An interesting difference between the AI boom and the dot com boom is that back then companies would IPO to raise capital and might have been forced to IPO when they were beyond 500 investors incl employees. These days these companies have already raised capital via private equity and the laws changed which allow them to not count employees with equity and have much higher investor counts. An IPO by either of the frontier labs companies is really a cash-out and payback on the capital they have borrowed. My thinking is the stock price would go down, possibly significantly, after an IPO.
Oracle's credit rating is hovering just above junk bond level. I think the capital agreements are similar to an Adjustable Rate Mortgage and monthly so if there's another rate increase from the Fed and Oracle can't find anyone to buy their bonds and/or their free cash flow continues to be negative then maybe that is a pin-prick in the stock bubble? Its sort of like if you keep bouncing around from 0% credit card offer for 12 months to another and then you stop getting 0% offers, then its time to face the music (those usually have a 3% transaction fee, at least a long time ago they did).
It's been a funny pairing of DotCom boom investors got burned buying paper with no revenue, and now investors complain about companies staying private till there is profit (or cashflow).
I've been saying for a while now to stay the fuck away from tech IPOs for exactly this reason: it's no longer "we need to raise capital so we can get bigger", it's "we're tapped out on growth, and the last round of private investors needs to be paid back, plus employees need their equity to be worth something in order to stay". Private markets extract all the value and then dump the husk on the public.
Assuming the NASDAQ / S&P aren't gamed into listing these at offering price to game the system into putting ~10% of people's retirement into a junk stock on day 1.
"Buy the index" isn't a solution for human influence and corruption, there are people in control of what goes into the index too.
When a stock hits the public market, it gets incorporated into broad market index funds (subject to the rules of those funds). If an overvalues stock gets dumped onto the public market, then crashes after being bought by index funds, then holders of those index funds are harmed.
You can get rich by stealing pennies, if you steal them from enough people. It is still wrong, even if no individual lost enough to care about going after you.
If enough people try profiting by stealling pennies, the cumulative losses are significant; and can threaten confidence in and the strenght of the market.
Are there examples of this and quantified losses? The only one I can think of is Nasdaq and SpaceX, but Nasdaq isn’t exactly an index lay people should be investing in, and it isn’t commonly owned anyway (maybe less than 1% own it). And I don’t think SpaceX has lost any Nasdaq 100 investors money yet.
I mean that used to be true until companies began strong-arming the indexes to include them on day 1 instead of having a mandatory 1-year waiting period. Now you're taking the bag whether you like it or not.
Ed Zitron might be too dismissive of the potential of LLMs but he seems to be the only popular journalist holding the companies accountable for wonky reports on their profitability.
Though, as time goes on I am increasingly sympathetic to Zitron's take on the potential of LLMs. Because 4 years later it's still just a potential that's perpetually around the corner. Most people I know in meatspace use it rarely, and, outside of coding, the ones who are expected to use it at work tend to resent it.
I'm not sure if I can express how different the dot com boom era was to someone who didn't live through it. America Online kindasorta brought the Web to everyday people's homes around about the beginning of 1995, except that at that time everyday people didn't necessarily even have computers at home. Four years later we were almost living in a different world. Darn near every college student had a computer connected to the Internet in their dorm room. Every office worker had a computer on their desk and conducted at least some of their work in a Web browser. My grandmother was trying to chat me up on instant messages so many times per day that it was starting to get annoying.
It really was quite remarkable. Compared to it, the AI "boom" has been more like a wet diaper fart.
> the ones who are expected to use it at work tend to resent it
And the population at large really resents it. I have not talked to a person under the age of 30 and outside of tech who has anything good to say about AI. That's a group that doesn't vote very reliably but if AI concerns motivate them to do so, they could have an overwhelming impact.
I can’t imagine a machine that solved navier stokes and a bunch of other outstanding problems in math being called “nothing more than a wet diaper fart”
I don’t know if/when AI will ever become indispensable to normies but it’s clearly making a difference in fields that are highly technical.
But that’s not what I said. Even forgetting the straight-up misquote, I think you may have misread the overall intent of the statement. I wasn’t trying to say that GenAI isn’t impactful. I was using colorful language to say that the impact after 4 years pales in comparison to the impact of the Web after 4 years.
The elephant in the room here is that these discussions are still happening at all. In 1999 there was no question of whether the web was going to completely reshape the world, because it already had.
> The elephant in the room here is that these discussions are still happening at all. In 1999 there was no question of whether the web was going to completely reshape the world, because it already had.
I think we're probably more in the early nineties for AI, rather than the late nineties where everyone (for certain values) is getting some value from it.
The tech is definitely going to have a large impact, but the inherent limitations of it (stochastic and expensive) means that we'll need a bunch of downstream tools and deterministic software to make the most of it.
Which means that a lot of "value" is gonna get vaporised before we reach that point.
> Posting this nice list of his failed predictions and bad analysis someone put together: https://danluu.com/zitron/
I like both Ed and Dan (for very different reasons), but I note that the financial side isn't really covered by Dan, and that's probably where Ed has been strongest (to the extent that much of the rest of the financial world is now picking up on it).
It’s hard to argue against what Zitron is focusing on, as it’s just numbers and simple analysis.
What Dan Luu is doing with that post is an act of misdirection (if not clumsy attempt at character assassination): avoiding the entirety of Zitron’s numbers and analysis (which can’t be refuted) and instead narrowly focusing on “predictions” that one can spin a hundred different ways.
If that’s convincing to the extent that you are dismissing Zitron’s numbers and analysis, you have a major critical thinking problem.
It always seems smart to be a contrarian/bear, but historically, betting against technology and technological progress is dumb.
Not only is he not a shareholder, but based on his opinions, he clearly isn't a user of modern models either. Somehow he still thinks they are party tricks and that "They will never be AGI," with no elaboration on _why_ they _can't_ be AGI, and why that actually matters for these companies (hint: it doesn't really matter much. As it turns out, you could replace a lot of humans with a modern LLM and have a reasonable comparative advantage in doing so, even though the models are ostensibly not "AGI.").
His predictions have also historically just been bad (https://danluu.com/zitron/). Even AI 2027, which is basically semi-fictional, is more accurate and worth paying more attention to than this guy. And the superforecaseters which actually made AI 2027 and actual futurists are generally higher signal.
> His predictions have also historically just been bad (https://danluu.com/zitron/). Even AI 2027, which is basically semi-fictional, is more accurate and worth paying more attention to than this guy. And the superforecaseters which actually made AI 2027 and actual futurists are generally higher signal.
As I noted upthread, those predictions don't cover the financial side, which is where a lot of the issues currently are.
Like, LLMs/AI are definitely going to have large impacts, but those impacts will take a long time to materialise (outside of software, and even there we'll need much better tools to make good use of agents), and the financial investment are predicated on near-term returns from the tech that seem wildly implausible to me (and others).
The bunch of crypto miners getting multiple billion dollar valuations should be a clue here that all is probably not well.
And fundamentally, I don't see model building as a very good business. It's incredibly expensive, and it's hard to maintain the margin in a very competitive space. This is probably good for the world, but very bad for the investors into these model companies.
Not being a 'party trick' and being 'AGI' are wildly different assumptions. Apart from AGI being a poorly defined term, I don't see any reason why the LLM methodology has the potential to scale to AGI.
There's other avenues which are probably needed to make the cohesive AGI picture come true, such as World Models. Simply learning from text and being able to generate language is not, in my mind, the end-all-be-all for intelligence.
They are a party trick, provided the party is software development. If your job does not have you sit behind a computer all day, their utility is minimal.
Wouldn't a "correction" be something _all_ AI users (not producers!) celebrate though? That would mean cheaper hardware and access to open models. The dotcom bubble keeps being compared, and the present we have now is a result of that post-dotcom bubble!
So I don't understand why there's fear and disagreement over a correction. Both sides should welcome it as there is some benefit to everyone. Well, most of everyone.
This article is fairly standard financial news boilerplate. "Market selloff, thesis is being tested" is an article that gets printed roughly twice a week regardless of what is going on, and is a template that's been run thousands of times.
But is it like collector gold coins -- those sell for a premium over their intrinsic gold for numismatic value .
I suspect there people who are betting they can arbitrage them, such as getting an allotment of them and renting them out. (I'm not talking about the scale of a single card in your house, but in DC operators or those who have racks leased already)
No, it's not. GPUs deliver value, real value, compute is the new oil(tm).
But also, it is. You can resell them.
I find it mind boggling that either AI is a bubble or AI is the next industrial revolution and no other options can be considered because reasons(tm).
Token costs are dropping, yes they are, but how much of that is better software and algorithms vs better hardware and that's too galaxy brain for the simpletons trying to predict what's next.
So what's next? IMO more tokens and lots of disappointment that more tokens alone neither cures cancer nor brings the apocalypse and further waves of innovation. AI continues apace.
Daily reminder that the Industrial Revolution was characterized in part by a series of market bubbles that formed in response to the introduction of exciting new technologies.
And it got us to today after we went through some things. And today is better than then what with the germ theory of disease, the genetic code, modern agriculture, and HVAC. Haters gonna hate, but there's nothing else they can do about that. But that won't stop them from trying, no sir!
Short term, AI is overrated. Long term, you ain't seen nuthin' yet... Now vote down that positivity, we're doomed! DOOMED!
Going on about a 30X PE ratio as excessive in an age of 336(TSLA), 178(PLTR), 154(AMD) etc...
You guys never stop.
But 20 is a classical and fundamentally healthy PE ratio for a healthy business and that would value Anthropic at ~$1.3T if that's what you're getting at and the early employees would all still be billionaires.
Mind you, I hate both Anthropic and OpenAI for their conduct in this space, but I wouldn't argue with the numbers. I'd argue with the blatant fearmongering towards regulatory capture but that's galaxy brain nonsense I know.
> But 20 is a classical and fundamentally healthy PE ratio for a healthy business
This essentially means that it takes twenty years to recover the price through earnings. The US market has been super strange since the GFC, and with higher interest rates, I don't see these prices continuing unless the machine god is invented real soon.
Listing a bunch of tech stocks’ high P/E ratios doesn’t seem likely to convince someone who’s criticizing tech stock P/E ratios for being to high that tech stock P/E ratios are not too high.
I gave up on convincing people the sky is blue or that they need to breathe if they don't want to suffocate a while ago. And now we all have personal chatbots to confirm every single one of our biases on demand, hooray!
But the tech industry PE ratio is about 28. Since 30 is crazy, what's a reasonable PE ratio and when are you going to short the NASDAQ on margin and make your first billion, you market genius you?
I’m not expressing an intent to short anything; I’m pointing out that that argument didn’t carry any rhetorical weight. A stronger argument for tech stocks tending toward higher P/E ratios than many other market segments would be something like suggesting they have higher upside potential because it’s still a growth industry.
I find it fascinating that your response to me pointing out that these two positions are not opposed because they can quite easily both be right at the same time is to use quite hyperbolic language to accuse me of holding hyperbolic opinions.
Rest assured that it did not, in fact, disabuse me of my belief that I’m at least partially successful in retaining a level head.
AI at the edge, i.e. running on your own hardware are going to be the new hotness.
But those data centers? Craters. Wouldn't be surprised if China buys them up for cents on the dollar and run expensive, polluting ops out of them instead of in China.
- The cost of protecting CoreWeave’s debt against default for five years rose by as much as 74.5 basis points to 1194.5 basis points, according to ICE Data Services. That’s the highest intraday level on record with swaps pricing in a more than 60% chance of default within the next five years.
- Meanwhile, some of the company’s bonds declined in the US junk market. Its 9% notes maturing in 2031 fell by about 2.5 cents on the dollar to 80.75 cents as of 11:00 a.m., pushing the yield on the debt to 15.3%, according to Trace. Its 9.25% bonds due 2030 dropped by about 0.75 cents to about 86 cents, pushing the yield to 14.3%. At the same time, some notes edged higher, with its 9.75% notes due 2031 rising 0.125 cents to 80.625 cents.
- “Debt investors are freaked out about CoreWeave’s cost of equity and equity investors are freaked out about its cost of debt,” said Bill Zox, portfolio manager at Franklin Templeton Fixed Income.
- However, D.A. Davidson analyst Gil Luria, who has referred to CoreWeave as “the subprime tranche” of AI, warned on a podcast that the stock could go to zero if the AI funding boom falters and the company becomes unable to pay its expensive debt obligations.
- Luria told Bloomberg Television on Wednesday that CoreWeave was “destroying value at every turn” by borrowing at such elevated rates.
- “If CoreWeave is borrowing at 9% and the best return they can get off of that is a 5% return, it doesn’t matter how much they scale,” he said.
- “Borrowing at very high interest rates in order to build data centers that are increasingly commoditized outside of the hyperscalers may not end up being a good investment,” he continued. “The more leverage we use to build data centers, the more risk we’re bringing into the system.”
Monitor CoreWeave equity and debt pricing to understand where we are in the AI capital bubble cycle.
Does it really matter to the big players? Aren’t they betting on AI providing super intelligence and therefore providing larger returns than the rates at which they can borrow?
Similar stuff happened with the unsustainable computer-leasing mania of the late 1960s. The president had to get involved so that they would stop borrowing more and more cash. I don’t think they really can stop borrowing if other competitors keep throwing cash on the AI build out. No one wants to be left behind and therefore will continue borrowing - even with the rising rates.
Think of it more in terms of market cap. Google is worth 4.3 trillion dollars. They're spending on AI to protect that and others are spending on AI in the hopes of becoming that. The golden goose is serving ads and the fight is over who gets to do that.
Yes, but 7-8 companies have shown they can build that goose. LLMs are going to end up like databases or operating systems. It's very hard to compete with free.
The value is the user data that collects overtime and using that to become sticky for customers
I mean, yes, that's a serious problem for the particular firms involved in it, but I don't really care which of them turn their owners into quadrillionaires. Ultimately, one of them is still going to eat us.
It matters if the debt they're attempting to originate doesn't math at elevated interest rates. For a personal example, how long can you last maxing out your credit cards at 20-30% interest on your wage income? Your earnings will support the debt load until the debt spiral eventually races ahead of your earnings.
In this context, debt load and interest costs continue to increase due to bond market fundamentals while earnings are not.
> First, the favorable impact of the artificial intelligence investment boom on economic activity and earnings will likely diminish significantly in 2027. That’s because what’s relevant for growth is how much investment is increasing, not its level. The increase in investment in 2026 will almost certainly be the peak. There aren’t sufficient resources — construction workers, electrical generation capacity, or chip manufacturing capacity - to increase investment by the same magnitude in 2027. Nor are the dominant hyperscalers likely to have the free cash flow and balance sheet capacity to sustain a bigger increase in investment in 2027 compared with 2026.
> Second, as the growth of investment spending slows, the growth in earnings of hyperscaler suppliers will falter, profit expectations will diminish and price-earnings ratios will shrink. The “picks and shovels” providers will suffer a double whammy - slower demand growth and profit margin compression. On the way up, higher demand leads to wider profit margins that sustain equity market valuations. On the way down, the outlook for earnings deteriorates quickly as the shortfall of demand relative to expectations is exacerbated by a collapse in profit margins.
> Third, as the investment cycle matures, the focus will shift to the returns that the hyperscalers are expected to earn on their massive investments. I suspect it will be difficult for the AI hyperscalers to generate sufficient revenue ($2 trillion or more per year) to generate the returns needed to justify an AI capital base that is likely to reach $5 trillion.
(I'd encourage you to read the entire piece, it was written by Bill Dudley, a former president of the Federal Reserve Bank of New York, and is too much to quote in its entirety)
I'm of the belief that the Fed raising rates won't do much at all. There's no reason to believe that raising rates will combat inflation coming from oil prices that fluctuate based on tweets and bombing, and it seems that at this point Coreweave is dealing with high interest rates not from the Fed raising rates, but because the market is waking up and realizing that there's a hell of a lot of hype in datacenters and that risk hasn't been priced accordingly.
Raising rates will have one main effect. Raising prices.
Most of the drivers right now of inflation are in-elastic demand. Raising the cost of borrowing, will not deter that spend.
What it will do, is actually drive up the cost of all business, and worsen employment simultaneously. You can typically see a rise in prices (inflation) following a rate hike. Even when it's needed to combat regular elastic demand side inflation, it takes time to have a positive effect. If the main driver of the inflation is inelastic, or rate insensitive as much of the driving force right now is, hiking rates will actually increase inflation.
The AI boom doesn't care what rate it has to pay, it will pay it, and keep driving things up.
What's needed is a significant policy shift, to make room in the markets again for things society needs financing for (ex. building homes, productive revenue generating businesses)
>What it will do, is actually drive up the cost of all business, and worsen employment simultaneously.
That's the point of raising interest rates. It's demand destruction. If raising interest rates didn't raise the cost of doing business, it wouldn't work.
>You can typically see a rise in prices (inflation) following a rate hike. Even when it's needed to combat regular elastic demand side inflation, it takes time to have a positive effect.
Are you sure it's not just reverse causation? eg. inflation has gotten so out of hand that bankers are raising interest rates to counteract it? After all, inflation isn't a random walk. If it's bad one month, it's likely to get worse the next, especially when central banks haven't raised interest rates yet.
The way I see it it's the other way around, rates need to be raised to adjust to the situation. If the actual yield of treasuries is higher than the rates set by the fed due to AI bonds flooding the market and investors requiring more pay from countries running a 5% deficit While in a boom cycle AND strong economy AND high inflation both with and without energy and food in the accounting then the fed just has to hike to not lose their credibility.
Now, I don't expect them to hike too much beyond what the markets price 1-2yr bonds at, that would, like you say do nothing as a lot of inflation comes from the oil disruption.
Never comes back for various reasons (structural demographics and savings glut having been drained, broadly speaking). Easy to see this based on forward curves of various sovereign debt durations.
Not investing advice, but you're potentially better off paying more towards the principal with excess cashflow versus waiting for lower rates that are unlikely to arrive.
One thing we know for sure -- A huge portion of the S&P 500 is exposed, the average consumer is pretty tapped for credit, and if people's stock accounts / 401ks go down they naturally retract their spending some. Which means a recession across all but necessities.
Plus loan rates are high meaning rational people either pay less for, or completely avoid buying houses, cars, and all the things that go with (furniture, yard maintenance tools etc)
Great comment. Additional downstream impact is that those who feel wealth ("wealth effect" [1]) from an inflated SP500 will no longer feel as wealthy and will reduce spend accordingly when a correction occurs, further slowing an economy at stall speed without AI capital investment.
And they will hold on to unrealistically high expectations on the value of their properties.
I was talking to a realtor I bumped into last month, he said that houses are staying on the market longer because sellers don't want to accept the offers they are getting. In some cases people are upside down or close to it.
It has been a long time since people have had to deal with the idea that their house will do anything other than increase in value. He said he thought up to a 50% "adjustment" in home prices is coming.
Presumably the creditors would then own the DCs. The price to build them is a sunk cost (meaning: not actually a factor in determining what they are now worth), so they would either sell them off for their new value or else run them at market rates.
The big question then becomes: what is the market value of these things? Is it even positive? Creditors will eat the difference between that number and whatever they actually lent (less what they've been paid back so far, if anything).
actually, because AI is the only industry that can monetize so high, interest rates will go up and all other companies except for AI will be hit. Change in traditional economics, humans can't understand exponential tech.
I am mildly invested ( figuratively and literally so take my next words with that in mind ) in the outcome and raised rates are certainly a possible outcome ( and oddly, one I was suggesting for the longest time as all but inevitable ), but, and it is not a small but, there is a reason those data centers are being built that that reason did not change much. I am just sayin'.
It is possible I was being too subtle. If the management and owners truly do believe that AI is the beginning of the end of the current ecosystem, then change in rates won't change that much, because it will be pushed through regardless.
An interesting difference between the AI boom and the dot com boom is that back then companies would IPO to raise capital and might have been forced to IPO when they were beyond 500 investors incl employees. These days these companies have already raised capital via private equity and the laws changed which allow them to not count employees with equity and have much higher investor counts. An IPO by either of the frontier labs companies is really a cash-out and payback on the capital they have borrowed. My thinking is the stock price would go down, possibly significantly, after an IPO.
Oracle's credit rating is hovering just above junk bond level. I think the capital agreements are similar to an Adjustable Rate Mortgage and monthly so if there's another rate increase from the Fed and Oracle can't find anyone to buy their bonds and/or their free cash flow continues to be negative then maybe that is a pin-prick in the stock bubble? Its sort of like if you keep bouncing around from 0% credit card offer for 12 months to another and then you stop getting 0% offers, then its time to face the music (those usually have a 3% transaction fee, at least a long time ago they did).
It's been a funny pairing of DotCom boom investors got burned buying paper with no revenue, and now investors complain about companies staying private till there is profit (or cashflow).
I've been saying for a while now to stay the fuck away from tech IPOs for exactly this reason: it's no longer "we need to raise capital so we can get bigger", it's "we're tapped out on growth, and the last round of private investors needs to be paid back, plus employees need their equity to be worth something in order to stay". Private markets extract all the value and then dump the husk on the public.
The "public" is completely unharmed as long as they stick to broad market index funds. Only the gamblers are harmed.
Assuming the NASDAQ / S&P aren't gamed into listing these at offering price to game the system into putting ~10% of people's retirement into a junk stock on day 1.
"Buy the index" isn't a solution for human influence and corruption, there are people in control of what goes into the index too.
When a stock hits the public market, it gets incorporated into broad market index funds (subject to the rules of those funds). If an overvalues stock gets dumped onto the public market, then crashes after being bought by index funds, then holders of those index funds are harmed.
Obviously diluted though by the 499 other stocks in the portfolio (using VFIAX/VOO as an example).
You can get rich by stealing pennies, if you steal them from enough people. It is still wrong, even if no individual lost enough to care about going after you.
If enough people try profiting by stealling pennies, the cumulative losses are significant; and can threaten confidence in and the strenght of the market.
Are there examples of this and quantified losses? The only one I can think of is Nasdaq and SpaceX, but Nasdaq isn’t exactly an index lay people should be investing in, and it isn’t commonly owned anyway (maybe less than 1% own it). And I don’t think SpaceX has lost any Nasdaq 100 investors money yet.
I mean that used to be true until companies began strong-arming the indexes to include them on day 1 instead of having a mandatory 1-year waiting period. Now you're taking the bag whether you like it or not.
Ed Zitron might be too dismissive of the potential of LLMs but he seems to be the only popular journalist holding the companies accountable for wonky reports on their profitability.
Though, as time goes on I am increasingly sympathetic to Zitron's take on the potential of LLMs. Because 4 years later it's still just a potential that's perpetually around the corner. Most people I know in meatspace use it rarely, and, outside of coding, the ones who are expected to use it at work tend to resent it.
I'm not sure if I can express how different the dot com boom era was to someone who didn't live through it. America Online kindasorta brought the Web to everyday people's homes around about the beginning of 1995, except that at that time everyday people didn't necessarily even have computers at home. Four years later we were almost living in a different world. Darn near every college student had a computer connected to the Internet in their dorm room. Every office worker had a computer on their desk and conducted at least some of their work in a Web browser. My grandmother was trying to chat me up on instant messages so many times per day that it was starting to get annoying.
It really was quite remarkable. Compared to it, the AI "boom" has been more like a wet diaper fart.
> the ones who are expected to use it at work tend to resent it
And the population at large really resents it. I have not talked to a person under the age of 30 and outside of tech who has anything good to say about AI. That's a group that doesn't vote very reliably but if AI concerns motivate them to do so, they could have an overwhelming impact.
I can’t imagine a machine that solved navier stokes and a bunch of other outstanding problems in math being called “nothing more than a wet diaper fart”
I don’t know if/when AI will ever become indispensable to normies but it’s clearly making a difference in fields that are highly technical.
But that’s not what I said. Even forgetting the straight-up misquote, I think you may have misread the overall intent of the statement. I wasn’t trying to say that GenAI isn’t impactful. I was using colorful language to say that the impact after 4 years pales in comparison to the impact of the Web after 4 years.
The elephant in the room here is that these discussions are still happening at all. In 1999 there was no question of whether the web was going to completely reshape the world, because it already had.
> The elephant in the room here is that these discussions are still happening at all. In 1999 there was no question of whether the web was going to completely reshape the world, because it already had.
I think we're probably more in the early nineties for AI, rather than the late nineties where everyone (for certain values) is getting some value from it.
The tech is definitely going to have a large impact, but the inherent limitations of it (stochastic and expensive) means that we'll need a bunch of downstream tools and deterministic software to make the most of it.
Which means that a lot of "value" is gonna get vaporised before we reach that point.
I don't consider him a journalist that I would listen to ever.
Posting this nice list of his failed predictions and bad analysis someone put together: https://danluu.com/zitron/
I think he's a broken clock that's right twice a day.
> Posting this nice list of his failed predictions and bad analysis someone put together: https://danluu.com/zitron/
I like both Ed and Dan (for very different reasons), but I note that the financial side isn't really covered by Dan, and that's probably where Ed has been strongest (to the extent that much of the rest of the financial world is now picking up on it).
That is fair but he is definitely right that Anthropic is not releasing a real set of numbers on their business
It’s hard to argue against what Zitron is focusing on, as it’s just numbers and simple analysis.
What Dan Luu is doing with that post is an act of misdirection (if not clumsy attempt at character assassination): avoiding the entirety of Zitron’s numbers and analysis (which can’t be refuted) and instead narrowly focusing on “predictions” that one can spin a hundred different ways.
If that’s convincing to the extent that you are dismissing Zitron’s numbers and analysis, you have a major critical thinking problem.
We are still waiting for at least one time of the day.
The sensibility remains intact when you’re not a shareholder.
It always seems smart to be a contrarian/bear, but historically, betting against technology and technological progress is dumb.
Not only is he not a shareholder, but based on his opinions, he clearly isn't a user of modern models either. Somehow he still thinks they are party tricks and that "They will never be AGI," with no elaboration on _why_ they _can't_ be AGI, and why that actually matters for these companies (hint: it doesn't really matter much. As it turns out, you could replace a lot of humans with a modern LLM and have a reasonable comparative advantage in doing so, even though the models are ostensibly not "AGI.").
His predictions have also historically just been bad (https://danluu.com/zitron/). Even AI 2027, which is basically semi-fictional, is more accurate and worth paying more attention to than this guy. And the superforecaseters which actually made AI 2027 and actual futurists are generally higher signal.
> His predictions have also historically just been bad (https://danluu.com/zitron/). Even AI 2027, which is basically semi-fictional, is more accurate and worth paying more attention to than this guy. And the superforecaseters which actually made AI 2027 and actual futurists are generally higher signal.
As I noted upthread, those predictions don't cover the financial side, which is where a lot of the issues currently are.
Like, LLMs/AI are definitely going to have large impacts, but those impacts will take a long time to materialise (outside of software, and even there we'll need much better tools to make good use of agents), and the financial investment are predicated on near-term returns from the tech that seem wildly implausible to me (and others).
The bunch of crypto miners getting multiple billion dollar valuations should be a clue here that all is probably not well.
And fundamentally, I don't see model building as a very good business. It's incredibly expensive, and it's hard to maintain the margin in a very competitive space. This is probably good for the world, but very bad for the investors into these model companies.
Not being a 'party trick' and being 'AGI' are wildly different assumptions. Apart from AGI being a poorly defined term, I don't see any reason why the LLM methodology has the potential to scale to AGI.
There's other avenues which are probably needed to make the cohesive AGI picture come true, such as World Models. Simply learning from text and being able to generate language is not, in my mind, the end-all-be-all for intelligence.
They are a party trick, provided the party is software development. If your job does not have you sit behind a computer all day, their utility is minimal.
You would have made a lot of money betting against technology back in 2000 ... assuming you got the timing right.
There are a bunch of ways that AI the technology could be massivly successful, while AI the industry crashes.
Wouldn't a "correction" be something _all_ AI users (not producers!) celebrate though? That would mean cheaper hardware and access to open models. The dotcom bubble keeps being compared, and the present we have now is a result of that post-dotcom bubble!
So I don't understand why there's fear and disagreement over a correction. Both sides should welcome it as there is some benefit to everyone. Well, most of everyone.
He is totally uninformed about even basic progress in the performance of these models. Seems like used free ChatGPT in 2022 and never updated once.
Looks like this time wasn't different after all.
This article is fairly standard financial news boilerplate. "Market selloff, thesis is being tested" is an article that gets printed roughly twice a week regardless of what is going on, and is a template that's been run thousands of times.
AI midterm is coming.
It sold off 1% yesterday and then went back up 0.5% today, hardly a signal.
In other AI news, 5090s are going for $9000 apiece. AI is so over, amIRight?
But is it like collector gold coins -- those sell for a premium over their intrinsic gold for numismatic value .
I suspect there people who are betting they can arbitrage them, such as getting an allotment of them and renting them out. (I'm not talking about the scale of a single card in your house, but in DC operators or those who have racks leased already)
It's not speculators, it's groups exporting to china where they're moving the chips to new cards for use in data centers. [0][1]
Probably related, china banned imports of the 5090d in may.
- https://videocardz.com/newz/heres-why-rtx-5090-costs-5000-ai...
- https://videocardz.com/newz/inside-chinas-mass-conversion-of...
No, it's not. GPUs deliver value, real value, compute is the new oil(tm).
But also, it is. You can resell them.
I find it mind boggling that either AI is a bubble or AI is the next industrial revolution and no other options can be considered because reasons(tm).
Token costs are dropping, yes they are, but how much of that is better software and algorithms vs better hardware and that's too galaxy brain for the simpletons trying to predict what's next.
So what's next? IMO more tokens and lots of disappointment that more tokens alone neither cures cancer nor brings the apocalypse and further waves of innovation. AI continues apace.
Daily reminder that the Industrial Revolution was characterized in part by a series of market bubbles that formed in response to the introduction of exciting new technologies.
And it got us to today after we went through some things. And today is better than then what with the germ theory of disease, the genetic code, modern agriculture, and HVAC. Haters gonna hate, but there's nothing else they can do about that. But that won't stop them from trying, no sir!
Short term, AI is overrated. Long term, you ain't seen nuthin' yet... Now vote down that positivity, we're doomed! DOOMED!
And even if right about the direction, and it comes to fruition with amazing changes/capabilities, investors can still be wrong about the price.
Roughly speaking Paying $2T for even $65Bn _profit_ is a bad deal, historically speaking. (~30x P/E)
s/investors/speculators/g
But also, ~20*$65B is ~1.3 billion and a healthy PE ratio just like Cisco. but that's not good for your narrative, is it?
Oh no! all those early founders who thought they'd be worth $3.8B will only be worth $2.6B! Whatever shall they do?
> but that's not good for your narrative, is it?
My narrative? What narrative is that?
and 30x 65 = 1950 which is pretty close to 2T.
I'm not sure what you're on about or if you're just taking the piss?
Going on about a 30X PE ratio as excessive in an age of 336(TSLA), 178(PLTR), 154(AMD) etc...
You guys never stop.
But 20 is a classical and fundamentally healthy PE ratio for a healthy business and that would value Anthropic at ~$1.3T if that's what you're getting at and the early employees would all still be billionaires.
Mind you, I hate both Anthropic and OpenAI for their conduct in this space, but I wouldn't argue with the numbers. I'd argue with the blatant fearmongering towards regulatory capture but that's galaxy brain nonsense I know.
> But 20 is a classical and fundamentally healthy PE ratio for a healthy business
This essentially means that it takes twenty years to recover the price through earnings. The US market has been super strange since the GFC, and with higher interest rates, I don't see these prices continuing unless the machine god is invented real soon.
Listing a bunch of tech stocks’ high P/E ratios doesn’t seem likely to convince someone who’s criticizing tech stock P/E ratios for being to high that tech stock P/E ratios are not too high.
I gave up on convincing people the sky is blue or that they need to breathe if they don't want to suffocate a while ago. And now we all have personal chatbots to confirm every single one of our biases on demand, hooray!
But the tech industry PE ratio is about 28. Since 30 is crazy, what's a reasonable PE ratio and when are you going to short the NASDAQ on margin and make your first billion, you market genius you?
I’m not expressing an intent to short anything; I’m pointing out that that argument didn’t carry any rhetorical weight. A stronger argument for tech stocks tending toward higher P/E ratios than many other market segments would be something like suggesting they have higher upside potential because it’s still a growth industry.
I find it fascinating that your response to me pointing out that these two positions are not opposed because they can quite easily both be right at the same time is to use quite hyperbolic language to accuse me of holding hyperbolic opinions.
Rest assured that it did not, in fact, disabuse me of my belief that I’m at least partially successful in retaining a level head.
Nothing would warm the cockles of my heart like OpenAI, Anthropic, SpaceX, and Tesla getting marked to market.
But it wouldn't be good for podcast culture, now would it?
AI at the edge, i.e. running on your own hardware are going to be the new hotness.
But those data centers? Craters. Wouldn't be surprised if China buys them up for cents on the dollar and run expensive, polluting ops out of them instead of in China.
And "business.com" sold for $7m in 1999... Stupidly inflated "assets" aren't a proof but if anything they prove the opposite of your point
Don't understand this, they're totally outclassed by datacenter tier cards so is it just hobbyists driving these prices?
Apparently a good chunk of 5090s have had their chips ripped off to make frankencards for those who can't get DC cards.
The GPU poor have zero access to the datacenter cards. But they can make some of it up in volume with multiple consumer cards because NUMA.
You're comparing a putative leading indicator to what's almost certainly a trailing indicator.
It should be expected the Fed will continue to raise rates.
US Producer Prices Up Most Since 2022 as Energy Pain Worsens - https://www.bloomberg.com/news/articles/2026-10-15/us-produc... - October 15th, 2026
CoreWeave Credit Risk Hits Record, Pricing 60% Odds of Default - https://www.bloomberg.com/news/articles/2026-10-15/coreweave... - October 15th, 2026
- The cost of protecting CoreWeave’s debt against default for five years rose by as much as 74.5 basis points to 1194.5 basis points, according to ICE Data Services. That’s the highest intraday level on record with swaps pricing in a more than 60% chance of default within the next five years.
- Meanwhile, some of the company’s bonds declined in the US junk market. Its 9% notes maturing in 2031 fell by about 2.5 cents on the dollar to 80.75 cents as of 11:00 a.m., pushing the yield on the debt to 15.3%, according to Trace. Its 9.25% bonds due 2030 dropped by about 0.75 cents to about 86 cents, pushing the yield to 14.3%. At the same time, some notes edged higher, with its 9.75% notes due 2031 rising 0.125 cents to 80.625 cents.
- “Debt investors are freaked out about CoreWeave’s cost of equity and equity investors are freaked out about its cost of debt,” said Bill Zox, portfolio manager at Franklin Templeton Fixed Income.
CoreWeave's Bonds Turn Distressed, Drawing 'Yikes' from Musk - https://news.ycombinator.com/item?id=50090791 - October 2026
- However, D.A. Davidson analyst Gil Luria, who has referred to CoreWeave as “the subprime tranche” of AI, warned on a podcast that the stock could go to zero if the AI funding boom falters and the company becomes unable to pay its expensive debt obligations.
- Luria told Bloomberg Television on Wednesday that CoreWeave was “destroying value at every turn” by borrowing at such elevated rates.
- “If CoreWeave is borrowing at 9% and the best return they can get off of that is a 5% return, it doesn’t matter how much they scale,” he said.
- “Borrowing at very high interest rates in order to build data centers that are increasingly commoditized outside of the hyperscalers may not end up being a good investment,” he continued. “The more leverage we use to build data centers, the more risk we’re bringing into the system.”
Monitor CoreWeave equity and debt pricing to understand where we are in the AI capital bubble cycle.
Does it really matter to the big players? Aren’t they betting on AI providing super intelligence and therefore providing larger returns than the rates at which they can borrow?
Similar stuff happened with the unsustainable computer-leasing mania of the late 1960s. The president had to get involved so that they would stop borrowing more and more cash. I don’t think they really can stop borrowing if other competitors keep throwing cash on the AI build out. No one wants to be left behind and therefore will continue borrowing - even with the rising rates.
It’s a tough situation for the FED to maneuver.
Think of it more in terms of market cap. Google is worth 4.3 trillion dollars. They're spending on AI to protect that and others are spending on AI in the hopes of becoming that. The golden goose is serving ads and the fight is over who gets to do that.
what's strange is that the more I interact with gemini and claude, the fewer ads I see.
No, the golden goose is ownership over the magic box that does work cheaper than people do.
That market is far bigger than the entire ad market put together, because its upper boundary is every dollar the world spends on white-collar payroll.
Yes, but 7-8 companies have shown they can build that goose. LLMs are going to end up like databases or operating systems. It's very hard to compete with free.
The value is the user data that collects overtime and using that to become sticky for customers
I mean, yes, that's a serious problem for the particular firms involved in it, but I don't really care which of them turn their owners into quadrillionaires. Ultimately, one of them is still going to eat us.
It matters if the debt they're attempting to originate doesn't math at elevated interest rates. For a personal example, how long can you last maxing out your credit cards at 20-30% interest on your wage income? Your earnings will support the debt load until the debt spiral eventually races ahead of your earnings.
In this context, debt load and interest costs continue to increase due to bond market fundamentals while earnings are not.
Bloomberg: It Won’t Take Much to Burst the Stock Market Bubble - https://www.bloomberg.com/opinion/articles/2026-08-20/us-sto... | https://archive.today/O9sQE - August 20th, 2026
> First, the favorable impact of the artificial intelligence investment boom on economic activity and earnings will likely diminish significantly in 2027. That’s because what’s relevant for growth is how much investment is increasing, not its level. The increase in investment in 2026 will almost certainly be the peak. There aren’t sufficient resources — construction workers, electrical generation capacity, or chip manufacturing capacity - to increase investment by the same magnitude in 2027. Nor are the dominant hyperscalers likely to have the free cash flow and balance sheet capacity to sustain a bigger increase in investment in 2027 compared with 2026.
> Second, as the growth of investment spending slows, the growth in earnings of hyperscaler suppliers will falter, profit expectations will diminish and price-earnings ratios will shrink. The “picks and shovels” providers will suffer a double whammy - slower demand growth and profit margin compression. On the way up, higher demand leads to wider profit margins that sustain equity market valuations. On the way down, the outlook for earnings deteriorates quickly as the shortfall of demand relative to expectations is exacerbated by a collapse in profit margins.
> Third, as the investment cycle matures, the focus will shift to the returns that the hyperscalers are expected to earn on their massive investments. I suspect it will be difficult for the AI hyperscalers to generate sufficient revenue ($2 trillion or more per year) to generate the returns needed to justify an AI capital base that is likely to reach $5 trillion.
(I'd encourage you to read the entire piece, it was written by Bill Dudley, a former president of the Federal Reserve Bank of New York, and is too much to quote in its entirety)
I'm of the belief that the Fed raising rates won't do much at all. There's no reason to believe that raising rates will combat inflation coming from oil prices that fluctuate based on tweets and bombing, and it seems that at this point Coreweave is dealing with high interest rates not from the Fed raising rates, but because the market is waking up and realizing that there's a hell of a lot of hype in datacenters and that risk hasn't been priced accordingly.
Raising rates will have one main effect. Raising prices.
Most of the drivers right now of inflation are in-elastic demand. Raising the cost of borrowing, will not deter that spend.
What it will do, is actually drive up the cost of all business, and worsen employment simultaneously. You can typically see a rise in prices (inflation) following a rate hike. Even when it's needed to combat regular elastic demand side inflation, it takes time to have a positive effect. If the main driver of the inflation is inelastic, or rate insensitive as much of the driving force right now is, hiking rates will actually increase inflation.
The AI boom doesn't care what rate it has to pay, it will pay it, and keep driving things up.
What's needed is a significant policy shift, to make room in the markets again for things society needs financing for (ex. building homes, productive revenue generating businesses)
>What it will do, is actually drive up the cost of all business, and worsen employment simultaneously.
That's the point of raising interest rates. It's demand destruction. If raising interest rates didn't raise the cost of doing business, it wouldn't work.
>You can typically see a rise in prices (inflation) following a rate hike. Even when it's needed to combat regular elastic demand side inflation, it takes time to have a positive effect.
Are you sure it's not just reverse causation? eg. inflation has gotten so out of hand that bankers are raising interest rates to counteract it? After all, inflation isn't a random walk. If it's bad one month, it's likely to get worse the next, especially when central banks haven't raised interest rates yet.
The way I see it it's the other way around, rates need to be raised to adjust to the situation. If the actual yield of treasuries is higher than the rates set by the fed due to AI bonds flooding the market and investors requiring more pay from countries running a 5% deficit While in a boom cycle AND strong economy AND high inflation both with and without energy and food in the accounting then the fed just has to hike to not lose their credibility.
Now, I don't expect them to hike too much beyond what the markets price 1-2yr bonds at, that would, like you say do nothing as a lot of inflation comes from the oil disruption.
I'm just waiting for us to hit the ZIRP/NIRP forever stage so I can refi my mortgage
Never comes back for various reasons (structural demographics and savings glut having been drained, broadly speaking). Easy to see this based on forward curves of various sovereign debt durations.
https://www.ustreasuryyieldcurve.com/
From a Savings Glut to a Savings Shortage - https://www.apollo.com/wealth/insights-news/insights/daily-s... - September 12th, 2026
Not investing advice, but you're potentially better off paying more towards the principal with excess cashflow versus waiting for lower rates that are unlikely to arrive.
https://www.hsh.com/prepayment-refinance-calculator.html
has anyone written about what happens if all these billion-dollar-a-pop AI datacenters are actually built
and then the finance paying workload never materializes?
so they just declare bankruptcy and what society just absorbs it?
One thing we know for sure -- A huge portion of the S&P 500 is exposed, the average consumer is pretty tapped for credit, and if people's stock accounts / 401ks go down they naturally retract their spending some. Which means a recession across all but necessities.
Plus loan rates are high meaning rational people either pay less for, or completely avoid buying houses, cars, and all the things that go with (furniture, yard maintenance tools etc)
Great comment. Additional downstream impact is that those who feel wealth ("wealth effect" [1]) from an inflated SP500 will no longer feel as wealthy and will reduce spend accordingly when a correction occurs, further slowing an economy at stall speed without AI capital investment.
[1] https://en.wikipedia.org/wiki/Wealth_effect
I agree, but also the housing market is already pretty dead because current owners have very low rates on their mortgages.
And they will hold on to unrealistically high expectations on the value of their properties.
I was talking to a realtor I bumped into last month, he said that houses are staying on the market longer because sellers don't want to accept the offers they are getting. In some cases people are upside down or close to it.
It has been a long time since people have had to deal with the idea that their house will do anything other than increase in value. He said he thought up to a 50% "adjustment" in home prices is coming.
Yes but, tariffs will keep replacement cost high due to materials. And if we keep ICE-ing out the hardest workers.. /shrug.
But if people start losing their jobs and cant/wont make the payments like in 2008 , could drop hard.
If you declare bankruptcy, your creditors absorb whatever losses are unrecoverable. So it would depend on who is lending them the money.
Presumably the creditors would then own the DCs. The price to build them is a sunk cost (meaning: not actually a factor in determining what they are now worth), so they would either sell them off for their new value or else run them at market rates.
The big question then becomes: what is the market value of these things? Is it even positive? Creditors will eat the difference between that number and whatever they actually lent (less what they've been paid back so far, if anything).
Rhymes with "1800s railroads".
> Rhymes with "1800s railroads".
If you had to redo one of the most expensive parts of laying all that track in order to use it, yeah.
Hardware that's specialized for building LLMs or for running LLMs isn't same sort of hardware you'd use to -say- serve up newegg.com.
I knew of the "Great Depression" but you made me learn much more interesting detail
* https://en.wikipedia.org/wiki/Panic_of_1873#Factors
actually, because AI is the only industry that can monetize so high, interest rates will go up and all other companies except for AI will be hit. Change in traditional economics, humans can't understand exponential tech.
I am mildly invested ( figuratively and literally so take my next words with that in mind ) in the outcome and raised rates are certainly a possible outcome ( and oddly, one I was suggesting for the longest time as all but inevitable ), but, and it is not a small but, there is a reason those data centers are being built that that reason did not change much. I am just sayin'.
>I am just sayin'.
What, exactly, are you just sayin'? Cause I'm having a stroke trying to parse your comment lol
It is possible I was being too subtle. If the management and owners truly do believe that AI is the beginning of the end of the current ecosystem, then change in rates won't change that much, because it will be pushed through regardless.