Data Center Boom May End Up Being 'Irrational,' Investor Warns (axios.com) 28
A prominent venture capitalist has warned that the technology industry's massive buildout of AI data centers risks becoming "irrational" and could end in disaster, particularly as companies pursue small nuclear reactors to power the facilities. Josh Wolfe, co-founder and partner at Lux Capital, compared the current infrastructure expansion to previous market bubbles in fiber-optic networking and cloud computing. While individual actions by hyperscale companies to build data center infrastructure remain rational, Wolfe said the collective effort "becomes irrational" and "will not necessarily persist."
The warning comes as Big Tech companies pour tens of billions into data centers and energy sources, with Meta announcing just this week a deal to purchase power from an operating nuclear station in Illinois that was scheduled to retire in 2027. Wolfe said he is worried that speculative capital is flowing into small modular reactors based on presumed energy demands from data centers. "I think that that whole thing is going to end in disaster, mostly because as cliched as it is, history doesn't repeat. It rhymes," he said.
The warning comes as Big Tech companies pour tens of billions into data centers and energy sources, with Meta announcing just this week a deal to purchase power from an operating nuclear station in Illinois that was scheduled to retire in 2027. Wolfe said he is worried that speculative capital is flowing into small modular reactors based on presumed energy demands from data centers. "I think that that whole thing is going to end in disaster, mostly because as cliched as it is, history doesn't repeat. It rhymes," he said.
transformers are quadratic (Score:4, Interesting)
People are quick to jump on the LLM bandwagon, but everybody who knows knows transformers are an algorithm that will be replaced (are already being replaced)
Everybody thinks "THIS IS THE WAY" no it isn't, we are still getting there, calm down and use that nuclear reactor for CO2 mining
hardware and software (Score:3)
This is still early days for AI. There's going to be an optimization curve where the energy cost of running the hardware is significantly reduced;
https://blogs.nvidia.com/blog/... [nvidia.com]
And the same thing with the software. Fewer tokens required for the same performance etc.
https://guptadeepak.com/comple... [guptadeepak.com]
But at the same time the demand will ramp up as more human work is offloaded to the machines, and this will potentially soak up all the gains of optimizations and swamp the infrastructure.
Re: (Score:3)
You mention consistent rends in semiconductors to reduce size and energy, but ignore the fact that the sheer mass of installed computing equipment has continued the entire time that this reduction in chip size and energy consumption has been going on
While there is an element of hula-hoop hysteria in descriptions of the continuing data center buildout, they will build out new datacenters as long as there is money in it, and IMO there will be demand for AI personal assistants for the mid-term 15-25 year horiz
Re: (Score:2)
>> there will be demand for AI personal assistants for the mid-term 15-25 year horizon
I don't doubt that's true. But the personal assistant may have largely been squeezed into the augmentation device you carry on your person and not require access to a data center.
Re: (Score:2)
the personal assistant may have largely been squeezed into the augmentation device you carry
I don't see that happening. Your data and everything you need the device you are carrying to do ends up requiring communicating with resources in the cloud. Also putting its basic functions in the datacenter will give the manufacturer more control of your hardware, and they want to make sure they are in the mix logging your every command, so they can sell all that data on their customers and charge a monthly s
Re: (Score:2)
>> everything you need the device you are carrying to do ends up requiring communicating with resources in the cloud
There will be wireless/cellular access to the big hardware when necessary of course but already you can load compressed models into devices as limited as a Raspberry Pi and get useful results for specific situations. I work with AI on that scale routinely and I can assure you it doesn't phone home to share the data anywhere.
Google sells the Edge TPU, much smaller than the Pi and runs mod
Re: (Score:2)
This is still early days for AI.
Please stop lying. AI is about 70 years old and even LLM tech is quite old. What is new is internet-scale piracy to get training data and NLP has gotten a lot better.
Re: (Score:2)
>> Please stop lying
You behave like such an abrasive asshole here.
>> AI is about 70 years old
OpenAI is about 10 years old, and several of the other implementations commonly in use these days are younger. Anthropic is 4 years old for example.
Re: (Score:2)
I just have had it with cretins pushing lies. Like you.
Re: (Score:2)
Whatever. Fuck off.
Yep (Score:2)
But current executives simply do not care. That is a problem for the next person.
Cloud Computing + Easy of Access = (Score:3)
The services that require building massive data-centers are now viewed as poor opsec. Cloud computing and offstite data-storage grow your attack surface exponentially. Hiring proficient coders to design in-house solutions for local data-storage is the only reasonable option.
Re: (Score:2)
Not only. As a second bad idea crapto is required for money laundering that powers ransomware.
Fiber optic bubble? (Score:2)
I guess I'm confused, how is a technology with a decades-long service life and is basically a capital investment subject to the same sort of label of 'bubble' compared to the explosive growth of something using a commodity model with obsolescence measured in years?
In 2020 I had to occasion to have some OS1 singlemode fiber installed back in 1994 terminated into splice cases and put into use. That fiber sat for basically a quarter century and was then usable when I needed it.
Where I work now I no longer hav
Re: (Score:3)
I worked at a nascent telco that was laying thousands of miles of fiber around 2000
Their stock took some hard hits, myself and a large percentage of the staff was laid off, and they only survived through competent merger and acquisition strategy
From a business standpoint, they certainly suffered from a bubble, but their fiber is still in the ground and they will exist for decades on their massive buildout 25 years ago. I suspect these datacenters will behave in a similar manner, there will be a bust, but th
Re:Fiber optic bubble? (Score:4, Interesting)
While fiber can sit for a quarter of a century without degradation, a data center would deteriorate into obsolescence in that same period, with zero use.
The capital investment in fiber doesn't lose value, beyond opportunity cost. The capital investment in a data center, or any unused building, is a total loss.
Re: (Score:2)
if I'm not having to spend $50k to go fifty feet between buildings.
That does seem insane.. for that kind of price you can probably mount a 50 foot rail between two buildings' roofs and plate that rail in gold, then strap a 288 strand count fiber bundle to the rail in 18 carat gold twisties.
Re: (Score:2)
I guess I'm confused, how is a technology with a decades-long service life and is basically a capital investment subject to the same sort of label of 'bubble' compared to the explosive growth of something using a commodity model with obsolescence measured in years?
I think you may be conflating ROI and immediate ROI. Investement-wise, what you're talking about is the equivalent of building housing or office space and then not being able to sell, rent, or lease it for a year or two after completion. Investors hate that shit.
Venture Capitalist Catches Up With Reality (Score:5, Insightful)
Hasn't most of the tech world outside of the AI prophets been saying this same thing since the hype cycle started? We've barely started on this AI treadmill, and the tech will change even if the bubble doesn't burst outright. Where we are today, the techbros have decided to build out Model T Ford building capacity at a "make sure every family owns six of these cars, even if there are only four people living in the house!" We've not just outpaced demand, we're building out as if we expect the demand that we're already outpacing will explode exponentially into infinity for the foreseeable future and beyond.
make them absolutely responsible for (Score:3)
Self Correction (Score:4, Interesting)
I don't know if this guy has been on vacation or living under a rock, but there was already a correction [reuters.com] last month. Microsoft dropped 2GW worth of DC leases (on top of the several hundred MW they did in Feb) which flooded the market with inventory. Two of my customers immediately dropped out of work that was being done on other data centers because they knew they would be able to pick up space sooner and for less money as a result. Everybody in the industry saw a pull-back. Where we were working on designing and selling inventory that was 24 months out now nobody wants to talk about anything that is further than 12 months away from being ready.
Tying new data centers to old nuclear plants has a whole host of other issues around it that make me think this will end up being a nothing burger (SMRs are another matter), but this supposed irrationality of a capital system working as intended seems ill conceived.
Correction (Score:4, Insightful)
There may have been a few companies that have spent "billions", at least in aggregate. But, there have been dozens of companies claiming to be "about to" spend billions, individually, on data centers. This is of course bullshit born out of irrational stock market activity and the desire to pump share prices.
Many of the major players that actually had the capital to spend billions on data centers, and said that they would, have already walked back their statements, paused, pumped brakes... The most prominent, that I can think of right now, being Microsoft. https://apnews.com/article/mic... [apnews.com]
AI 2027: A Realistic Scenario of AI Takeover (Score:2)
imagination (Score:2)
Here's the first sentence for mid 2025: Advertisements for computer-using agents emphasize the term âoepersonal assistantâ: you can prompt them with tasks like âoeorder me a burrito on DoorDashâ or âoeopen my budget spreadsheet and sum this monthâ(TM)s expenses.â They will check in with you as needed: for example, to ask you to confirm purchases.
That first sentence represents something far from the capabilities of any AI we have. Sure there are programs that
I lived through the telecom busto of early 2001 (Score:2)
you know, the bust that preceded the Intyernet bubble bust of mid 2001
Fixed carriers overinvested in fibre backbone deployments, meanwhile, mobile operators overpaid for 3G spectrum, leaving them with a massive financial burden AND little money to buy telecom kit, which in turn brought down the telecom kit providers...
I lost a great job opportunity because of that...
Anywho... the Datacenter boom, be it for Cloud, AI, or CryptoMining is looking A LOT like that...
A Hyperscaler/datacenter burst is nigh, repen
Not comparable (Score:2)
compared the current infrastructure expansion to previous market bubbles in fiber-optic networking and cloud computing.
The problem with the fiberoptic build-out was that it was speculative expansion in the hopes that the expanded capacity would eventually be used. When that capacity wasn't used, then the expansion became a bubble. The current data center explosion is added capacity that is already being used.
Cloud computing growth rates are indeed slowing down. However, current growth rates for AWS and Azure are 15-20%. Microsoft has said that their growth rate is currently gated by supply rather than demand. So, I wou
adding computational and electrical capacity is (Score:2)
Adding computational and electrical capacity is one of the most rational things that can be done with that money they're willing to burn... i mean invest. There is nothing presumed about the growing demand on the grid by data centers... it was an issue PRE-AI boom.... Is it just taken as normal when people who happen to have some cash speak ignorance and have the memory of a fucken gold fish? Why did this idiocy get any air time on the news. Sure, he said it at a conference.. but stupid is stupid. Not