollars of revenue is what is represented, or at least the, the time change is represented from this sample in the red bars, and you can see it's leveled off. Okay? So rapid growth, and then, but total revenues have leveled off.

That's for multiple reasons. One is that there's competition from Chinese labs. There's competition between the top US labs. They've all shipped models that are more efficient than earlier versions. And it could also be that people are not getting enough economic value from the tokens, and they've, they've started monitoring more what they spend.

But any case, this is not showing the kind of rapid growth that's required for this ten X to happen over the next few years. Okay? Now, of course, again, this could just be a little bump on the road an error, an air pocket or something, and maybe like after September, these data are up till September.

Maybe in the coming months, this massive growth will just resume, and we will hit this 10X growth in revenue within the next few years. But this is definitely an alarming development. It's also alarming for the Anthropic IPO, which I'll talk about as well. So there's sort of two separate issues here.

One is the companies that are buying chips, building data centers, are their investments gonna turn out to be good ones, have a high rate of return? Secondly, if you buy a share in OpenAI or Anthropic, is that gonna turn out a few years from now to be a good investment? So there's actually two issues that we wanna discuss here, and they're, they're actually separate issues.

You could have a scenario, this was actually discussed by one of the leading AI bulls at this meeting that I've been at. He made the point on his slides that the Cap, because I think he's heavier into CapEx than into the labs themselves. His point was, even if the open models out-compete the closed models on, say, price or, or for whatever reason or they, they compress margins for the closed models, still the CapEx people can win because still many, many tokens will be generated, and they will be paid for generating those tokens regardless of which models are used.

Right? So, so there's two issues here. We could have a scenario where the CapEx people, people who are bulls on AI CapEx win, but investors in ANT and OpenAI lose. That, that could happen or maybe both types of investors win. Okay. Here's another analysis. I have several slides of this type, so I, I don't won't dwell on them.

But these are, these are more estimates. In this case, this is specialized to Anthropic specifically. So again, this is an external analysis. Anthropic is preparing for an IPO and some numbers have are now available from their S-1 filing, which is in preparation for the IPO. That filing only covers their revenues up till July. Up till July is when we see this extremely, you know rapid growth here.

But this is specific estimate from external analysis of what Anthropic's revenue the term we use is annualized revenue run rate, ARR, which is an extrapolation from a point in time. Like, oh, if we earned this much, the amount that we earned in the trailing month, if we or trailing week, if we earned that for the rest of the year, how much would we make?

For an entire year, how much would we make? So that's the ARR, ARR. And so investors are looking heavily at, at that ARR and trying to extrapolate it into the future. What this slide is saying is that for Anthropic specifically, this external entity that did the analysis thinks there's been a flattening in revenue growth or ARR growth for Anthropic, and so that's why these, these bars look flat.

If that were true, that would be and if that were to continue, that would be extremely alarming for investors in ANT and similarly for investors in OpenAI because these companies still lose money. So there's actually a scenario where if they don't experience rapid growth in revenue, they will literally run out of money as a company and the, the, the valuation, valuation of the company could go down, you know, by a factor of ten or something.

That-that's like a worst-case scenario, but that did happen in the dot-com bubble. So in that earlier tech bubble, there were companies that lost ninety percent or more of their value because the, the value during the bubble was built on extrapolated revenues that never materialized.

Okay? This is a combined I think this is also ramp data. This is combined spend on the two different labs. I think the golden-ish bars are Anthropic and the green bars are OpenAI. And you, you can see that in aggregate, well, for individually and in aggregate their the, the rapid growth has slowed down in the later part of twenty twenty-six post after the summer.

Okay? So again, alarming information. This is what everybody who's thinking about the Anthropic IPO that's what they're loo focused on. Okay? Also, if you're interested in CapEx, if you're interested, if you're buying Nvidia shares, you know, again, all-almost all of the revenue that's flowing through to Nvidia it in a sense originates from people buying tokens to run it, to use, which is a byproduct of using AI.

And if this is flattening out, it could be It could be a negative signal for semiconductors, memory chip makers, people who are building data centers. Okay. Here's another graph from Ramp. And what this graph shows is that, that if you segment companies by how much AI they use, and you look at the top one percent, the companies that really heavily use AI, they're spending seven thousand dollars a month per employee to buy tokens.

That group, the top one percent, dominates all AI expenditures. So most AI expenditures are actually coming from this very select group of companies that are spending seven thousand dollars a month on average. That group of companies is spending seven thousand dollars a month per employee on AI.

Whereas the median company that uses Ramp to, to do their, internal monitoring of financial, finances, et cetera they're only spending twelve dollars per employee. So the median company is spending twelve dollars a month per employee on AI. That's kind of like having an OpenAI subscription, the base subscription, twelve dollars a month.

for maybe for half your employees, you buy a subscription, that would be about twelve dollars a month. that's to be compared with seven thousand, so per month. So, so there's a clear hierarchy in who's actually spending on AI. So it's interesting to know how this is working out. So it doesn't seem like most companies in the US at currently are getting a lot of value from AI.

I would guess in this top one percent group, it's almost all companies that develop software, that do a lot of software development, and that's why they're spending so much money. This is a slide which is particularly relevant if you're interest in interested in the CapEx build-out, if you wanna understand different a-aspects of the data center build-out to supply the forecasted or predicted future demand for AI tokens.

And what's shown here is the implied amount of electrical power That is required under certain assumptions. So if you look at projected GPU sales and you say, "Oh, well, eventually we're gonna have to power all those GPUs," that gives you a certain amount of electrical power that you need, and if that doesn't materialize, your GPUs will not be activated.

Right? So if I look on the bottom here, it's the dark blue bar is, North American power based on accelerator sales volumes. Right? So if, if you say this projected sales volumes for Nvidia will be materialized this, this graph only goes to twenty twenty-eight. Let's say in the next two years their projections are materialized, then those blue bars are you know, fifty-four gigawatts in twenty twenty-eight, thirty-five gigawatts in twenty twenty-seven.

That amount of power has to be added to the grid or at least has to, has to be allocated to these chips in order to run them. And that's a substantial amount of energy. I think the, the number to think in your head is one to two hundred gigawatts. That's the total amount that goes to residences, to people's homes.

So you're talking about an amount of electricity production which is a bit less than, but comparable to all the electricity which is consumed by families in their homes currently. So can we can that much energy be found for AI just in the next two years? Okay. Many people find this implausible, okay, either because the, the, the gross, the aggregate supply of electricity production for the total grid just can't, can't grow that fast, or we can't reallocate to data centers taking it away from factories and, and families.

There's a question of whether this is gonna materialize. If this doesn't materialize, the people who bought those GPUs will not be able to turn them on. Okay? Or maybe these sales will not materialize for Nvidia because people realize, "Hey, I can't actually use these GPUs. I don't have the power. We haven't been able to build out enough data centers and I can't actually turn the chips on."I think that, this is worth looking at because even if everything goes right in the sense that demand for AI tokens grows, as we've been discussing there's a 10X growth in, you know, aggregate dollar demand for AI tokens, it might be the case that the CapEx build-out due to physical constraints of electricity or permitting of data centers overcoming public opposition to data centers, that can't keep up.

So the AI boom could bust not because of lack of adoption by companies or economic benefit from AI tokens. It could be that on the physical build-out, we're not able to add enough power, build enough data centers to supply the demand. So, so this whole thing could fail in multiple ways, right?

Even if demand materializes, maybe the physical CapEx can't keep up. Maybe the physical CapEx is accomplished, but demand is not there and, and the people who paid for the physical CapEx build-out lose money. Okay? So something to keep an eye on. These things are at a level where they affect the whole economy now.

You're talking about percents of total GDP, and so if all of this melts down, it's going to be similar to the 2008 financial crisis or the 2000 dot-com bubble, where it could cause a very, very deep recession or even a depression.

Here's a slide which shows, OpenAI's financials, and they're already valued on the private markets at roughly a trillion dollars, both they and Anthropic. So they're able to raise money at an implied valuation of roughly a trillion dollars. They're hoping to IPO at a premium to that at one and a half or two trillion dollars. Sam Altman can go on the private markets and raise, say, $100 billion at a trillion-dollar valuation, you know, and suffer, say, a 10% dilution for the existing shareholders.

He can already do things like that, but he, he has to do things like that because if you look at their cash burn rate, you can see that already in 2027, they have a negative, according to this analysis they have a negative ending cash balance in these quarters, 2027 quarters. So that would mean that, you know, they're spending more money than they're making from selling tokens.

And if they don't raise money, if they don't bring in on the order of $100 billion, they can't make it through 2027. So they, they have to raise money. Now, if they delay their IPO, as they have in the case of OpenAI, they're gonna have to raise this on the private markets. Anthropic is still, as of, as of now, as far as I know, still planning to have an IPO probably in November, and they will raise they need to raise comparable amounts of money here, as, as shown here, on the public markets And so again, whether this turns out to be a good investment depends on those revenues materializing, not just for the industry as a whole.

They can't materialize and go to open source models. They have to materialize and go to OpenAI and Anthropic in order for this trillion-dollar valuation of OpenAI or two trillion dollar valuation of Anthropic. In order for that to have been a good investment for someone who, say, bought the shares at IPO, in order for that to be a good investment, the 10X or even greater than 10X growth in total AI revenue has to materialize, and it has to go to these labs at significant margins.

It can't be the case that the Chinese models or open models come and competewith them and drive down the margins, in which case maybe a lot of tokens are consumed and the hyperscalers are happy, but OpenAI and Anthropic run out of cash, right? So that's a, that's a scenario that could actually happen.

One of the things that, you know, Jensen Huang has announced, I heard a little bit of color about this at the meeting you know, they're pushing very hard to release their own open source models. Not very people many people know this, but NVIDIA has an open source model called Nemotron, which I think they're going to release a frontier-level open source model, I think called Megatron, and that will directly compete with Claude and GPT.

Why is NVIDIA doing this? Well, it's because NVIDIA doesn't wanna be beholden to a handful of top AI labs. They wanna control their own fate, and one way they can control their own fate is to have a very powerful, competitive, open source model, which no matter what happens to OpenAI or Anthropic, there will still be AI available which will run on NVIDIA GPUs.

And I don't think they can fail at this. I think they have to they will succeed because in the worst case, they can just take Chinese open source models, which are completely open weight, and just improve them, clean them up, say like, "Oh, these are safe now. They don't have any communist propaganda in them.

They will talk about Tiananmen Square to you." Right? So N-NVIDIA could do that even if they didn't have the capability internally, which I think they will have. I think they will hire enough people that they can compete, maybe not at parity with OpenAI, Anthropic, but not too far behind, kind of like the Chinese labs.

And they will ensure that there are trustworthy American or US open source models which are at the frontier. It's in their interest to do this, but that is a problem for OpenAI, Anthropic, the margins for those companies. Also, similar projects probably happening at Microsoft et cetera. So, again even if the overall AI boom plays out as the bulls hope, it doesn't mean that OpenAI and Anthropic are gonna be the winners there.

Here's an example of some an entity that's already being identified as a loser. So Oracle is one of the hyperscalers, probably the one whose circumstances are least favorable. This is the price of credit default swapsCDX on Oracle debt. So Oracle has taken on a huge amount of debt to build out its data centers, and they've announced some negative outcomes, like problems getting their data centers in New Mexico up and running et cetera, et cetera.

And you can see the market is now starting to price in, you know, a non-trivial probability of default or impairment of the debt issued by Oracle or Oracle through SPVs and special vehicles for its data center construction. And this is what happens when bond investors actually start looking carefully at what's happening, not equity investors who are vibes-based.

Even CIOs of big funds but who are mainly equity investors, they can be super vibes-based, okay? Just, "Oh, I believe the Nvidia story. I don't really understand the technical details of AI, but I just really believe in AI." And, and they can, they can make investment decisions on that basis or valuation decisions based on that based on that kind of reasoning, and same with venture capitalists.

But when bond investors get involved, generally they're more sober, they're a lot more conservative, and they're more numbers-driven. And here you're seeing the bond market, or in this case, the credit default swap market, which is insurance on bonds, starting to become worried about the situation at Oracle.

And, and something like this could happen for the debt issued by all the hyperscalers for their build-outs, even for, you know, maybe not for a company like Google, which prints tons of money, but, you know, eventually you could see stuff like this or some of this debt being rated as near junk, not investment grade, right?

So something to look out for if you're watching for the if you think this, this could be a bubble and it's gonna pop, these are the kinds of things that you wanna be looking at. That is roughly what I, what I wanted to cover. For people that are in finance, I probably went through this in such a painfully slow pedagogical way.

You probably were bored, and you, you maybe knew all of this. For people who are not steeped in finance maybe this was useful for you to understand what we're really looking at here in terms of this AI bubble/boom. If you were listening to this on audio only, might wanna go to the YouTube video or to I think I'll probably post these figures on my Substack as well. They're also available on my X feed, so you can, you can see what it was that I was talking about. So thanks a lot for joining me. I'll see you next time on Manifold.