That's the amount of economic value AI will have to deliver to justify all the investment:
The AI industry must generate $6 trillion in annual revenue by 2031 to keep funding the infrastructure needed to meet anticipated demand, and this will require imaginative new uses beyond simply enhancing employee productivity.
That $6 trillion figure comes from the 2026 annual Global Technology Report from management consultants Bain & Company, which put its thinking cap on to figure out how the AI sector can possibly meet this challenge.
It says the massive demand for AI compute infrastructure has revived the hardware industry. Some of the fastest growing segments include high-bandwidth memory (HBM) used in GPUs, advanced packaging and custom silicon, and application-specific integrated circuits (ASICs) all scaling rapidly.
As The Register reported last year, AI spending is actually keeping the US economy out of recession, with datacenter infrastructure and model development providing the only significant growth areas.
Bain points to the "arms race" among hyperscalers for AI capacity, claiming that capital expenditure by Microsoft, Google, Amazon, Meta, and Oracle could hit $780 billion for the whole of 2026, nearly five times the level seen just three years earlier.
By 2031, it estimates that annual spending on AI infrastructure could reach $1.5 trillion, a figure that is not far off the $1.6 trillion forecast separately by analyst firm Omdia. The $6 trillion comes from the Bain's assumption that capital expenditure will amount to about 25 percent of industry revenue, an ambitious but reasonable percentage based on trends among cloud providers, it says.
Bain expects that existing applications of AI will grow, but estimates these will add up to a total between $1.2 trillion and $1.8 trillion in revenue. This includes consumer AI, via subscriptions and advertisements, plus enterprise AI, through software development, sales, marketing, customer service, and IT operations.
That leaves a remaining $4.2 trillion of additional revenue for the industry to find from somewhere.
[...] "The debate today is fixated on employee productivity. The economics of AI infrastructure demand trillions in new revenue beyond productivity gains. What the industry needs is a wave of innovation that will dwarf what mobile and cloud unlocked," said David Crawford, chairman of Bain's global Technology practice.
[...] There are reasons to think the AI industry is not going to hit $6 trillion in annual revenue by 2031. The Register has published numerous reports indicating that AI rollouts are just not paying off, such as this one, this one or this one.
At the same time, doubts are being cast over whether all those AI infrastructure projects will be realized. A report from investment bank Jefferies found only half the US datacenter capacity scheduled for 2026 is actually under construction and work is yet to begin on as much as 80 percent of the 2028 pipeline.
Another report from the same company suggested that chip manufacturing constraints will limit how many planned AI server farms can be brought online over the next several years.
(Score: 5, Funny) by turgid on Wednesday October 07, @09:27AM (29 children)
Will there be the mother of all firesales when it all falls over, like the dotcom crash only much bigger?
Will we be able to buy cheap hardware? What will happen to all the companies which have spread themselves too thinly?
Will governments bail them out, like the 2008/2009 banking disaster?
I'm getting friendly with Claude. It would be a shame if he had to leave.
I refuse to engage in a battle of wits with an unarmed opponent [wikipedia.org].
(Score: 5, Interesting) by zocalo on Wednesday October 07, @12:15PM (17 children)
Also, to hell with any bailouts; it's not like any of the AI businesses that might fail are essential services in the way that it was claimed that banks were in 2008 (YMMV on how correctly). Same as with anything; if you've pinned all your prospects (financial wellbeing, business model, whatever) on a single horse in this race rather than having a Plan B, and ideally a Plan C & beyond as well, then it's entirely on you if it falls before the finish line. You might want to check just how exposed your pension plan is on all this crap though, and maybe get a few things adjusted to de-risk it a bit though - especially if you're heavy on stock market index trackers because they'll have a truckload of AI-linked stocks like Google, Micron, & Nvidia in them right now.
UNIX? They're not even circumcised! Savages!
(Score: 4, Interesting) by JoeMerchant on Wednesday October 07, @12:30PM (2 children)
>unless you really think three quarters of the world's population (or their employers) are going to pay approx $1,000 per annum just so they can all break even, let alone make a profit.
Is this $6T per year in the initial buildout, or sustained going forward?
If it's only for the next 3-5 years, that's not unimaginable. .com and the subsequent fiber buildout attracted roughly $4T in investment based on pets.com and hope for the future.
The US railroad buildout of the 1800s was equivalent to roughly $5000-6000 per capita in 2026 dollars.
🌻🌻🌻🌻✌️ [google.com]
(Score: 4, Interesting) by zocalo on Wednesday October 07, @03:03PM (1 child)
For their fever dream to work, line needs to keep going up so while the initial buildout may eventually slow, the BAU operating costs are going to accrue throughout that process before they start to plateau (ignoring inflation on utilities costs and assuming newer hardware does more with the same power input). That seems to be the fundamental difference between this and the original dotcom bubble; the latter was a significant up-front CapEx transitioning into a substantially lower OpEx with a reasonably long cycle time on hardware renewals. By comparison, AI seems to have an on-going OpEx that is proportionately *much* higher and, to keep up with the Joneses, you're going to have to keep upgrading the hardware to keep your premium tier competitive, which means a higher sustained CapEx as well.
So yeah, if all goes to plan, the rate of DC builds might start to slow, but their compute contents will need to be continually churned to keep up with the state of the art, so that keeps annual CapEx high, and if you look at how power consumption of GPUs has evolved over the last few generations; that line most definitely has kept going up and I doubt that's going to be much different for specialised AI processors, so it's also quite doubtful that the OpEx will actually plateau even if DC builds stop altogether.
UNIX? They're not even circumcised! Savages!
(Score: 2) by JoeMerchant on Wednesday October 07, @04:58PM
Yeah, these are the kinds of numbers that start to break models based on historical ratios and patterns. I'm sure the "experts" are doing their best crystal ball gazing to account for that, but what I see is: predicted revenue of $1.5T and expecting a 4x spend based on... well, their best crystal ball gazing.
It's like when Warren Buffet goes to sell his position in a company - things work differently when you're not only the biggest fish in the pond, but you account for a signifcant proportion of the pond volume all by yourself.
I think the main thing that needs to happen to fix the ratio is to stop giving away free trials to make AI videos of naked children with animal tails and ears and similar heavy usage with very niche / limited future monetization value. By now we should start to see some below 1.0 cost/value ratios in certain fields - continue to encourage growth there, and start to discourage it where we're above 5:1 and showing no signs of falling.
🌻🌻🌻🌻✌️ [google.com]
(Score: 3, Interesting) by VLM on Wednesday October 07, @03:54PM (7 children)
Oh, I have opinions, but I'll stick strictly to verifiable numbers and their obvious implications.
I have a client addicted to ClickUp which is like AI powered todo/knowledgebase/workflow. It is subjectively pretty cool. They're addicted to the tune of $660/person/yr aka $55/person/month. They gave me a contractor guest account I don't know how that works financially maybe just some IRS workaround for them to not providing my tools directly, I don't think I cost them the full $55/month, in fact I think I'm "free". Every small business I know of including my own, uses or even pays for Canva (sorta AI sorta graphics art cloud) is like $300/yr per person for the fancy pants version, which is pretty nice for presentation generation. I think I make more than $300 revenue off good presentations, how much of that is due to Canva vs my high level of manly charms is unclear of course LOL. Add some Claude/Gemini/etc. So yeah $1K/yr in at least some fields is not irrational estimate at all. I would be willing to pay for ClickUp if I didn't have a free contractor account at this part time client. I'm pretty close to just doing it for my own business purposes for "workflow automation and integration". I guess SaaS has gotten to the point where a small timer can have fancy "workflow automation" tools if they're smart enough to profitably use them... interesting.
Does this client get more than $1K of value per employee per year? As near as I can tell... yeah they do it is pretty cool and I could see many other companies spending like that profitably. That's the good news that at least some people are paying $1K and making a profit off it.
The bad news is I googled around, and ClickUp has about 20M "users" some of which are probably free tier or not the fancy "enterprise integration SSO" options like this client, etc. So less than 20M actually paying, compared to world population of 8B+ well that is a BIG problem.
Most jobs worldwide likely have $0 profitable use potential for AI. I think in the long run AI will end up like Ansys HFSS, annual license costs like $5K last time I checked, and its possible to make a hell of a profit off that expense, but less than 1% of the population can make a profit off that $5K annual license. Maybe thats too industry specific. 99% of the population can't understand what HFSS is supposed to do much less use it. Mathematica annual professional license for a business licensed contractor like me is a bit over $2K/year and I don't subscribe but I've occasionally thought about it. Personal pricing is only like $200/annual but I'd mostly want to use it for work so ...
My point of the above is I think the percentage of the population who can generate more revenue than AI costs is pretty darn low, and thats the real problem. A couple thousand here and there is pocket change in some industries, the problem is the subscriber pool in those fields is like 1 in 10000 maybe as much as 1 in 100 at most so budgeting on "the entire world will subscribe" is pretty funny.
Right now as a consumer if you're smart enough in certain industries and willing to do it, you can make a lot more than $1K/year by spending only $1K/yr on AI. The problem is 99.9% of the population will absolutely NOT be playing along LOL so financial estimates assuming 100% of the population will subscribe is pretty funny. Also I think there's roughly 0% of the population who can make more than $100K/yr off spending $100K/yr on AI, so they can't just overcharge the whales like they do for math and electromagnetic field solvers and to a lesser extent CAD. I don't think there's any way they could give AI away for free for 99.99% of the population and charge me $10M/year where I'd make more than $10M/yr profit.
(Score: 2) by JoeMerchant on Wednesday October 07, @07:02PM (6 children)
Another problem in the "less than 20M actually paying vein - if everybody (all 8B) is paying $1000 per year ($3 per day, half a cup of coffee lately) - how are they ALL making more than $1K per year additional profit? Are they growing additional crops for less inputs and labor? That might fly, but - as a global population really all it's buying them is free time, because people only eat what they eat. In entertainment, they're using fewer people to sell more movies - again, free time for the workers, and I suppose that's free time the workers can consume more entertainment in, but... where's the money coming from to buy the entertainment. Are we getting higher pay AND more time off? Seems like maybe we should be, at least to the tune of maybe $5 per day...
🌻🌻🌻🌻✌️ [google.com]
(Score: 2) by VLM on Thursday October 08, @03:21PM (5 children)
Use aggressive levels of technology to coordinate the local farm co-op, AI replaces all the office workers sounds bad if you're an office worker, but sounds good if you start up a company in the off season using AI...
(Score: 2) by JoeMerchant on Thursday October 08, @03:42PM (4 children)
> AI replaces all the office workers sounds bad if you're an office worker, but sounds good if you start up a company in the off season using AI...
The logic behind the economy needs to shift from 1700s "fruits of your labors" to ... any number of other things.
Raw capitalism already has a winner-takes-all issue which we have only partially mitigated.
Communism as implemented in Russia and China post WWII named certain people (party officials) winners and the losers really didn't have any incentives to try to be anything but losers and late stage Western capitalism is headed there fast.
Working Scandinavians like to bitch about their taxes, and yet they self-report some of the highest happiness metrics and are objectively measured as some of the highest quality of life and "developmental index" countries by outsiders.
If AI actually pans out to "shake things up" significantly, I'd much rather use the opportunity to remodel economies closer to what seems to be working FOR THE MAJORITY OF THE PEOPLE rather than heading back to Tsarist Russian economic models.
🌻🌻🌻🌻✌️ [google.com]
(Score: 2) by VLM on Thursday October 08, @04:21PM (3 children)
I'd propose its reasonable to expect AI to move the opposite, rather than one org owns and runs all, the natural result of "AI helps out" will help out little companies more than big companies.
I personally have friendships with two canva users. I do my computer nonsense so I use AI in canva to make presentations mostly. Another is two brothers running a car mechanics shop and they do ALL their graphics art marketing using Canva AI and the other is what amounts to a music instrument teacher and she also does all her social media nonsense using Canva AI. They make a lot higher profit off AI than a giant company that already has a marketing department.
I suspect the short term effect of AI will be big companies firing most of their employees, but the long term effect is those "corporate office" permanently unemployable people will just fling up little businesses.
There's a lot of "friction" in business that AI can eliminate, so the car mechanic that's graphics arts illiterate can, with AI, still run a car mechanics shop, and the music teacher lady who's social media illiterate can, with AI, still be a music teacher lady. I suppose I should be doing something more productive than shitposting on SN with my AI provided spare time because I don't waste as much time on presentation slides LOL, I'm half retired maybe without AI I wouldn't be able to be retired at all, who knows.
Big orgs are always inefficient and very bad at using new opportunities. Little orgs (including sole proprietorships) will be the wave of the future at least WRT AI use. I wonder if ANY large companies make net positive revenue or profit off AI, the profit is probably all at small companies/orgs.
Would old SEARS dept stores make more money off AI or the two brothers down the road with their mechanics shop? My bet is on the bros. I've seen what big companies do with information technology; it ain't pretty like making sausage.
(Score: 2) by JoeMerchant on Thursday October 08, @05:29PM (2 children)
> the natural result of "AI helps out" will help out little companies more than big companies.
Academically, I agree with that first blush analysis.
Having spent 20+ years in small companies and the last 10+ in a big company... I'm not sure which way AI is going to tilt the playing field. Things are already so heavily slanted toward big company advantage, and AI promises to gut a lot of the redundant bureaucratic work that big companies shackle themselves with where little companies more tend to ignore those things at risk already anyway. Then we get into the ability of big companies to negotiate big agreements where little companies are stuck ordering from the menu - that could go very poorly for the little companies if the big companies start successfully negotiating for menu prices to be raised...
I'm still seeing factors of 100:1 and more in "token pricing" between various ways you can access and pay for AI. At the moment, my private subscription account is still quite useful at $20 per month (though it just took what looks like a 33% reduction in available token-time for the same $20/month rate last month), and meanwhile I can do similar work on our corporate Cursor account and burn $80 in a single day. I think it's fair to say: pricing models aren't settled yet.
> two brothers running a car mechanics shop and they do ALL their graphics art marketing using Canva AI and the other is what amounts to a music instrument teacher and she also does all her social media nonsense using Canva AI
Here, I wonder what net effectiveness either of those business are getting from AI graphic art - I mean, compare their marketing campaign with 1990s block text laid out using a simple editor - are they actually winning customers away from competitors using old-school graphics? Are they "creating a larger market" by having cool graphics? Even assuming the AI graphics aren't recognizable slop that generates backlash, if they are "maximally effective" evocative imagery with clear communication of the actionable deal-closing steps... how much does that improve customer procurement vs a cheap pre-AI marketing campaign?
We have been working with a local events promoter, and we happened to run a killer event a couple of months ago, hundreds of attendees (vs baseline of maybe 50), and there were AI generated flyers involved - but they weren't new for the good event, AI flyers flopped on earlier events, and later ones too. So much of what makes an event attract a large crowd, or not, is far beyond our control as promoters. Maybe back in the 1980s you could purchase TV or radio advertising time that would make a big difference, but now you have social media - just like all your direct and indirect competition does.
I might see AI "blitzing" social media, so you create one message and your agent reformats it and promotes it to Farcebook, Xitter, BlueballsSky, appropriate Reddit forums, Instasham, and all those other things I understand even less - that _might_ help, some. I will note that I engaged with a shareware promotional software in the 1990s and got my warez listed on hundreds of shareware guides and distribution CDs, resulting in more website traffic and downloads, but still zero actual donations.
>They make a lot higher profit off AI than a giant company that already has a marketing department.
Is that still true when the giant company downsizes 90% of their marketing staff that's working in AI acceleratable roles?
> those "corporate office" permanently unemployable people will just fling up little businesses.
My impression is that the majority of the unemployed aren't going to have enough personal capital to fling up significant businesses and they're going to end up engaged in low level activities like MLM schemes and menial paycheck jobs. All that pandemic and post-pandemic bitching and moaning about "nobody wants to work anymore" will slowly turn around as people who want to work have no choice but to accept the shit jobs that the moaners are offering.
> the car mechanic that's graphics arts illiterate can, with AI, still run a car mechanics shop, and the music teacher lady who's social media illiterate can, with AI, still be a music teacher lady
I know/knew plenty of car mechanics who got all the business they could handle without graphics arts literacy, and music teachers who are more than fully booked through word of mouth. Music teachers who can make a living remotely, with no travel via video conference would be an interesting twist - requiring a lot of competitive edge - but face to face, no you hold your hand like _this_ teaching doesn't seem to be suffering out-competition from Instagram savvy individuals.
> I should be doing something more productive than shitposting on SN with my AI provided spare time because I don't waste as much time on presentation slides LOL
I'm shitposting while waiting for my LLM agents to chew through their next slice... such a revolutionary wonderful quality of life improvement, eh?
> half retired maybe without AI I wouldn't be able to be retired at all
Now that I'm 96%+ work from home, I'm not sure if I'll ever trade my paycheck for retirement. If this is "work" - I can handle that a lot longer than I was planning to in my 40s.
> Big orgs are always inefficient and very bad at using new opportunities.
That's a mostly true observation, but I'm witnessing dramatic uptake of LLM agents across our big inefficient org. We're filling out more BS forms and documentation faster, more completely and more accurately than ever before - and it's actually working better than it used to as a result. The larger corporate mechanism isn't accelerating very rapidly, but the cogs are spinning faster and smoother than before and it's kind of inevitable that the big beast will speed up eventually. Mostly we're still paralyzed as ever about knowing what we should be doing - that's nothing new at all, but now when a decision is finally reached we are effectively acting on it faster.
> make net positive revenue or profit off AI
The big org I am in has like 80% market share in our sector, we're always driving for more, but at a certain point there's just no more blood to squeeze out of those turnips. I suppose our people could become so effective that we could start serving multiple sectors with a single team, but that's a needlessly risky strategy for a small gain. Feet on the street are still required to close deals and keep those little startups under our bootheels, AI isn't changing that.
> I've seen what big companies do with information technology
My last boss was all jazzed about Power BI Dashboards, and hiring contractors / consultants. I don't think it matters what kind of tools you give people like that, you're usually going to end up spinning wheels with no traction. He finally left to do something with a partner elsewhere, they had big plans to sell software to a niche market, but by the time they did the market research they ended up writing articles about the market and selling them to trade magazines for more profit than they ever made with the software (vaporware) they were planning to develop based on the market research. They didn't hit the point too hard in the articles, but the truth they found in their market was: the customers are loyal to the misery they experience with their current software providers, they have tons of ideas of what needs to improve, but don't trust others to actually deliver on promises of making those improvements - so sales traction for startup competitors is, very challenging.
Lots of people will be "rolling their own" software solutions in the coming 5-10 years, I wonder if that won't be destroying more market for sales and support of such software solutions than it enables for startup software providers?
Ironically, my first "real" job, through the 1990s, was in-house software developer - doing for them what programmers armed with LLM agents are likely to do for a lot of companies in the future.
🌻🌻🌻🌻✌️ [google.com]
(Score: 2) by VLM on Friday October 09, @03:49PM (1 child)
They're small timers, they don't "do" marketing if its not "type a prompt into canva and click the order button if you like the flyer cards"
Confusion of having data with having 'actionable' data. If its not actionable, its tire spinning. "Old Timers" couldn't get any data at all so its a bragging point to have data even if its useless.
I've found so far that people who can program can program a wee bit more with LLMs as a tool, but people who can't program continue to be unable to program even with LLMs.
Pretty much any software bug can be described in English prose and you'll have people typing that into LLMs and needing a programmer. A race condition is a race condition in any language including English, see also concurrency deadlocks, etc. Its just English Majors and Art History Majors can't debug those and CS grads can, even if they program in English now instead of C++. So thats my view from the trenches so far. Its even more basic than "esoteric CS stuff" there's non-programmers who refuse to accept GIGO. "But can't I just prompt the AI until it works?" Uh no.
(Score: 2) by JoeMerchant on Friday October 09, @06:01PM
> people who can program can program a wee bit more with LLMs as a tool,
Well, for my personal projects, I've built things about 5-10x more complex / time intensive than I used to build in less time and for much less focus-effort than I used to spend on the simpler stuff.
For my professional code reviews, I'm about 100x more thorough than I used to be when I use LLM agents to do the reading of the pull requests. And my professional code output tends to be lower effort on my part while being more thorough about documentation, edge case and test coverage. There are aspects that don't change much, but overall - I'd say it's impactful.
> but people who can't program continue to be unable to program even with LLMs.
Yeah, there's a lot of grey around that: people who could program but never bothered to before because they didn't have the time. We also have "programmers" who were dismal before and they're still pretty dismal.
> A race condition is a race condition
Coincidentally, those dismal programmers tend to generate race conditions - even in Rust - and LLM reviews are pretty helpful at finding, and fixing, them. With the LLM reviews, I can suggest targeted functional fixes to dismal code whereas in the past I'd be frustrated down to the core of the system and often want to throw it all out and start over more sensibly.
> "But can't I just prompt the AI until it works?" Uh no.
Well, I kind of just said that that is working for me - sometimes. Are targeted fixes better than architectural re-implementations? Depends on the situation.
🌻🌻🌻🌻✌️ [google.com]
(Score: 3, Interesting) by JoeMerchant on Wednesday October 07, @09:04PM (3 children)
>For sure it's going to be like the dotcom crash
Not specifically like the .com crash - .com money all ran to fiberoptic infrastructure investment, laying fiber that stayed dark for 20+ years after people sobered up after going "all in" on pets.com and similar.
Over-investment and financial crash/panics are a tale as old as time. Each has its own twists.
One thing that is a bit different about this AI runup is the scale - it's far bigger than .com, more on par with late 1800s railroads. Unlike railroads, the "iron" being invested in is shorter lived, but more fungible - applicable to different fields and sources of income.
A crash is near inevitable, but the nature of the crash is likely to be a bit unique.
🌻🌻🌻🌻✌️ [google.com]
(Score: 1) by khallow on Thursday October 08, @12:27AM (2 children)
Actually, it probably is somewhat bigger [yahoo.com], but not a lot bigger than dotcom at present. The link is to the lifetime Wilshire 5000 with logarithm scaling. I like this index because it provides a good sense of scale - it's total capitalization of 5000 publicly traded businesses and provides a decent sense of how things fluctuate relative to the overall market rather than just out of context numbers.
Going from the 1991 dip in the index to the 2001-2002 peak is about two thirds of the run up from the 2009 trough in the real estate crisis through to the present day. Similarly, the runup to the dotcom peak is as steep as the present runup in the index.
(Score: 2) by JoeMerchant on Thursday October 08, @01:43AM (1 child)
When comparing events over long timescales, the reference measurement scale is important.
Dollars are clearly naive, you need to adjust for inflation.
But, when talking about how big a bubble is, it's not just inflation adjusted dollars, it's more a percentage of GDP. Global GDP is 2-4x (depending on how you want to look at it) what it was 30 years ago.
Then there's a question of whether or not per-capita should factor in. Even though the population boom was "solved" 30+ years ago, global population has still added +43% in those 30 years.
🌻🌻🌻🌻✌️ [google.com]
(Score: 1) by khallow on Thursday October 08, @02:25AM
Logarithm helps a lot with making comparisons where large shifts occur due to inflation or per capita viewpoint. The reference measurement scale becomes differences between states.
(Score: 0) by Anonymous Coward on Thursday October 08, @02:28AM (1 child)
Those involved are not necessarily stupid. It's like a legal pyramid scheme. You can make a lot of money as long as you get out early enough.
And often if they don't, they get a bailout and get to keep their bonuses etc anyway.
(Score: 1) by khallow on Friday October 09, @12:18AM
Greater fool theory.
Executive bonuses are hard to take away once they're paid.
(Score: 3, Interesting) by JoeMerchant on Wednesday October 07, @12:17PM
I also wonder... when this all shakes out, how much of this nifty hardware will become accessible for "ordinary" home PC use?
Will this high performance componentry remain in the domain of $20K-$100K computers (which, adjusted for inflation, aren't that different from 386 prices in 1990) - or will it be headed "downmarket" after the initial costs of production capacity are recouped? Or, will we face a "memory cartel" that keeps profits high and supply constrained (for your own safety, of course, to keep "dangerous tools" out of the hands of the little people)?
🌻🌻🌻🌻✌️ [google.com]
(Score: 3, Interesting) by ikanreed on Wednesday October 07, @02:28PM (5 children)
I don't know, do you want a GPU that's useless for graphics?
(Score: 2) by turgid on Wednesday October 07, @02:34PM (2 children)
Why would it be useless for graphics?
I refuse to engage in a battle of wits with an unarmed opponent [wikipedia.org].
(Score: 2) by VLM on Wednesday October 07, @04:13PM (1 child)
Quite a few accelerator cards have no or minimal graphics port outputs. You can buy some "GPUs" with no outputs at all.
Famously, you can do cool non-graphics things with GPUs.
Some cards belong in a Proxmox cluster like mine where you can run virtualized graphics card games that think they have an accelerator because they do, its just not connected to anything. Or CAD. I don't have GPUs in my cluster at all, been thinking of getting some ebay specials for the experience of setting up GPU passthru. I don't know exactly how that works but time and funding permitting I will.
https://pve.proxmox.com/wiki/NVIDIA_vGPU_on_Proxmox_VE [proxmox.com]
I like how they call out using the "RTX PRO 6000 Blackwell Server Edition" as a test bed, I'd only need about 9 of them and they're retailing for about $10K so naaah this ain't happening anytime soon. But its sooooo cooooool.
Someday I'll be one of those guys buying this stuff on ebay for pennies on the kilobuck for home lab fun times. Just not today.
You can also mine crypto which probably stopped being profitable a decade ago, but who knows maybe once again? If the hardware were free and I could run a profit off solar power during the day, maybe?
(Score: 0) by Anonymous Coward on Saturday October 10, @06:10AM
A couple of people I know from work have their own racks, yes the big ones, mainframes, blade servers, routers, the whole kit and kaboodle.
You might be thinking 'who would want that'. Well, people who have worked with IBM mainframes for decades, and still do. Networking, Cyber, Security. The type of people who think it is fun to build their own little datacentre in their garage. Just for kicks.
The equipment was cheap. Like, really cheap. As in, my company gutted two really old data centres, built for mainframe and midrange, and just fire saled the lot. Racks. Blade servers. Everything. Well, just about everything.
One guy was taking blade servers to turn them into home computers and routers for friends and family.
Just imagine the mountains of hardware these data centres might spew forth when they bite the dust. Now imagine the tax write downs/offs. I wonder how much they can get back after.
(Score: 1, Interesting) by Anonymous Coward on Thursday October 08, @11:11AM (1 child)
It's not useless for me to run all the models I can't fit now. AI is useful but not to the extent they're spending.
(Score: 0) by Anonymous Coward on Saturday October 10, @06:16AM
I got a new computer last year. Nothing too fancy. 64GB ram, 17GB drive space, I9 CPU, RTX 4060. Wish I had got 128GB ram - thought I would upgrade later.
It's next to useless for AI. I could run some things, but I'd probably be better off just using the cloud. It would need twice the ram, at least, and that is just not happening. Sad times. Can't wait for this bubble to burst.
(Score: 4, Insightful) by Thexalon on Wednesday October 07, @04:54PM
See, this is why it's important to know that we shouldn't be simply burning down data centers, we should be doing a smash-and-grab to extract as much valuable hardware as we can from them first.
At the very least, I suspect RAM will become massively cheaper when the bubble bursts.
For reference, $6 trillion is approximately the same scale as the entire US federal budget.
"Think of how stupid the average person is. Then realize half of 'em are stupider than that." - George Carlin
(Score: 3, Touché) by Reziac on Thursday October 08, @04:50AM
The mother of all fire sales presumes that most of this hardware actually exists.
And there is no Alkibiades to come back and save us from ourselves.
(Score: 4, Interesting) by driverless on Thursday October 08, @01:47PM (1 child)
Trump has significant AI holdings so yes, the USG will bail them out. I'm sure the High IQ! president will come up with a solution like printing more money to make up for the current shortfall to fund it all.
(Score: 1) by khallow on Friday October 09, @12:57AM
He would need to stay in office long enough for that to happen. I presently think there's a large chance the bubble holds together till after the 2028 election due to Trump administration machinations.
One thing I find sad about this bubble as compared to the dotcom bubble is that then President Clinton almost balanced the federal budget for a year (financial year 2000) on the strength of the dotcom bubble. Here, the federal deficit is presently around 6% of US GDP. Federal tax revenue won't get much better than it is now! So what will happen when the bottom drops out of AI?
So we'll see a one-two combo: efforts to bail out AI combined with a massive budget deficit due to the subsequent recession.
(Score: 3, Funny) by Anonymous Coward on Wednesday October 07, @11:09AM
... if we can make billions?
(obligatory quote)
(Score: 1, Touché) by Anonymous Coward on Wednesday October 07, @11:19AM (1 child)
Don't forget who Bain & Co are -- https://en.wikipedia.org/wiki/Business_career_of_Mitt_Romney [wikipedia.org]
(Score: 3, Insightful) by Thexalon on Wednesday October 07, @04:43PM
I mean, that's certainly who they were, but he's had a much better grift going for a long time, first as governor of Massachusetts, then as a presidential candidate, and lastly as a senator from Utah. As a businessman you have to come up with the schemes yourself and raise the capital to do it with, whereas as a politician the people come to you with money they want to give you to be a part of their schemes.
And if you peruse OpenSecrets, you can see how remarkably cheap a lot of politicians can be bought for.
"Think of how stupid the average person is. Then realize half of 'em are stupider than that." - George Carlin
(Score: 4, Interesting) by Barenflimski on Wednesday October 07, @12:45PM (2 children)
These articles get so much traction, the writers just can't wait to write them. These are basically the same voices that said that solar power prices would never ever match coal. Wouldn't surprised me if these guys just want to make money shorting the market.
The reality is that the cost is going to plummet to deliver these services. The hardware demands are going to become less and less. As this happens, and they need less data-centers, they may take a small hit to cancel a contract, but that's not these Trillions these folks like to throw around. Once these are small enough and work well enough, everyone will be paying for part of this with every new Microsoft OS license.
On top of that, I find it wild that all the doomer and gloomers read this stuff daily, salivate over it, and then also tell their local folks to ban data centers, knowing full well that if they do that, they could help bring this "everything goes to hell" scenario to life, assuming nothing changes. I mean, if they want a recession, go ahead and will it into existence.
These companies are investing billions in bringing their compute costs down. The models are already becoming smaller, ie: less footprint and faster. The hardware being developed will bring power and compute prices down significantly. That leads to needing smaller footprints, so likely they wont need all of these datacenters by 2031, which is a win-win for everyone.
Another scenario that could just as easily play out is that these large companies invest in our power grid, bring massive power online, while at the same time figuring out how to do this AI with less, therefore leading to some of the lowest energy prices we have ever seen.
(Score: 4, Touché) by VLM on Wednesday October 07, @04:17PM
Energy prices are usually at generational lows during economic collapses. "Low prices" sound good on paper, but you really don't want to be there when it happens. Think like 1933, 2008... If we ever get $32/barrel crude again you'll enjoy a few minutes paying your bill but the other 23 hours of the day will not be a good time LOL.
(Score: 2, Insightful) by khallow on Thursday October 08, @12:10PM
The problem with that is that they're investing trillions. That becomes the cost.
A fair bit of that is people who want a reality where they are right. Self-fulfilling prophecies are a sleazy way to make that happen.
(Score: 5, Funny) by looorg on Wednesday October 07, @06:44PM (1 child)
Good luck with that. Perhaps Musk, Zuck and the others should start an OnlyFans page .... So the ROI on this is poor to say the least. Also so expensive that they can't really afford it so they are now borrowing heavily to try and keep up. Keeping up with something that is going to be $4.2 trillion in the hole in a few years. All the years to that point are also going to be in the red. They better come up with one hell of a app or service that will make up that massive short fall.
Who the f*ck is lending these muppets money? It would probably be better if they just bought one of Elons flamethrowers and just burned it to the ground.
(Score: 2, Interesting) by khallow on Friday October 09, @01:01AM
Some of this is circular borrowing. Spreadsheet asset # goes up, spreadsheet liability # goes down.