The AI dooooooom thread

[crazy canuck]

This century has been a profound disappointment.

Yep.

And it became disappointing almost immediately. There were very few times when it even teased us that things might work out well.

I dunno, up till 2008 things were looking to be on track.

No, things were looking to be on track up until September 11, 2001.

[crazy canuck]

Four of the tech industry's wealthiest companies made it clear this week that their spending on artificial intelligence was not about to slow down.

But the outlays from Google, Meta, Microsoft and Amazon — which all raised their spending by billions of dollars, saying they needed to meet demand for A.I. — are increasingly feeding concerns that the tech industry is heading toward a dangerous bubble.

Artificial intelligence remains an unproven and expensive technology that could take years to fully develop. How much companies will ultimately get back in return from A.I. products like chatbots is unclear. And smaller companies pursuing A.I. gold, financial analysts pointed out, are not nearly as wealthy.

Last week, the Bank of England wrote that while the building of data centers, which provide computing power for A.I., had so far largely come from the cash produced by the biggest companies, it would increasingly involve more debt. If A.I. underwhelms — or the systems ultimately require far less computing — there could be growing risk.

For the rest of the article

https://www.nytimes.com/2025/10/31/technology/ai-spending-accelerating.html?unlocked_article_code=1.xk8.4zu2.pTkWIH--g3Cj&smid=url-share

[Jacob]

The Atlantic on Here's How the AI Crash Happens

[Josquius]

This century has been a profound disappointment.

Yep.

And it became disappointing almost immediately. There were very few times when it even teased us that things might work out well.

I dunno, up till 2008 things were looking to be on track.

No, things were looking to be on track up until September 11, 2001.

Maybe things were different in NA, but in the UK things were still very much on the up despite the period of Islamic paranoia.
The economy continued to improve, life steadily became better, deprived areas rebuilt, etc...
It really wasn't until the financial crisis that things truly broke and austerity sent things plumetting.
This is shown by data and my own experiences.

[Sheilbh]

I think it very much depends where you're looking from - the famous elephant curve to include post-crash:


Most of us, like most of the Western middle class happen to broadly be in that B section and it's not been great. A great example from Branko Milanovic, is in Italy as they've not grown over the last two decades. The lowest decile of Italians in 1988 were in the 73rd global percentile; by 2008 lower income Italians are now down to the 56th global percentile. By now they're probably lower (possibly even in the bottom half of the world). There are similar processes across the west, Italy's just an extreme. Obviously for some real income actually fell, but for many it didn't - it's just the rest of the world is growing faster in a fundamental and transformative way. I think this is tied to both migration and tourism - and the pressures they're producing - we're moving from a world of the 90s when they were basically luxury goods that were limited to one where they are available to a huge global middle class of which we are just one part.

In other respects I similarly don't know. There's techno-pessimism and understandably, but the lives of many millions of people have been transformed by technology in the last twenty five years. Some of that reflects very admirable policies - first to mind is PEPFAR and the huge downgrading of AIDS as a global health risk from the turn of the century. But others just spreading of technology like phones across Africa and India helping change people's lives and also tying people in to global networks of remittances but also the famous example of farmers being able to get better, more up to date info from their phones. The world is probably less democratic within states compared to 2000, but power is distributed more democratically between states (because of that economic story) - I think there's going to be a dynamic between those two and both the national and individual desire for recognition.

I think there are the conditions of technology and more equal global distribution of wealth and power for a better and more hopeful world (albeit one that I think would be disorienting for the West). But also for a far more alarming future. The thing that freaks me out most to be honest is the increasing fusion of state and corporate power around the world.

[Razgovory]

This century has been a profound disappointment.

Yep.

And it became disappointing almost immediately. There were very few times when it even teased us that things might work out well.

I was telling my coworker, who is only 18, how when I was 18* the biggest political crisis was the President getting a blow job.  She had no idea what I was talking about.  She didn't know Bill Clinton was impeached and how most everyone just laughed it off.  I miss the days when politics was simply stupid rather than stupid and scary.

*Technically I think I was 16 when that happened, but you get the idea.

[Sheilbh]

The Atlantic on Here's How the AI Crash Happens

I don't fully get the focus on data centres as the risk to be honest. It seems to me like it's investing in the physical infrastructure of the digital world. There's been a 500% increase in data centres over the last 10-20 years. I think if we build compute power we will find uses for it. But als othere's increasing regulatory and consumer pressure for data localisation which means there's a need for expanded capacity everywhere.

It may not end up being AI but - I don't know - it sort of feels a bit like worrying about overcapacity in the context of China building EVs or solar. I'm not sure there is such a thing at this point. As I say it may not ultimately be used for AI as we currently know it. But I think there may be something to the analogy discussed by Krugman and Wolf in that piece I linked to earlier of electricity - it actually took about 20 years after being able to commercially generate electricity to work out how to use it in a commercially viable way. And I feel this sort of thing is like building the grid of the next stage of our economy even if we're not fully sure at this point how to use it (and that the companies betting on the physical infrastructure are, at this stage, probably making a better gamble than those working on the end use products).

I fully get the concerns around debt and the weirdnesses going on in the debt markets at the minute (particularly the relationships of private debt markets with the real finance sector) and the incestuousness of money flowing around the AI world. But I think data centres are maybe just a way to physically explain that rather than a source of risk in themseleves?

[The Minsky Moment]

A data center is just a physical location housing equipment.  What makes the data center different from say a power plant is that the equipment depreciates MUCH faster.  We still draw power from plants that have been functioning for decades.  There are some upgrades and replacement of equipment of course, but it's not like in computing where 4 year old components are laughably obsolete. An overbuilt power plant still supplies power even if it brings in less revenue then expected. An overbuilt data center quickly becomes an indoor junkyard.

Another example would be the telecom crash of 2000: companies like Worldcom and Global Crossing built real capacity that had real value; but it didn't stop a very hard crash when they got over-extended.  And again - high capacity cables take quite a bit longer to hit true obsolescence than GPUs.

The piece confuses the issue because financial crashes result from weaknesses in the financing.  The piece alludes to this but what raises red alarms is the circular vendor financing at extreme levels.  Nvidia gives credit to Open AI, etc.  which then uses that money to buy Nvidia chips, which raises Nvidia's reported income and financing capacity, which then prompts Nvidia to do even more ambitious vendor financing.  All fine unless there is a hiccup in the music . . .

Minsky categorized the business cycle as one that progresses through differing stages of finance - starting with "hedge finance" where businesses can get financing backed by current income only with a margin for security. As the boom progresses, it is possible to get financing up to the full amount of current income projected into the future. Last stage is "Ponzi finance" where current income flows aren't enough to service loans and the lender is counting on future growth in market cap to be repaid.

Because Open AI and most of the startups are "pre-earnings" companies, the billions of vendor financing they are getting are extreme manifestations of Ponzi finance.  The counter to this is that: (1) the AI startups are able to access the capital markets for real money, and (2) a lot of the demand is coming from far more substantial backers like MSFT or Amazon that have ample resources.  That is somewhat reassuring but not really.  Capital markets provide ample financing until they don't and the faucet can be closed off quite quickly and brutally. As for big players like MSFT or Amazon, they may look imposing. But Amazon is basically AWS - the rest of its business, however impressive looking, is small potatoes financially.  To what extent is AWS's profitabilty come to be dependent on the AI boom?  Similar questions could be asked of Microsoft and the other big tech groups.

When panics come, they can be just as circular and self-reinforcing as the booms at their origin. Right now, we are witnessing the fallout from a vendor financing driven bankruptcy at First Brands involving a few billion.  We are already deep into the hundreds of billions in the US AI boom.

[Zoupa]

What could possibly go wrong?

https://www.businessinsider.com/even-top-generals-are-looking-to-ai-chatbots-for-answers-2025-10

Even top generals are looking to AI chatbots for answers

It's not just the civilian corporate executives and white-collar workers who are leaning into the generative AI boom at work. Military leaders are diving in too.

The top US Army commander in South Korea shared that he is experimenting with generative AI chatbots to sharpen his decision-making, not in the field, but in command and daily work.

He said "Chat and I" have become "really close lately."

"I'm asking to build, trying to build models to help all of us," said Maj. Gen. William 'Hank' Taylor, commanding general of the 8th Army, told reporters during a media roundtable at the annual Association of the United States Army conference in Washington, DC, on Monday.

Taylor said he's using the tech to explore how he makes military and personal decisions that affect not just him but the thousands of soldiers he oversees. While the tech is useful, though he acknowledged that keeping up with the pace of such rapidly developing technology is an enduring challenge.

"As a commander, I want to make better decisions," the general shared. "I want to make sure that I make decisions at the right time to give me the advantage."

AI in the military

Commanders like Taylor are focused on fast decision-making and how AI could provide an advantage because of a thought process popular with military leaders known as the "OODA Loop." The theory, developed by US fighter pilots during the Korean War, posits that troops who can move decisively before the enemy does — and observe, orient, decide, and act— often have the advantage on the battlefield.

The US military is embracing artificial intelligence with a recognition that decisions in future combat may need to be made faster than humans can make them.

The former Secretary of the Air Force said last year that he doesn't think the people saying that AI technology is "going to determine who's the winner in the next battlefield" are "all that far off." He also wrote that with the advancement of highly automated, highly autonomous kill chains, "response times to bring effects to bear are very short."

Predicting what future war will look like, he said that "we're going to be in a world where decisions will not be made at human speed. They're going to be made at machine speed."

AI is being integrated into drone tech, targeting, and data processing, among other capabilities — an AI algorithm has even piloted a modified F-16 through a simulated dogfight — but the military use of AI is not restricted to combat platforms.

Special Operations Forces, for instance, have sought to "reduce the cognitive burden of our operators" through the use of AI tools for paperwork, situation reports, concepts of operation, managing key supply and logistics demands, and other back-end work.

Operators have employed AI to analyze Pentagon doctrine, improve search functions, and make it easier for personnel who are transferring to a new location or position to catch up on the job and requirements quickly.

There are clear applications at the leadership level as well. Bianca Herlory, the Joint Staff AI lead, said at a panel event in April that "AI can significantly enhance the Joint Staff's ability to integrate and analyze global military operations, ultimately enabling better, faster decisions."

Using generative AI also comes with questions, especially in decisions at the command level. The Pentagon has urged caution as troops and leaders explore these tools, warning that generative AI can leak sensitive data. It can also produce deeply flawed answers if not adequately trained, and that could prove risky and even problematic if commanders use it to inform certain high-stakes decisions.

Things can go wrong with everything, and I imagine even more so in the military than other places, given that rival countries are doing their best to make things go wrong for you.  That's not a reason enough to let your rivals be the first ones to try out revolutionary technologies.

You can't outsource thinking.

[Iormlund]

The Atlantic on Here's How the AI Crash Happens

I don't fully get the focus on data centres as the risk to be honest. It seems to me like it's investing in the physical infrastructure of the digital world. There's been a 500% increase in data centres over the last 10-20 years. I think if we build compute power we will find uses for it. But als othere's increasing regulatory and consumer pressure for data localisation which means there's a need for expanded capacity everywhere.

It may not end up being AI but - I don't know - it sort of feels a bit like worrying about overcapacity in the context of China building EVs or solar. I'm not sure there is such a thing at this point. As I say it may not ultimately be used for AI as we currently know it. But I think there may be something to the analogy discussed by Krugman and Wolf in that piece I linked to earlier of electricity - it actually took about 20 years after being able to commercially generate electricity to work out how to use it in a commercially viable way. And I feel this sort of thing is like building the grid of the next stage of our economy even if we're not fully sure at this point how to use it (and that the companies betting on the physical infrastructure are, at this stage, probably making a better gamble than those working on the end use products).

I fully get the concerns around debt and the weirdnesses going on in the debt markets at the minute (particularly the relationships of private debt markets with the real finance sector) and the incestuousness of money flowing around the AI world. But I think data centres are maybe just a way to physically explain that rather than a source of risk in themseleves?

In addition to what Minsky said, there's also the matter of sheer scale. Even if one of these companies hits the jackpot, they have to eventually return a profit. Given that investments are now in the trillions, how high will an eventual subscription go? What happens when you've gutted your staff and rely on the AI service vendor to operate your business?