As with all of these financial manias, the goal is to get the big pension funds to come on board so that you are "too big to fail" (i.e., you can pressure the government for a bail out when it all comes crashing down). The people leading the charge are well aware of 2001, 2008, 2020 --- they know the game: privatise the profits and socialise the losses.
Agreed. When the bubble bursts, the weak will be discarded and the strong agents distributed and consolidated among The Big Tech Bros in a replay of Sub Prime. What I'd like to know is, how does the government bail out work? Is it as simple as adding a zero? I've been waiting to see the "cliff" for years and it always seems to recede just over the horizon.
"Data centers in the middle of nowhere are useless without AI."
While I agree with the overall thrust of the article, this one isn't quite true. Even aside from AI, demand for cloud resources has generally continued to increase, and it mostly doesn't matter where those are physically located.
Now that being said, the kind of hardware fine tuned for AI is often not optimal for general purpose computing. It's not going to be a 1 to 1, turnkey conversion. Still, the data centers themselves have SOME value in a post-AI market. Is it in the zillions like it's being rated now? No. But it's not nothing.
I recently had to repair my iPhone XR. Since I don’t use the cloud to back up data, all contacts and photos were lost. So? No big deal, there was much data no longer useful. Gradually I added back the data that WAS useful as text messages from friends and relatives were received after repairs.
As for the photos, like eliminating ghosts, no big deal. If I want a trip down memory lane, that’s my brain’s domain. 😎
Besides we need to call out and refer to “Data Centers” for what they are; data warehouses for creating the future panopticon to watch us, the prisoners.
One silver lining is the newfound public acceptance of nuclear power needed for big computing power. If Three Mile Island will make my phone smarter, fire it up!
Yeah, as far as I know there’s nothing about data centers that restricts them to AI. To talk about them “going dark” if AI turns out to be a nothingburger sounds to me like worrying that we shouldn’t build electrical plants or highways because people might decide to stay home and read by candlelight.
Yes we do but as Bismarck said, "the only thing we learn from history is that we learn nothing from history!" Of course the banksters learned that they got bailed out and will again.
The money engineers and conjurers have become more fantastical than the invisible man in the sky conjurers ie purveyors of religion. AI makes my research so much easier that it makes me even lazier. With the experience of Deep Seek training their model for a small fraction of Open AI I don't see the US being able to compete with AI anymore than anything else. No doubt we'll see even more nefarious uses than the friends and lovers machines it's doing now. Of course it can blow up financially but I worry more about what it can do to a very sick society and I don't see anything positive. We're social animals that need real relationships and movement through nature, ALL difficult and so easy to let the machine do it for us. It's like some kind of opiate, making you feel good but depriving you of a REAL life for convenience.
My experience with Google Gemini includes a considerable amount of subjective bias. We need to remind ourselves that just because an answer is derived from AI doesn’t ensure its accuracy.
Oh don't get me wrong, I'm not saying other models are necessarily more accurate. Quite the opposite. LLMs have major inherent problems that simply cannot be solved. Everything they say is a "hallucination", it's basically coincidence that some of it happens to be correct. (This is an oversimplification, but hopefully gets the point across.)
All I meant was that while Deepseek may have an efficiency advantage, I don't believe it's any more accurate.
The power requirements alone should at least slow down this AI dream. I'm thinking of the Black Mirror episode where young adults were locked in huge warehouses to pedal generators so elites could have warm fondue.
Open AI is a question mark, the rest of the companies involved actually have solid businesses - unlike the late 90's bubble.
There's also a national security aspect that wasn't part of the 90's internet bubble or the mid-aughts real estate bubble. Whoever gets to self-improving AI first will be in an unimaginably controlling position, so it had better be us and not China.
Perhaps then if the tech is great but does not produce enough revenue to meet and surpass the costs, will government step in and give some sort of bailout.....is the technology too big to fail? Not being snarky, you make a great point
Not too big to fail, too important to risk falling behind. The government can't do much except cheer from the sidelines, it's really up to the innovators to keep us ahead.
Super excellent point. Sure, Webvan and pets.com died, but they took huge bites out of healthy companies, as you state. Could/most-likely-will happen again.
The numbers are off by orders of magnitude. Subsidies are driving data center growth, AI growth is a media fallacy. Before AI they said it was bitcoin.
Spend 5 minutes doing a back of the envelope calculation. How many tokens per day are processed? How much energy per token? Compare with current data center energy usage? AI is only using a fraction of 1% of the data center capacity.
Back in the bitcoin days, you could divide the daily total data center energy usage by the daily number of bitcoin transaction and it would cost something like $10,000 in electricity to make a single bitcoin transaction according to the media reports.
What is really happening is the political parties have rigged up subsidy regimes to pay the data centers to consume power, so they can't build them fast enough. Most likely they are running infinite loops to generate usage reports that are turned in for subsidies.
A big giveaway is that the data center power draw is constant 24/7/365. If it were human generated usage, the load would drop at night and on the weekends.
They could be queueing up processes to run on off hours. Data clean up and management, model training and testing, etc.. All of these can run automatically and be queued to fill in down time.
Where's the leverage? Yes, investments can go bad. But the only thing worth noting from this article is that it won't cause a banking crisis because the banks have stayed at arm's length.
I don't see how all these investments will work out well, but I didn't think Amazon would ever be much more than a book retailer. The argument that hapless pension funds will lose money isn't that compelling -- should every retirement dollar be invested in commercial real estate? What is your investment alternative?
Yesterday there was an article on Bloomberg on how private credit was getting squeezed by banks in the lending space because lending spreads have compressed (on syndicated loans) so much that the math doesn't work for PC.
Here was a quote, "That has left many private credit funds with idle cash - and falling returns - as they struggle to invest the billions they raised during the boom years."
That scares me.
My concern isn't so much that retirement dollars get invested in these A.I. strategies, it is the amount of the investment, and left to their own devices, Private Asset managers will over do it, by a lot. Just my opinion.
In my experience when fund concepts don’t work out and the investment thesis can’t be realized, the funds often close up and return investor dollars or fold into a different fund, for which investor approval is needed, so that quote doesn’t concern me. This stuff happens all the time.
Not just your opinion, Eric. The push for retail investors to get access to PE markets means the same, that there will be excess money that must be invested, so very iffy deals will be done.
See Oracle's 500 debt-to-equity ratio and $325 billion of corporate debt, plus the off balance sheet / private credit debt structures. It's only a problem if the ultimate lenders are "systemically important", and taxpayers are forced to bail them out (again). The banks are not arms length: they are funding private credit. And pension funds and insurance companies are getting in on the game, too.
This is madness. The world economy is built on commerce driven by working people. The house of cards collapses when companies racing to leverage AI have no more customers. We are doomed.
"Data centers in the middle of nowhere (MoN) are useless without AI." That's not quite true. If the MoN has cheaper resources (energy, labor, real estate) and sufficient fiber links to the Internet backbone, the MoN is a good place for a data center.
An insightful essay but there's one assumption in it I question. Right now use is high because it's essentially free (and Google and others push it on the unwary user). Is there any reason to believe use will continue, much less grow, once the actual costs of this technology is charged to the end user? My take is no.
Innovative tech always becomes profitable eventuality, doesn’t history show this? I read some business reports saying current efforts in AI will replace 1 in 12 workers. I had an interaction with a “customer advocate” AI on the phone at United Healthcare who couldn’t resolve my issue so transferred me to a live person in another company. Hilariously the live person couldn’t tell that it was an AI and asked it for its name 😭. The real tech development “space race” is in the field of developing “super AI General Intelligence”, not LLM AIs. Read the book “If anyone builds it, everyone dies” a good primer on the AI in general & Super AI General Intelligence. Scary stuff
Regardless, it will push innovation forward, probably to everyone’s ultimate benefit as has been the case with most innovation. And I would expect any military oriented R&D to continue and probably be profitable enough to justify ongoing investment.
This is really concerning, very high risk. I’ve actually been considering divesting from the stock & bond markets, not least because of this exact situation.
However, we should remember that not all AI is large language models and much of the less famous concepts/applications are far more interesting.
Glass half full, maybe this time it works out better — maybe AI becoming less compute intensive ramps up competition and there is a robust market. Hell, maybe the super exciting (in 2019) 5G network investments that have been floundering will finally become as necessary and useful as they were promised to be.
you have to wonder how much vital infrastructure trillions of dollars might buy. Tertiary water treatment is somehow a luxury, while phantasmal data priorities are on track to get trillions of dollars.
As with all of these financial manias, the goal is to get the big pension funds to come on board so that you are "too big to fail" (i.e., you can pressure the government for a bail out when it all comes crashing down). The people leading the charge are well aware of 2001, 2008, 2020 --- they know the game: privatise the profits and socialise the losses.
Agreed. When the bubble bursts, the weak will be discarded and the strong agents distributed and consolidated among The Big Tech Bros in a replay of Sub Prime. What I'd like to know is, how does the government bail out work? Is it as simple as adding a zero? I've been waiting to see the "cliff" for years and it always seems to recede just over the horizon.
Is AI advanced enough that we can ask it if this is all a good idea?
“Beignet” is a perfect name: yummy but no nutritional value and you’re starving again in 10 minutes!!
And completely empty inside.
But tasty!
"Data centers in the middle of nowhere are useless without AI."
While I agree with the overall thrust of the article, this one isn't quite true. Even aside from AI, demand for cloud resources has generally continued to increase, and it mostly doesn't matter where those are physically located.
Now that being said, the kind of hardware fine tuned for AI is often not optimal for general purpose computing. It's not going to be a 1 to 1, turnkey conversion. Still, the data centers themselves have SOME value in a post-AI market. Is it in the zillions like it's being rated now? No. But it's not nothing.
I recently had to repair my iPhone XR. Since I don’t use the cloud to back up data, all contacts and photos were lost. So? No big deal, there was much data no longer useful. Gradually I added back the data that WAS useful as text messages from friends and relatives were received after repairs.
As for the photos, like eliminating ghosts, no big deal. If I want a trip down memory lane, that’s my brain’s domain. 😎
Besides we need to call out and refer to “Data Centers” for what they are; data warehouses for creating the future panopticon to watch us, the prisoners.
Yeah, who needs art? Oh, you’re just a “picture taker”. Never mind, carry on…
One silver lining is the newfound public acceptance of nuclear power needed for big computing power. If Three Mile Island will make my phone smarter, fire it up!
Yeah, as far as I know there’s nothing about data centers that restricts them to AI. To talk about them “going dark” if AI turns out to be a nothingburger sounds to me like worrying that we shouldn’t build electrical plants or highways because people might decide to stay home and read by candlelight.
"How will rating agencies such as S&P and Moody’s arrive at their ratings?"
I'm pretty sure we already know the answer to that one...
Yes we do but as Bismarck said, "the only thing we learn from history is that we learn nothing from history!" Of course the banksters learned that they got bailed out and will again.
The money engineers and conjurers have become more fantastical than the invisible man in the sky conjurers ie purveyors of religion. AI makes my research so much easier that it makes me even lazier. With the experience of Deep Seek training their model for a small fraction of Open AI I don't see the US being able to compete with AI anymore than anything else. No doubt we'll see even more nefarious uses than the friends and lovers machines it's doing now. Of course it can blow up financially but I worry more about what it can do to a very sick society and I don't see anything positive. We're social animals that need real relationships and movement through nature, ALL difficult and so easy to let the machine do it for us. It's like some kind of opiate, making you feel good but depriving you of a REAL life for convenience.
My own experience of Deepseek so far is that it's rather overhyped. It's just a faster way of getting the wrong answers.
My experience with Google Gemini includes a considerable amount of subjective bias. We need to remind ourselves that just because an answer is derived from AI doesn’t ensure its accuracy.
Oh don't get me wrong, I'm not saying other models are necessarily more accurate. Quite the opposite. LLMs have major inherent problems that simply cannot be solved. Everything they say is a "hallucination", it's basically coincidence that some of it happens to be correct. (This is an oversimplification, but hopefully gets the point across.)
All I meant was that while Deepseek may have an efficiency advantage, I don't believe it's any more accurate.
No misunderstanding. I totally agree.
A friend mentioned to me the other day that she gets mail with a disclaimer at the bottom: “This was generated by AI so there might be errors.” !?!?
I think I'll start using AI as a universal excuse.
I read that there will actually be AI porn. So kind of like anime porn, but with “life-like” participants. That’s just fucked up.
Never thought naked women could be replaced.
The power requirements alone should at least slow down this AI dream. I'm thinking of the Black Mirror episode where young adults were locked in huge warehouses to pedal generators so elites could have warm fondue.
They should get the Slovenian cyclist, Tadej Podecar, on a bike. His power output could warm all the fondue in Switzerland.
🫕
Open AI is a question mark, the rest of the companies involved actually have solid businesses - unlike the late 90's bubble.
There's also a national security aspect that wasn't part of the 90's internet bubble or the mid-aughts real estate bubble. Whoever gets to self-improving AI first will be in an unimaginably controlling position, so it had better be us and not China.
Perhaps then if the tech is great but does not produce enough revenue to meet and surpass the costs, will government step in and give some sort of bailout.....is the technology too big to fail? Not being snarky, you make a great point
Not too big to fail, too important to risk falling behind. The government can't do much except cheer from the sidelines, it's really up to the innovators to keep us ahead.
How are "innovators" and government rent seekers different in this case? If the government finances the innovators, costs will triple.
Private capital from many sources is doing the financing.
Not to be all gloomy and doomy but...
Microsoft is one of those solid companies now. And it was also solid in 2000. Still it lost 75% of its market value between 2000 and 2009.
But it always generated cash. Just because the multiple came down, it didn't mean that Microsoft couldn't finance its projects.
Super excellent point. Sure, Webvan and pets.com died, but they took huge bites out of healthy companies, as you state. Could/most-likely-will happen again.
The numbers are off by orders of magnitude. Subsidies are driving data center growth, AI growth is a media fallacy. Before AI they said it was bitcoin.
Spend 5 minutes doing a back of the envelope calculation. How many tokens per day are processed? How much energy per token? Compare with current data center energy usage? AI is only using a fraction of 1% of the data center capacity.
Back in the bitcoin days, you could divide the daily total data center energy usage by the daily number of bitcoin transaction and it would cost something like $10,000 in electricity to make a single bitcoin transaction according to the media reports.
What is really happening is the political parties have rigged up subsidy regimes to pay the data centers to consume power, so they can't build them fast enough. Most likely they are running infinite loops to generate usage reports that are turned in for subsidies.
A big giveaway is that the data center power draw is constant 24/7/365. If it were human generated usage, the load would drop at night and on the weekends.
They could be queueing up processes to run on off hours. Data clean up and management, model training and testing, etc.. All of these can run automatically and be queued to fill in down time.
But AI is only a fraction of 1% of the total data center usage. So what is the cpu being spent on?
Infinite loops can also be queued up, and that would be consistent with the near 100% cpu occupancy, 24/7/365, at every single data center.
Where's the leverage? Yes, investments can go bad. But the only thing worth noting from this article is that it won't cause a banking crisis because the banks have stayed at arm's length.
I don't see how all these investments will work out well, but I didn't think Amazon would ever be much more than a book retailer. The argument that hapless pension funds will lose money isn't that compelling -- should every retirement dollar be invested in commercial real estate? What is your investment alternative?
Yesterday there was an article on Bloomberg on how private credit was getting squeezed by banks in the lending space because lending spreads have compressed (on syndicated loans) so much that the math doesn't work for PC.
Here was a quote, "That has left many private credit funds with idle cash - and falling returns - as they struggle to invest the billions they raised during the boom years."
That scares me.
My concern isn't so much that retirement dollars get invested in these A.I. strategies, it is the amount of the investment, and left to their own devices, Private Asset managers will over do it, by a lot. Just my opinion.
In my experience when fund concepts don’t work out and the investment thesis can’t be realized, the funds often close up and return investor dollars or fold into a different fund, for which investor approval is needed, so that quote doesn’t concern me. This stuff happens all the time.
Not just your opinion, Eric. The push for retail investors to get access to PE markets means the same, that there will be excess money that must be invested, so very iffy deals will be done.
See Oracle's 500 debt-to-equity ratio and $325 billion of corporate debt, plus the off balance sheet / private credit debt structures. It's only a problem if the ultimate lenders are "systemically important", and taxpayers are forced to bail them out (again). The banks are not arms length: they are funding private credit. And pension funds and insurance companies are getting in on the game, too.
Appreciate the contributors to Racket like Eric Salzman. 🙌🏽
This is madness. The world economy is built on commerce driven by working people. The house of cards collapses when companies racing to leverage AI have no more customers. We are doomed.
"Data centers in the middle of nowhere (MoN) are useless without AI." That's not quite true. If the MoN has cheaper resources (energy, labor, real estate) and sufficient fiber links to the Internet backbone, the MoN is a good place for a data center.
In my field, especially arcana I know pretty well, AI is wrong a good deal of the time. I just ignore it at this point.
An insightful essay but there's one assumption in it I question. Right now use is high because it's essentially free (and Google and others push it on the unwary user). Is there any reason to believe use will continue, much less grow, once the actual costs of this technology is charged to the end user? My take is no.
that is the fear.....great tech but is it profitable?
Innovative tech always becomes profitable eventuality, doesn’t history show this? I read some business reports saying current efforts in AI will replace 1 in 12 workers. I had an interaction with a “customer advocate” AI on the phone at United Healthcare who couldn’t resolve my issue so transferred me to a live person in another company. Hilariously the live person couldn’t tell that it was an AI and asked it for its name 😭. The real tech development “space race” is in the field of developing “super AI General Intelligence”, not LLM AIs. Read the book “If anyone builds it, everyone dies” a good primer on the AI in general & Super AI General Intelligence. Scary stuff
Regardless, it will push innovation forward, probably to everyone’s ultimate benefit as has been the case with most innovation. And I would expect any military oriented R&D to continue and probably be profitable enough to justify ongoing investment.
This is really concerning, very high risk. I’ve actually been considering divesting from the stock & bond markets, not least because of this exact situation.
However, we should remember that not all AI is large language models and much of the less famous concepts/applications are far more interesting.
Glass half full, maybe this time it works out better — maybe AI becoming less compute intensive ramps up competition and there is a robust market. Hell, maybe the super exciting (in 2019) 5G network investments that have been floundering will finally become as necessary and useful as they were promised to be.
you have to wonder how much vital infrastructure trillions of dollars might buy. Tertiary water treatment is somehow a luxury, while phantasmal data priorities are on track to get trillions of dollars.