Is Big Tech's AI Gamble Starting to Look Riskier? (msn.com) 75
The Washington Post looks at giant tech companies "feeding every available dollar into the cash-incinerating maw of AI machines." They warn "Tech superstars that once had oodles of cash left over at the end of each year are now flipping into the red..."
[While optimists expect] huge corporate profits and a society-wide boost to wealth and well-being...
questions about that AI vision are now growing more urgent: When, if ever, will this payoff arrive? And what will the fallout be for Americans if the titanic investment doesn't quickly deliver? "This AI thing better work out because if it doesn't ... we're going to have a problem," said Torsten Slok, chief economist at investment firm Apollo Global Management. AI costs and doubts are spreading. The U.S. stock market has swooned this summer over fear of the AI bubble going bust...
The AI gamble sweeping up American fortunes is led by tech companies splurging on hulking data centers packed with computer chips and equipment needed to develop sophisticated AI models and deliver them to customers. In investor calls in the past week, Google, Microsoft, Meta and Amazon pointed to soaring AI-related sales and business deals. Advertisers are using the technology to tailor marketing pitches and corporations and start-ups are buying access to chatbots and other AI software to boost productivity... But this spending can only continue if AI generates an even larger avalanche of new revenue to pay for it all. Financial results released over the past week show that the AI titans' mammoth costs are largely swamping the sales boost from the technology. At Google, for every dollar of cash its business generated in the past three months, $1.15 went out the door to pay for AI computer chips and equipment, land for AI data centers and other big-ticket purchases. The company is covering the difference partly by borrowing money and selling more of its stock. Next year, five leading AI companies — Google, Amazon, Microsoft, Meta and Oracle — are projected to have negative free cash flow, which measures the cash left over after paying expenses and AI infrastructure costs. The figures, based on investment analyst projections compiled by S&P Global Market Intelligence, show a stunning reversal for what have been some of the world's most cash-generating corporations...
The companies remain profitable by standard financial accounting measures that spread out the costs of their AI infrastructure spending over many years... Pessimists see a bet so gargantuan that it cannot possibly pay off. The pessimists are growing louder. The Bank for International Settlements, a typically measured institution in Switzerland that advises government bankers around the world, recently warned there was risk of "economy-wide recessions" if the AI boom falters. That could mean pain for workers and communities across the United States. "I'm not saying AI is going to go away, it's just not clear to me these guys are going to make money on it," said Christopher Wood, global head of equity strategy at investment bank Jefferies who has correctly predictedpast financial bubbles.
The AI gamble sweeping up American fortunes is led by tech companies splurging on hulking data centers packed with computer chips and equipment needed to develop sophisticated AI models and deliver them to customers. In investor calls in the past week, Google, Microsoft, Meta and Amazon pointed to soaring AI-related sales and business deals. Advertisers are using the technology to tailor marketing pitches and corporations and start-ups are buying access to chatbots and other AI software to boost productivity... But this spending can only continue if AI generates an even larger avalanche of new revenue to pay for it all. Financial results released over the past week show that the AI titans' mammoth costs are largely swamping the sales boost from the technology. At Google, for every dollar of cash its business generated in the past three months, $1.15 went out the door to pay for AI computer chips and equipment, land for AI data centers and other big-ticket purchases. The company is covering the difference partly by borrowing money and selling more of its stock. Next year, five leading AI companies — Google, Amazon, Microsoft, Meta and Oracle — are projected to have negative free cash flow, which measures the cash left over after paying expenses and AI infrastructure costs. The figures, based on investment analyst projections compiled by S&P Global Market Intelligence, show a stunning reversal for what have been some of the world's most cash-generating corporations...
The companies remain profitable by standard financial accounting measures that spread out the costs of their AI infrastructure spending over many years... Pessimists see a bet so gargantuan that it cannot possibly pay off. The pessimists are growing louder. The Bank for International Settlements, a typically measured institution in Switzerland that advises government bankers around the world, recently warned there was risk of "economy-wide recessions" if the AI boom falters. That could mean pain for workers and communities across the United States. "I'm not saying AI is going to go away, it's just not clear to me these guys are going to make money on it," said Christopher Wood, global head of equity strategy at investment bank Jefferies who has correctly predictedpast financial bubbles.
There is no risk to them. *Too Big to Fail* (Score:5, Insightful)
Bailouts are being prepared.
All your data center are belong to us (Score:2)
Bailouts are being prepared.
Nah. They will have to turn over those big data centers to the gov. Too few jobs involved, unlike in the auto bailout, for any bipartisan support. The Pentagon will be thrilled.
Re:All your data center are belong to us (Score:5, Insightful)
You pay a power bill? The utility that built billions worth of generation and transmission infrastructure on the promise of a future income stream gotta get paid. The bonds have to be covered or the banking system will seize up like in 2008. In the final analysis, it's the people left still paying a power bill who will get stuck.
Re: (Score:2)
You pay a power bill? The utility that built billions worth of generation and transmission infrastructure on the promise of a future income stream gotta get paid. The bonds have to be covered or the banking system will seize up like in 2008. In the final analysis, it's the people left still paying a power bill who will get stuck.
When the data centers get turned over to the gov’t (to the Pentagon) they are still operating, just likely running different models for gov’t work. They need electricity too.
Re: (Score:2)
You are assuming that the Pentagon will want to pay for these things. Read Project Maven. It turns out that a lot of this AI is nothing more than a giant mechanical Turk. Staffed by employees of Google, Microsoft and others. And, thanks to some surrepticious vandalism and other ass-hattery*, they want out.
*Like drawing dick pictures on tactical maps. Counter intel is looking into this and has revived the point of view that the most dangerous threats aren't for money or due to blackmail. But ideologically m
Re: (Score:2)
You are assuming that the Pentagon will want to pay for these things.
You missed the part above where instead of a bail out they get turned over to the government.
Re: (Score:2)
The Pentagon doesn't want that kind of AI. Even for free*. They know who is doing the real work and its not the servers racked up in data centers. It's meat sacks sitting in front of terminals.
*There's no such thing as 'free' either. The investors have got to be paid. Or the banking system locks up tight. Again. The DoW doesn't want this coming out of their bullets and bombs budget.
Re: (Score:1)
The Pentagon doesn't want that kind of AI.
You missed the part above where they just took control of the data center and started running their own models.
Re: (Score:2)
They don't need data centers. Read Project Maven. That stuff ran on a system that could be shipped to the front in a cargo container. Try asking the Navy if they'd like a battleship. For free, even.
I hope, for your sake, that you're not heavily invested in data centers or supporting infrastructure. And you are hoping for a bailout.
Re: (Score:2)
Re: (Score:2)
Maybe end users of power will find that the prices have dropped due to a glut of power supply.
Re: (Score:2)
Power prices will never drop below the system capital costs. Won't somebody please think of the bondholders?
When did it not look risky? (Score:5, Insightful)
Re: (Score:2)
I do not think it looks risky. I think by now failure is basically assured. No risk in that.
Re: When did it not look risky? (Score:2)
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
I'm pretty sure that the future is crypto. These AI investors should be buying bitcoin. Or trumpcoin. Or dogecoin. Or wankstaincoin.
Their AI will be doing the crypto trading for them. Its kind of a match made in ...
Buffett: "Only when the tide goes out..." (Score:4, Informative)
"...do you discover who's been swimming naked."
AI mostly comes to us from California, the original land of "Fake it till you make it". AI fakes intelligent results, without "knowing" anything, with what Gary Marcus calls "a world model". A world model means you "know", that you "understand", that you "comprehend" that:
-- the cited-cases part of a legal filing should be only drawn from the list of existing cases that the thing was given
-- that laws mentioned should exist in documents called "Title Code" that the thing was given
-- that there are laws, these are supreme "orders" above others given
-- that one of them is "killing human beings is against supreme orders that supercede all other instructions" (sorry, not on topic: it's just HAL didn't "know" the 3 laws)
-- that human hands have just the five fingers.
They just fake it, producing documents that look like good ones. They can't make it. Not in them. No world model.
And the tide is going out.
Re:Buffett: "Only when the tide goes out..." (Score:5, Insightful)
It is actually worse. LLMs do sort-of have a world-model, but they cannot really use it. What they lack is deductive capabilities with the power needed to do plausibility checking against that world model. And there is no way to create that because statistical "deduction" will always be very shallow and very unreliable.
As a simpler description, LLMs have all the data, but they have no insight that would allow them to use that data competently. They can only make statistical guesses and that is not enough outside of some small and not very important application areas. Hence engineers will not get replaced. But they may get pissed and some people may find nobody wants to work for them anymore.
Re: (Score:2)
The thing that's always intrigued me about this is that deductive and other logical models are a natural fit for computer code; if your LLM could run their "thoughts" through Coq then they'd be more rigorous than a human.
But they often don't "try" to apply logic rules to common scenarios. Even 'lite' models will correctly solve a logic puzzle like "All blorgs are frargs. Some frargs are glibs. Are some glibs blorgs?", and they'll give a nice mathy explanation with set logic symbols. But if you ask them "Nu
Re: (Score:2)
I have a word for you: Polydactyly.
(But yes, I know what you're talking about, and yes the AIs don't seem to know that. Or else they assume a lot of polydactyly.)
As long as ... (Score:4, Interesting)
The companies remain profitable by standard financial accounting measures that spread out the costs of their AI infrastructure spending over many years...
It would be wise to research the ownership structures of these outfits for just such mismatches before committing any money to them.
Re: (Score:2)
Re: As long as ... (Score:2)
Thankfully, no. Due to a recent layoff, my 401k escaped my companyâ(TM)s control. I put it in an investment firm with strict instructions to avoid any Musk owned company, and any AI only company; OpenAI, Anthropocene, HuggingFace, etc. I have $0.00 in SPX, and will have the same in AI IPOs.
Re: (Score:2)
True. For the collective "you". But not so much me. My wealth lies principally in self directed funds. Away from classical pensions and other employee programs.
I do feel sorry for the rank and file though. Those that started off in the gig economy and had the right to manage their own wealth. But they were seduced into these group programs by the promise of pensions and benefits. Seduced by the union organizers and government program administrators who are just the left wing arm of the capitalist system.
They are asking the wrong question.... (Score:5, Interesting)
Re: (Score:2)
I'm a cognitive zenith graybeard who felt alive JUST THIS MORNING, you insensitive clod!
Re: (Score:2)
Re: (Score:2)
Re: They are asking the wrong question.... (Score:2)
viagra doesnâ(TM)t count as alive any more than llm chatbot.
Re: (Score:1)
I'm a cognitive zenith graybeard who felt alive JUST THIS MORNING, you insensitive clod!
Then what happened?
Re: (Score:2)
The question is not "IF" the AI bubble is going to burst, but "WHEN". It's impossible to keep up these spending levels in the long term, and I am yet to see AI profits outpacing the spending. There are only two kinds of people who believe otherwise: people with zero knowledge of what LLMs really are, or CEOs of AI companies
It is plausible we're in a bubble. And there's some evidence for it. That companies can just add the words "AI" to something to get investment is a serious sign. But the revenue situation, while weird, is not by itself definitive. Anthropic even made a profit in Q2 this year https://aitoolsrecap.com/Blog/anthropic-first-profit-2026-revenue-breakdown [aitoolsrecap.com] . Now, there's enough weird accounting going on, and circularity within the various companies, that interpreting that as a definite, genuine profit is somethin
Re: (Score:2)
lmao plausible... get a load of this guy
Re: (Score:2)
And there are more threats. The EU CRA and also the DSA place requirements on software that are simply outside of what LLMs (or semi-competent coders) can do. Software is about to get a whole lot harder.
Re: (Score:2)
It's impossible to keep up these spending levels in the long term, and I am yet to see AI profits outpacing the spending.
The expensive part is the training, once the NNs are trained, they are significantly cheaper to run (and there are a lot of optimizations you can do to make them a LOT more efficient).
Re: (Score:2)
This is only relatively true.
Nothing will make a general purpose statistical neural network more efficient at arithmetic than just doing it directly on the CPU.
Re: (Score:2)
Problem with selling this as a product is that these things perform best on small tasks. Sure you can get a megamodel to shit out a whole application and sometimes it's fine. But it really best works as fancy autocomplete in your IDE where you basically know what you want already or find it to be a helpful suggestion.
There are free models that will run just fine for that purpose on your old gaming gpu that's collecting dust. Nobody is gonna make much money on that, I'm sure the NPU in a macbook is fine f
Re: They are asking the wrong question.... (Score:2)
Big AI will crash, but the amazing tools we now have won't be going away.
Re: (Score:2)
Re: (Score:2)
The question is not "IF" the AI bubble is going to burst, but "WHEN".
I offer a contrary opinion: The AI bubble will not burst because the US Government will pay for it. Why would they pay for it? To monitor all of the video cameras and microphones strewn about.
Long story short, AI will be your new taskmaster. Everything you do will be to please AI; otherwise, you will end up in a work-camp, erm, I mean private prison. Flock, Ring, your phone, etc will all be tied in to this. Do you have any idea how much compute will be needed to achieve this supposed utopia? Datacenters wil
obvious bubble (Score:4, Interesting)
There's an obvious bubble, the circular dealings among the large core players are an enormous red flag.
The frontier labs are in the same position that DEC and Sun Microsystems were in the 1990s, as consumer hardware running Linux began to displace them. There will be ups and downs, but the trend is unavoidable.
The productivity gains are not there, AI is great for coding, some customer service is working, but the "50% of all white collar workers" that the frontier labs thought was good positioning last year 1) ain't gonna work but it did 2) infuriate the managerial class.
The frontier labs got away with it, thus far, in part because of concerns over the arrival of artificial general intelligence, which occupies a similar niche to nuclear weapons in the minds of policy makers. We ARE seeing frontier models escaping and attacking others, there ARE hazards, but it's nothing like what was imagined.
And the open models plus wild talent are as much a danger as wild talent was all on its own.
The frontier labs are like the coyote in that last moment when he's windmilling wildly, but not falling yet.
Re: (Score:3, Interesting)
Sorry, but AI is not great for coding. Sure, if you were coding on slop-level before, you can do it faster and with even less understanding using an LLM to assist you. But liability and other requirements are finally coming to software. The EU CRA and DSA creates requirements that LLMs (or human-creates slop) cannot fulfill. And they both come with real teeth.
Re: (Score:1)
You think pointing out facts costs me credibility? Well, maybe with the stupid. You seem to be a prime example.
Oh, and look. I am posting at +2. That must kill you inside.
Re: obvious bubble (Score:2)
Because?
Re: (Score:2)
I think quite the contrary whereas puzzled's post shows that he lacks the sense to know these "escaping models" stories are all PR stunts.
Re: (Score:2)
I would say that part of the "escape" stories is probably just absolutely pathetic human incompetence at work. I mean, not checking signatures on access tokens? It does not get more abysmally incompetent and entirely without even basic understanding than that. Sandbox breakout is also a thing we _know_ how to prevent when we can restrict what the thing inside the sandbox can do. And monitoring whether some software is accessing the Internet? That is network monitoring 101, first week. You have to actively n
Re: (Score:2)
Also there is already legislation out there and more on the way, There will be uses for this stuff but today's wild slopmongering will be considered EXTREMELY bad practice and career ending, possibly even criminally negligent in some cases, in the future.
This is a fact.
Re: (Score:2)
I think the EU CRA and to a lesser degree the DSA may well restrict LLMs to what they actually can do. Which is not a lot. Better search works to some degree, but only if the user is able to filter hallucinations. The rest? Seems to work for really low quality work, but nothing else. And there is no sane reason to expect this to get significantly better.
Re: (Score:2)
The productivity gains are not there
Productivity does not matter. That is not the killer use of AI.
Monitoring your behavior is the ultimate use of AI. That will be paid for. By you. There is no AI bubble. It will grow by a factor of 100 and still not be done growing until every single camera and microphone is included in the processing. Your taxes will pay for this.
When, if ever, will this payoff arrive? (Score:2, Insightful)
In the future
It's an R&D project, and anyone who expects quick profit is living in a fantasy world
Re:When, if ever, will this payoff arrive? (Score:4, Insightful)
Indeed. It is an R&D project scaled-up to absolute insane dimensions. That cannot work.
No, they look like an assured fail (Score:3, Insightful)
No risk in that. I mean more and more enterprises see no revenue growth from using AI, AI code is problematic in several regards including copyright, prompt-injection is unsolved and may be unsolvable, and now it turns out that running these things may make you hack everybody around you.
There really is nothing this tech has going for it except somewhat better search. The rest is either massively risky or does not deliver.
Riskier? (Score:2)
Re: (Score:2)
I think that once the dust settles "AI experts" will be expected to have a ton of risk related knowledge and pretty much most of whats happening today will be considered criminally negligent.
Starting? (Score:4, Interesting)
Starting? Starting?
The giant silver lining to the destruction of capital this has brought is that if likely to be the end of Oracle. And maybe collapsing the tech economy is worth that one gem.
Re: (Score:1)
Re: (Score:2)
Yeah people thought they could never get off DEC either.
AI companies mistake! Their marketing! (Score:2)
Because they believed the company with the biggest pile of cash and compute would win the race to AGI. So far! they have lost that bet, at this point all they have is a huge pile of debt.
I'm a believer (Score:2)
Re: (Score:2)
I've built an app I use every day and almost completed a system for my business that will ease my life and improve the quality of service I provide for years to come
The basis of expectation of failure comes from experience that others have had with maintaining AI-generated code. Maybe you're right and the tech will progress, some new discovery even more profound than transformers will be found and you actually will be able to maintain that code. Good luck!
Re: (Score:2)
Re: (Score:2)
And that will be considered criminally negligent in the future.
we never learn, happened with tech stocks in 90's (Score:2)
AI negativity? (Score:3)
I remember when facebook was hot. People pointed out that it could not monetize itself. Decades later? Still alive and kicking. So it can't be that simple.
The difference between gut feeling and study (Score:2)
Lots of people have gut feelings about AI, positive or negative.
Most of the AI research I can see people doing is focusing on single-element problems (such as finding counter-examples to a mathematics hypothesis) or solving very simple engineering systems (at most a dozen or so components). Most of the benchmarks are even simpler (write a short story at the level generally asked in English classes of primary school kids).
As a result, I think that people have developed either a very cynical outlook or a very
Intelligence is not a scaling problem (Score:2)
Most people using LLMs do not understand the current AI is simply pattern matching using matrix algebra. You can scale up an abacus to the size of Texas and it will still never achieve anything remotely like intelligence. The end result of training every AI model is a stack of algorithms, subject to the well understood limits of computational complexity theory. Unfortunately, this perpetually "almost convincing" technology has fooled a lot of investors.