There is an AI bubble. It will burst. When is unknown, but think about this

[Bernardo de La Paz]

1. AI is a BFD.

2. AI is only going to get stronger. That is a main reason why it is a BFD. But not by any smooth function. There will be fits and starts.

3. AI hopes and adoption will get bigger until they get smaller. The latter phase is an AI winter: a reduction in funding and emphasis. There have been several AI winters so far, over the past 50 years, each less severe than the one before. The coming one will follow the same pattern. It will be milder and not as long as the last one.

4. After the AI winter will come an AI summer. That is another reason this is a BFD. You ain't seen nuthin' yet.

5. This late stage AI summer does have a bubble. Let me count the ways:

A: Valuations are very high: Nvidia is over 30 times projected earnings and that is kind of mild and probably the best justified of the whole bunch of companies. Palantr is one of the worst in regards to valuations, but it is memish because of its tRumpian connections and the hard right politics of the foul-mouthed CEO. The pure AI companies are making pennies for every dollar they burn.

B. The stock market is top-heavy with the Magnificent Seven, which all, more or less, are in head-over-heels on AI. Nvidia is the purest AI play but at some time next year or two or three after it may face an industry wide retrenchment and be stuck with an over-supply. Wang is an everyman-to-everyone business genius at moving a brilliant company forward, but retrenchments are very different. Tesla is betting on self-driving, overpromised, especially with their cheaper technology. Waymo seems safer. Amazon is minting money with AWS and is going to make its own chips (good luck). Meta is all user (advertisers) and cattle (who do get a unique service). Google may be split and seems undervalued but may be a good horse on the outside. Apple is habitually late, but always aims to be as much best as possible which is a strategy hard to bet against.

B2. All of those heavyweights are investing mind-boggling sums of money in data centers and research and marketing, the whole nine yards, but especially data centers and the power supplies for them (reactors, disadvantaged neighbourhoods, suffering Texan utility customers). So not only is it a significant stock market bubble but also, and partly because, it is a significant part of the economy. A contraction in capital expenditures will have ripple effects which will reduce advertising expenditures, leading to further contraction in cap-ex.

C. AI has been over-promised and over-sold. Each AI summer has delivered on the over-promise of the previous AI summer and made it's own over-promises. The current AI summer is delivering on the speech recognition and synthesis promised in the previous summer. It is now promising replacing human agents and decision makers. The next AI summer will perfect human assistants with limited decision making ability, sometimes broad decision making where no direct human issues are involved.

D. Businesses are beginning to realize that:
. . i) Performance is problematic. Faults are too numerous.
. . ii) Applicability is limited in modes and areas. Human supervision and interaction can be heavy beyond basic chores / actions.
. . iii) Deployment is expensive and requires planning and testing.
. . iv) A recent study (by MIT?) has shown that companies are not seeing as many benefits as they thought they would.

But

You ain't seen nuthin' yet. It will astound us as much as rock music astounded our parents and grandparents.

{ bucolic_frolic asked the question that got me going: https://www.democraticunderground.com/11216481 }

[bucolic_frolic]

1. I saw on a Bravos Research video on YouTube yesterday

If you invested $10,000 in AMD in 2015, it would be worth $1,000,000 today.

If you invested $10,000 in NVDA in 2015, it would be worth $3,000,000 today.

Fortunes will be made.

[Bernardo de La Paz]

14. And fortunes will be lost

For every $10,000 in AMD or NVDA there's tens of thousands lost in dozens of other companies.

[bucolic_frolic]

18. True, which is why you just can't randomly speculate or use hot tips from GoogleFi

Investors need to understand the company, strategy, landscape, value proposition, growth path, unique tech expertise. A rocketship proposition, in other words. And at a price that makes sense.

I recall Micron MU in the mid-80s at 15 cents a share. Many pundits screamed buy! But you had to do it, to understand they why behind their motivation. And their tech wasn't even that unique, several companies made memory chips. But Micron was the best of the lot in a growing niche of demand.

[Alice Kramden]

2. The cost to the environment is very high,

Due to water and electricity consumption

[madville]

3. A major hurdle is energy

AI and quantum computing consume massive amounts of electricity 24/7. China is now constructing coal and nuclear power plants dedicated solely to data centers. Amazon’s new data center in Pennsylvania requires 800 MW of power availability, enough to power for 200-300 thousand households.

AI and quantum computing are the biggest threats to the environment we’ve seen in any of our lifetimes. The race between China and the US in these fields will likely require and consume more power in the next decade than has been generated in human history up to this point.

[Bernardo de La Paz]

15. Yes to your first paragraph. Your last sentence is untrue hyperbole. . . . nt

[madville]

24. By 2028 data centers will consume 45%

Of all power generated in the U.S. China is on the same path. In 10 years it will be multiple times that. Microsoft and Meta are in the process of building their own nuclear power plants to power AI, many can’t comprehend how fast this is going to move.

[Bernardo de La Paz]

25. Do you have a link for your 2028 figure? Biden's DOE says 7 to 12 pc by 2028

DOE Releases New Report Evaluating Increase in Electricity Demand from ...
Dec 20, 2024 The report finds that data centers consumed about 4.4% of total U.S. electricity in 2023 and are expected to consume approximately 6.7 to 12% of total U.S. electricity by 2028. The report indicates that total data center electricity usage climbed from 58 TWh in 2014 to 176 TWh in 2023 and estimates an increase between 325 to 580 TWh by 2028.

[highplainsdem]

4. Rock music was, is, and always will be a good thing. Generative AI, the kind of AI hyped now, is the

most harmful non-weapon tech ever developed - and the sooner the bubble bursts and takes out the AI companies and the venture capitalists keeping them afloat, the better.

[cbabe]

5. Citizens turn against ai data centers

https://futurism.com/small-towns-ai-data-centers

Small Towns Are Rising Up Against AI Data Centers
"We don’t want to be the next Data Center Alley."

MAY 4, 11:00 AM EDT
by Joe Wilkins

…

The facilities working behind the scene to fuel the AI revolution are bulky, noisy, and hog resources like electricity at a staggering scale. Data centers have been blamed for placing a huge burden on local electrical grids and water tables that were only designed for small-town homes, not state-of-the-art industrial facilities.

That being the case, it's no surprise that local opposition to data centers is growing rapidly in rural areas of states like Indiana, Virginia, Missouri, and Illinois, where AI developers are flocking to lock down cheap land and generous tax breaks. And as local organizers successfully repel the monstrous sites, their battleplans are being chronicled and shared with other activists engaged in the fight.

Hyper scale data centers bring few benefits to communities," claims Peaceful Peculiar, a grassroots campaign that successfully repelled a Diode Ventures data center in Peculiar, Missouri. "Dozens of communities around the nation severely struggle from the presence of data centers. It's only about maximizing their profits, so they choose properties as close as possible to a large power supply."

The group's Facebook page, "Don't Dump Data on Peculiar," once welcomed input from activists and organizers in other communities, according to a profile by the Washington Post. Now after successfully fighting off the Diode facility last October, the organizers of Peculiar have become a resource to fellow activists as far away as Indiana, Idaho, Georgia, and Texas.

… more …

[tinrobot]

6. The 2000 dot-com bubble didn't kill the internet.

This bubble won't kill AI.

Hopefully, it will temper some people's expectations about the technology and make some billionaires slightly less rich.

[Bernardo de La Paz]

16. True, but I find it interesting AI cap-ex is much bigger than dot-com cap-ex in 2000

I heard this on CNBC a couple of days ago and I don't have the details as to what from 2000 was being compared (2025 cap-ex vs some kind of 2000 internet investments), but the comparison was on inflation adjusted dollars. It put 2000 completely in the shade, maybe by an order of magnitude.

[OC375]

7. Probability

A.I., isn't... It's just a probability machine. It gives you the response most likely to be correct based on whatever data sets it sucks up. It bears no responsibility for what it produces. It's GIGO. Tomorrow with a different set of data may yield a totallt different answer, not becuase it learned, but becuase it simply has volumetrically more input added to make a response from. To say it overpromises, is an understatement. It does seem to be making us dumber, lazier and less skeptical in general though, so there's that. I'd call A.I. just automation at this point.

[Mossfern]

12. LUDDITES UNITE !!!

[Bernardo de La Paz]

17. Like humans

If you immerse a child in a toxic red hatter environment you'll get a different adult than one immersed in a progressive environment.

[hunter]

8. With any luck AI will crash so badly that investors will be scared off for decades.

Hopefully it's not the sort of crash that kills multitudes, but it could be.

[Bernardo de La Paz]

19. Like the dot-com crash? Didn't scare off investors for even one decade. Some kind of crash, but not decades deep.

If the bubble bursts or deflates in 2026, investment will be coming back by 2030, more or less.

For example, Netflix launched about 1997 with DVD by mail, but launched online streaming in 2007. Expect similar timing but somewhat faster because the AI bubble has a better foundation at this point. However, getting to the next level (akin to 2007) is going to be a bit more difficult than the optimistic bulls are projecting.

[hunter]

27. I can imagine AI severely damaging a regional electric grid during a deadly heatwave.

This could be caused by AI electricity demands, grid control software that was produced with AI assistance, incompetent employees who were hired with AI assistance, or a combination of all these things.

I can also imagine AI crashing large rockets or weapons into heavily populated areas, or blowing out water and natural gas distribution systems.

AI is also capable of triggering psychopathic behavior in politically powerful individuals who control police and military forces.

AI is already killing people piecemeal -- in car accidents, by encouraging people to commit suicide, or promoting violence, etc. It simply hasn't hit the big time yet.

The dotcom bust wasn't a similar situation.

[Bernardo de La Paz]

29. I can imagine things too. . . . nt

[hunter]

30. I don't know about you, but my imagination, and even to a certain extent my paranoia...

... has kept me out of a lot of trouble.

[Bernardo de La Paz]

32. My imagination has made me money, and kept me out of much trouble too. . . . . nt

[hunter]

33. AI as it currently exists will end up like some kind of chronic disease.

People have made lots of money on chronic diseases.

[Jack Valentino]

9. A.I. uses too much electricity......

[Melon]

10. Name a major technological breakthrough that the world back away from due to...

Environmental concerns or anything else. I think the genie is out. China doesn’t care. They are starting countless new coal burning power plants a year. The US can’t back away without getting behind. There is no reverse.

You can see the jobs this will take quickly. I’d never hire someone today to write a resume. Chat GPT can do it in a minute. Business headshots from $350 photographers…nah. Business analysts…already getting covered by AI. I don’t know how we could go backward. The companies are just learning how to monetize all of this.

[CentralMass]

26. China consumes a lot of power. They also lead the world in reneawble energy investment.

"China led global renewable energy investment in 2024, with over $290 billion, nearly three times the investment from the United States."

"By the end of 2024, China's total installed renewable energy capacity reached 1,889 gigawatts (GW), which accounted for a record high of 56% of the nation's total installed power capacity. This includes 887 GW of solar power, 521 GW of wind power, and 436 GW of hydropower."

[Melon]

28. China also has 1200 coal fired plants vs US 195

95% of coal fired plants are in China and they are adding more every year. China produces 4x as much carbon dioxide green house gas emissions as the next closest country(8x more than the USA). If you’ve been to China you see the destruction of their environment. They increase green tech but it has only reduced their production of Co2 by 5%.

China added the most coal fired power plants ever….in 2024.

[hunter]

31. More importantly they are building nuclear power plants...

... that will eventually make most other energy resources obsolete.

[Quixote1818]

11. It's not nearly as big of a bubble as the Dot-com bubble was because AI companies are making a crap load of sales

The Dot-com bubble was very real. Most of the companies at that time wern't making much money.

Search results:

Differences from the Dot-Com Bubble

Real revenue growth already exists – NVIDIA, Microsoft, and others are making tens of billions directly from AI, unlike 1999 when most internet companies had little or no revenue.

Wider adoption – AI is being used in healthcare, finance, logistics, software, chip design, and consumer apps. The internet in 1999 was still dial-up and mostly just email + websites.

Infrastructure is useful even if hype fades – The fiber optic cables from the dot-com era ended up enabling the modern internet. Similarly, today’s GPU clusters, data centers, and AI models won’t just vanish — they’ll underpin future productivity.

Corporate demand is real – Unlike 1999’s consumer-driven internet craze, AI is being heavily adopted by Fortune 500 companies to cut costs and boost productivity. That gives it a stronger base.

Is AI in a bubble?

Short term: Yes, there are bubble-like pockets — some small AI startups with extreme valuations will likely crash, and GPU demand could overshoot before stabilizing.

Medium term (2–5 years): We could see a “shakeout” like the early 2000s, where only the strongest players (NVIDIA, Microsoft, Google, etc.) keep thriving while many smaller firms collapse.

Long term (10+ years): AI is almost certainly here to stay and may be as transformative as electricity or the internet. The companies that survive could dominate global markets for decades.

Bottom line:
AI today looks part bubble, part real revolution. The market may experience a dot-com-style correction, but unlike 1999, the technology is already delivering real profits and utility. That means the long-term winners are likely to be even bigger than the internet winners (Amazon, Google, etc.), but investors should expect volatility and shakeouts along the way.

[Bernardo de La Paz]

20. Excellent analysis. Thank you. . . nt

[Xolodno]

13. AI is a useful tool, that's it.

What it does is simply takes something that used to be a bit time consuming and makes it easier and faster. We all had our various "templates" to get work done faster. But the thing is, that's when the work began. AI doesn't find insights, it doesn't see patterns in a market that intuition would hone in on, it doesn't recognize things that could be a problem later, etc.

CEO's are being sold on AI because a vendor is telling them how much money they can save by eliminating employee's. And given I've worked with a few C Suite people, they aren't the brightest sad to say. All they are looking at is thier bonus and not looking how the company is shrinking and about to have an economic catastrophe.

An old manager of mine gave me a call, he got the boot and not being replaced. He, like I did, worked in the guts of the company, finding out where we could increase sales, where to expand, what training was needed. Now, completely gone because the CEO thinks AI can do that for them. It won't.

[Bernardo de La Paz]

21. It is missing the boat to dismiss it as a tool. The internet was/is a tool, but it was * transformative *.

AI will be even more transformative because it is wider reaching. It does not require the internet; autonomous robots are autonomous.

Today's AI is very primitive compared to what is coming, just like internet apps in 2000 were primitive compared to 2020 apps.

[Ron Green]

22. It hurts people and the planet.

There really is no meaningful upside.

It’s a doom machine.

[hatrack]

23. Precisely - and how will it be powered? By coal and gas, because they're cheap, and renewables are gay, woke and trans.

Other items to keep in mind as consequences of this bubble and associated policy/investment decisions:

1. Any chance of reducing American GHG output in any meaningful way (not that there was much of a chance of it anyway) is now gone. Globally, we hit 430 ppm of atmospheric CO2 in May of 2025 - the only month so far on the instrumental record at or above that total. Next year we'll see five or potentially six months at or above 430, and from 2027 on, we'll never see a month below that figure.

2. We can expect substantially higher electrical rates across the country, with some exceptions depending on state laws governing electrical utilities. For investor-owned utilities, the costs of additional infrastructure, generation and transmission will be passed along to consumers. Retail consumers, who pay substantially higher rates than industrial users (i.e. AI and crypto farms) will pay far more proportionally.

3. More and bigger grid failures are also likely. Rapidly spiking demand and lax regulation are part of the problem, obviously, especially in places like Texas. Other factors are also in play. One is increased vulnerability of the electrical grid to increasingly violent weather - derechos, ice storms and stuck jet streams in winter, for example. Another issue is manufacturing. If you're a utility and in the market for transformers, place your order now - depending on size and design, delivery dates for transformers in the US range from one to three years.

4. Enjoy water shortages and high water bills? You're going to love what data centers do to counties and states in arid parts of the country. Maricopa County (Phoenix) is second only to northern Virginia in terms of data center projects. Arizona has lost 25 million acre-feet of groundwater in 20 years, while farmers are still using flood irrigation. Upstream, Powell is at less than 29% of capacity, Mead just above 31% of capacity, and the long-term drought that's hammered the southwest for 20 years shows no sign of changing (it appears to be driven by large-scale slow-motion alterations in the Northern Pacific currents and climate).

"Taking your eye off the ball" doesn't even come close to describing this disastrous misallocation of money and talent. At least Jack got some magic beans, which is more than we'll receive.

[bucolic_frolic]

34. Hey thanks for the credit

I ask questions because I don't know how to answer them!