News: 0180002366

  ARM Give a man a fire and he's warm for a day, but set fire to him and he's warm for the rest of his life (Terry Pratchett, Jingo)

'Stratospheric' AI Spending By Four Wealthy Companies Reaches $360B Just For Data Centers (msn.com)

(Saturday November 08, 2025 @11:34AM (EditorDavid) from the big-big-data dept.)


"Maybe you've [1]heard that artificial intelligence is [2]a bubble poised to [3]burst ," [4]writes a Washington Post technology columnist . "Maybe you have [5]heard that it [6]isn't . ( [7]No one really knows either way, but that won't stop the bros from jabbering about it constantly.)"

"But I can confidently tell you that the money being thrown around for AI is so huge that numbers have lost all meaning."

> The companies pouring money in are so rich and so power-hungry (in multiple meanings of that term) that our puny human brains cannot really comprehend. So let's try to give some meaning and context to the stratospheric numbers in AI. Is it a bubble? Eh, who knows. But it is completely bonkers. In just the past year, the four richest companies developing AI — Microsoft, Google, Amazon and Meta — have spent roughly $360 billion [8]combined for big-ticket projects , which included building [9]AI data centers and stuffing them with computer chips and equipment, according to my analysis of financial disclosures.... How do companies pay for the enormous sums they are lavishing on AI? Mostly, these companies make so much money that they can afford to go bananas...

>

> Eight of the world's top 10 most valuable companies are AI-centric or AI-ish American corporate giants — Nvidia, Apple, Microsoft, Google, Amazon, Broadcom, Meta and Tesla. That's according to tallies from S&P Global Market Intelligence based on the total price of the companies' stock held by investors. My analysis of the S&P data shows that the collective worth of those eight giants, $23 trillion, is more than the value of the next 96 most valuable U.S. companies put together, which includes many still very rich names such as JPMorgan, Walmart, Visa and ExxonMobil. No. 1 on that list, the AI computer chip seller Nvidia, last week become the [10]first company in history to reach a stock market value of $5 trillion. That alone was more than the value of entire stock markets in most countries, Bloomberg News [11]reported , other than the five biggest (in the U.S., China, Japan, Hong Kong and India)...

>

> All the announced or under-construction data centers for powering AI would consume roughly as much electricity as 44 million households in the United States if they run full tilt, according to a recent analysis by the Barclays investment bank as [12]reported by the Financial Times. For context, that's nearly one-third of the total number of residential housing units in the entire country, according to U.S. Census Bureau [13]housing estimates for 2024.



[1] https://www.cnbc.com/2025/08/18/openai-sam-altman-warns-ai-market-is-in-a-bubble.html

[2] https://www.washingtonpost.com/technology/2025/09/30/ai-economy-investment-bubble/

[3] https://www.wired.com/story/ai-journalism-worst-thing-about-ai/

[4] https://www.msn.com/en-us/technology/artificial-intelligence/the-ai-spending-frenzy-is-so-huge-that-it-makes-no-sense/ar-AA1Q132J

[5] https://www.cnbc.com/2025/10/29/powell-says-ai-different-from-dotcom-bubble-major-source-gdp-growth.html

[6] https://www.derekthompson.org/p/why-ai-is-not-a-bubble

[7] https://www.washingtonpost.com/business/2025/10/03/ai-will-trigger-financial-calamity-itll-also-remake-world/

[8] https://www.washingtonpost.com/technology/2025/10/30/google-meta-ai-data-center-spending/

[9] https://www.washingtonpost.com/technology/2025/10/21/nvidia-ai-factories/

[10] https://www.washingtonpost.com/technology/2025/10/29/jensen-huang-nvidia-trump-china-trade/

[11] https://www.bloomberg.com/news/articles/2025-11-02/nvidia-is-worth-5-trillion-here-s-what-it-means-for-the-market

[12] https://www.ft.com/content/2b849dbd-1bef-4c26-aa11-2cb86750d41e?shareType=nongift

[13] https://www.census.gov/data/tables/time-series/demo/popest/2020s-total-housing-units.html



Re: No biggie (Score:2)

by blue trane ( 110704 )

When the Fed prints money to buy sinking assets, does any taxpayer get debited? In 2020 when the Fed increased base, high-powered money by 40%, and inflation only briefly touched 9%, does that show printing money is a net gain?

Imaginary assets like hallucinations? (Score:2)

by shanen ( 462549 )

Can't figure out your focus, but it may have something to do with virtual money (such as recursive futures). The vacuous Subject you inherited and propagated didn't help at all. Care to clarify your points? Or at least your main point? Maybe something about fiat currencies, old or new?

Re: (Score:2)

by evil_aaronm ( 671521 )

I wonder if these rich sorts will use AI to figure out how to solve society's problems so that they - the rich sorts - don't have to use AI to figure out the best approach to defend themselves from angry, desperate hordes intent on righting the balance of wealth. Sort of a foresight - slash - "an ounce of prevention" thing.

Re: (Score:3)

by Ostracus ( 1354233 )

Democratizing AI would be the best thing, because right now it's a rich persons game with the rewards in their hands. If open source can free the world then something similar for AI would be good.

Re: (Score:2)

by kertaamo ( 16100 )

Why do you say "Russia's attempts at sieging cities is only going to get better"?

That may well be true but at the same time our attempts to defend ourselves will get better too. We have access to AI as well you know.

Re: Because AI solves everything? /s (Score:2)

by toutankh ( 1544253 )

> For example, the traveling salesman problem. Add another city, the cost of an answer goes up exponentially. In the 2000s, you could solve it with genetic algorithms, and get a good answer. It isn't the Answer (tm) that is perfect... but it is good enough. Now, LLMs can take into accounts more variables, and get closer.

I have been conducting research on combinatorial optimisation for more than two decades. What you say is not true.

First, genetic algorithms are not a good approach for solving the TSP. If you want an exact solution then the best methods are based on branch and cut, for example [1]Concorde [uwaterloo.ca]. It can solve surprisingly large instances. If you want something close enough, this particular implementation of the Lin-Kernighan heuristic works surprisingly well and way faster than any genetic algorithm.

Second, and way m

[1] https://www.math.uwaterloo.ca/tsp/concorde.html

Meanwhile... (Score:3)

by rsilvergun ( 571051 )

You're going to do a 60 minute commute one way and you're just now noticing the bridge you're stuck under is crumbling because we haven't built infrastructure in 30 years.

It's almost as if giving all of our money to the rich because we were busy being distracted by moral panics was a bad idea.

The funny thing is when I say moral panic everybody agrees with me but if I mention the specific moral panics that caused you to give all your money to the rich then everybody gets upset.

Re: Meanwhile... (Score:2)

by blue trane ( 110704 )

Do the bottom 90% even have enough money that the rich want it, or can they make ten times tbat amount just by investing in the fictitious goods of stock markets, while we get a basic income to satisfy our basic physical needs?

It won't matter once the power goes out (Score:2)

by silvergig ( 7651900 )

I don't think you'll be able to add an additional 33% of the USA's usage of power over a couple years and think that it won't end badly. And renewable power like solar would require massive grid storage since this stuff is running 24x7.

This will pop simply because the resources just aren't there, without even taking into account all of LLM's bullshit hallucinations and shit data output.

Re: (Score:2)

by Ostracus ( 1354233 )

That Ion QC story earlier shows promise in helping with that problem.

Re: (Score:2)

by silvergig ( 7651900 )

> That Ion QC story earlier shows promise in helping with that problem.

Data Centers being built right now are not going to benefit from a tech that could still be 20+ years away from mass production. Even if those new computers were available right now, they won't be just a snap-in for these data centers, which are being built specifically around GPU architectures and GPU cooling. Data centers being constructed and used right would probably need to be all but razed to accommodate an entirely new paradigm for computing.

Re: It won't matter once the power goes out (Score:2)

by blue trane ( 110704 )

If you look at a Sankey diagram of US energy production and consumption, will you too see that residential electricity usage is less than 15% of total electricity demand? So does data center usage only represent a 5% increase in demand, which given the 10% electricity generation surplus shown on eia.gov's energy facts explained page, is totally doable given current production?

[1]https://flowcharts.llnl.gov/si... [llnl.gov]

[2]https://www.eia.gov/energyexpl... [eia.gov]

So is this panic over electricity supply wildly overblown, a pure h

[1] https://flowcharts.llnl.gov/sites/flowcharts/files/2024-12/energy-2023-united-states.png

[2] https://www.eia.gov/energyexplained/us-energy-facts/

It's all based on the assumption that... (Score:2)

by MpVpRb ( 1423381 )

...progress can only be made by using massive and ever increasing computing power

One efficient algorithm could change everything

`Lasu' Releases SAG 0.3 -- Freeware Book Takes Paves For New World Order
by staff writers

...
The central Superhighway site called ``sunsite.unc.edu''
collapsed in the morning before the release. News about the release had
been leaked by a German hacker group, Harmonious Hardware Hackers, who
had cracked into the author's computer earlier in the week. They had
got the release date wrong by one day, and caused dozens of eager fans
to connect to the sunsite computer at the wrong time. ``No computer can
handle that kind of stress,'' explained the mourning sunsite manager,
Erik Troan. ``The spinning disks made the whole computer jump, and
finally it crashed through the floor to the basement.'' Luckily,
repairs were swift and the computer was working again the same evening.
``Thank God we were able to buy enough needles and thread and patch it
together without major problems.'' The site has also installed a new
throttle on the network pipe, allowing at most four clients at the same
time, thus making a new crash less likely. ``The book is now in our
Incoming folder'', says Troan, ``and you're all welcome to come and get it.''
-- Lars Wirzenius <wirzeniu@cs.helsinki.fi>
[comp.os.linux.announce]