That's true, but it's bipartisan policy. The Trump admin obviously is fighting wind, though solar is just too useful to be able to be stopped even by them. But it's also popular among environmentalists to fight solar projects https://v2.courthousenews.com/greens-fight-mojave-desert-sol...
And Indian Point in NY is the classic example of how nuclear was supposed to be replaced by renewables but was replaced by gas (and many solar installations in NY were blocked so we're not coincidentally here): https://www.theguardian.com/environment/2024/mar/20/nuclear-...
So, it's political in that our policy is in line with what most people want: fewer renewable generation stations. But there's no outside enemy to conquer here. We've just collectively decided on this equilibrium.
The primary bottleneck to this growth is the availability of electricity.
The bottleneck for building some AI datacentres and switching them on is electricity, sure, but that's not what drives growth. There also needs to be demand for the additional capacity; people need to be waiting for capacity to catch up so they can do the useful work that grows [society|GDP|something] that they aren't doing right now.
There's also very likely to be diminishing returns from additional capacity if we're near or over the limit of productive use. And there's the opportunity cost of what could have been done with that [money|land|electricity].
This is a much more complicated system than "people say they need more AI -> build datacenter -> power datacenter -> magical growth!"
We should have been building more nuclear. We’re not going to upgrade civilization with windmills and putting on sweaters. Think of how much power we’ll need for millions of robots?
TFA isn’t about consumer usage, it is about training the next generation models. An interesting thought I heard recently is that a SOTA model has about the same parameters as the number of synapses in a Golden Retriever’s brain that are not dedicated to biological processes like breathing. Most of that should be wrapped in double quotes, don’t take it literally! But that number is about 100x lower than the same synapses in a human brain.
If the next order of magnitude costs 40B, I wonder if it’s even possible to get to the one after.
Cheap ubiquitous distributed power systems will change the world in many weird ways. Watch small modular nuclear offer home installation for ~$reasonable and getting cheaper every year.
Fast forward 20 years from the advent of essentially infinite energy results in WWIII and a new “Great Detente” but only after all the assholes have wreaked all the havoc they can.
There are dark days ahead but ultimately a brighter future. Sucks to live through that transition phase though.
Question for the experts: does the power crunch mean that AI hyperscalers will turn off previous generation GPU datacenters to free up power for their new Vera Rubin GPUs?
This is part of why the AI bubble will burst. The only way to make the profit numbers backing the loans to AI companies is to get increased capacity, and the capacity requires energy, and the energy won't arrive in time, only partly due to all the factors here, and partly because building transmission and generation is speculative and can fail for a number of reasons.
US administration can try to pull a China and basically remove all regulatory barriers (following existing playbook of "do whatever we want and wait a year or two for the courts to catch up and stop us"). It'll create havoc that will make people very upset (more so than the people that already protest DCs in their backyards). But even then, it's construction on varied terrain and property over long distances; you can't predict exactly how that will go. Triple the estimated timeline and that is probably doable, but current AI investment likely can't wait that long, unless somebody can pull additional hundreds of billions out of a hat to extend lines of credit or a ponzi-scheme-esque paying-creditors-with-newly-lent-money. In that time the market will realize the hype was hype, the gains were modest, they'll start divesting, and then the house comes down.
One way around that might be to deploy thousands more gas turbines and make rural air quality look like 2010 Beijing. It will probably happen if things get really tight, and we'll see how the current administrations's base responds; if they stick it out, the market gets a reprieve.
The idea is to bring the data centers, power generators and energy supply together in the ~same physical space so the only thing you have to transmit is data. Moving energy is way more expensive than moving information.
I think Elon Musk would have happily built gas turbines on high-rises if we let him build them in the middle of the city to power those electric cars. I don't think that's a good idea, though.
Yeah seemingly this is not a problem with data centers. I am sure it was astroturfing all along. Also you never hear about climate change even though that's the cause of the heat wave in Europe and the US
The politicization of AI has been a huge damper. NYS put a 1 year moratorium on datacenters with some municipalities doing so as well. My town did a 1 year ban, despite us being a retarded location for a data center (more expensive land than surrounding areas, less water access, far from power). US has also been lagging in power capacity, because of the anti-growth segment of politics. More nukes.
This should include the AI companies trying to sell their product to the government, suggesting it is a strategic asset in need of protectionism, and playing the PR game of "our product will let us own the world, we're so awesome you should be scared and invest."
Even rural localities in blood-red states are opposing datacenters. Noise, water electricity usage coupled with few permanent jobs and little tax-income due to subsidies have made them unpalatable to many.
The retardation has mostly been a case of AI companies fearmongering, lobbying the govt and the govt not reassuming people against job losses. Local governments passing moratoriums is just common sense populism.
"AI data centers" are so non-central and located in places so data-sparse (yet rich in cheap electricity) I consider them more likely fronts for crypto mining.
> But if AI is a race with national security implications, they won’t be fast enough.
I really don't think this is the case. As far as I can tell there are two national security things where LLMs have some utility; mass surveilance (ick!!!) or software security. To me that does not justify a huge infrastructure buildout considering the implications of said infrastructure.
42 comments
[ 5.1 ms ] story [ 68.9 ms ] threadAnd the reason current US policy opposes clean, renewable energy is --- purely political.
And Indian Point in NY is the classic example of how nuclear was supposed to be replaced by renewables but was replaced by gas (and many solar installations in NY were blocked so we're not coincidentally here): https://www.theguardian.com/environment/2024/mar/20/nuclear-...
So, it's political in that our policy is in line with what most people want: fewer renewable generation stations. But there's no outside enemy to conquer here. We've just collectively decided on this equilibrium.
The bottleneck for building some AI datacentres and switching them on is electricity, sure, but that's not what drives growth. There also needs to be demand for the additional capacity; people need to be waiting for capacity to catch up so they can do the useful work that grows [society|GDP|something] that they aren't doing right now.
There's also very likely to be diminishing returns from additional capacity if we're near or over the limit of productive use. And there's the opportunity cost of what could have been done with that [money|land|electricity].
This is a much more complicated system than "people say they need more AI -> build datacenter -> power datacenter -> magical growth!"
If the next order of magnitude costs 40B, I wonder if it’s even possible to get to the one after.
Fast forward 20 years from the advent of essentially infinite energy results in WWIII and a new “Great Detente” but only after all the assholes have wreaked all the havoc they can.
There are dark days ahead but ultimately a brighter future. Sucks to live through that transition phase though.
US administration can try to pull a China and basically remove all regulatory barriers (following existing playbook of "do whatever we want and wait a year or two for the courts to catch up and stop us"). It'll create havoc that will make people very upset (more so than the people that already protest DCs in their backyards). But even then, it's construction on varied terrain and property over long distances; you can't predict exactly how that will go. Triple the estimated timeline and that is probably doable, but current AI investment likely can't wait that long, unless somebody can pull additional hundreds of billions out of a hat to extend lines of credit or a ponzi-scheme-esque paying-creditors-with-newly-lent-money. In that time the market will realize the hype was hype, the gains were modest, they'll start divesting, and then the house comes down.
One way around that might be to deploy thousands more gas turbines and make rural air quality look like 2010 Beijing. It will probably happen if things get really tight, and we'll see how the current administrations's base responds; if they stick it out, the market gets a reprieve.
https://www.chevron.com/newsroom/2026/q2/chevron-signs-20-ye...
The idea is to bring the data centers, power generators and energy supply together in the ~same physical space so the only thing you have to transmit is data. Moving energy is way more expensive than moving information.
2. Companies realising that they are burning half million to get nowhere
3. Circular investment scaring investors
4. And more recently, companies hiring people back coz the AI aimed to replace humans, created more problems than solved them
5. Memory cartel falling apart, again, they did the same thing during 2000s
6. China is making good ML free, supply and demand, destroying the US tech token business model
7. Even META has too much computer power and no enough use for them.
Those are the main reasons why AI buildout is not just slowing down but falling apart faster than expected.
Premise perverted, FUD averted, false-victimhood denied.
A datacenter did not murder your family.
The real "AI" success story will be the person that makes an IRL backrooms theme park in the husk of a datacenter.
Or: laser tag park, the vests you wear are in part old tpu/gpu components.
Until free/at cost car charging is provided at work when the sun shines you're just moving the place where combustion happens.
This may be a worth while trade in downtowns where the main form of car pollution is engine exhaust at an average of 20mph.
We put the cart before the horse, lets stop talking about growth and focus on making it useful first lmao
This should include the AI companies trying to sell their product to the government, suggesting it is a strategic asset in need of protectionism, and playing the PR game of "our product will let us own the world, we're so awesome you should be scared and invest."
Of the items on this chart, I would say AI data centers are providing the least amount of value per % of GDP spent.
And 1930s public works the highest.
I really don't think this is the case. As far as I can tell there are two national security things where LLMs have some utility; mass surveilance (ick!!!) or software security. To me that does not justify a huge infrastructure buildout considering the implications of said infrastructure.