Inadvertently the opposition to AI and AI data centers may save the tech industry. It will stop them from spending insane amounts on AI, when there is no prospect for any substantial revenue.
I see it the way you do, just don't think for a minute they'll ever agree with us. They seem to see themselves as victims whenever they encounter any obstacle to their goal of total domination over the planet.
Oaktown: Yes, and "the people" who just want to live their lives in their homes and communities without "interruption" from the Arrogant Gaslighting Idiots (AGI) among us, are probably thought of as just so many backward neanderthals.
All in this thread--see my post above (or somewhere here) where the quotations are from FORTUNE 500 Digest today--showing several of the BIG companies already investing in data centers and "infrastructure." Of course, they COULD be finding ways to self-energize and actually help the people around those centers (shall we laugh now or just wait till later?)
But it seems to me, and I speculate here, that the power brokers are just going ahead with "WHAT I WANT" and to hell with the "local" protest. Let them scream. And apparently the U. S. Government, such as it is, is not on the side of "the people."
The problem, as I see it, the "power brokers" assume that the stuff they are building - AI and data centers - will make them money. They do not do now, and likely they won't in the future.
So, if locals stop them, they will save the "power brokers" money.
Richard Bielak: I saw all that big money/corporations as listed in that Fortune 500 Digest investing in data centers as a clue about the influence of local protests: None.
Franz-Xaver Huber: What's that saying? "Power corrupts. Absolute Power Corrupts Absolutely"? And there is this REALLY STUPID IDEA that the more money and power you have, the more intelligent and even wise you are. Even when some of these "entrepreneurs" have a streak of what is obviously DUMB LUCK, they tend to think it's "something about me."
They will always say they are talented, intelligent and working hard to justify their wealth and income/profits. That's part of the narrative. They don't want to share anything.
AI is the latest revolution in capitalist production, renewing (they hope) an economy of overproduction. It follows on from previous revolutions; computers, communications, aircraft, the automobile, the train and so on, until we run out of planet.
If humans can’t even get along in a bountiful paradise for which they are uniquely adapted by aeons of evolution, there is no way in Hell they will survive in any significant numbers on Mars.
If they can’t get along in the Garden of Eden, they won’t last more than a brief time in a place like Mars where everything needed for survival is scarce.
“Tribes” will be raiding and fighting one another (to the death) for those resources pretty much from day one.
The bipartisan angle is the real signal here -- polarized issues get litigated in courts and campaign ads for years without changing much, but bipartisan opposition moves zoning boards and utility commissions fast, and that's where data center buildout actually lives or dies. Worth watching state-level power-purchase-agreement fights over the next two quarters, not the 2028 election -- local utility rate cases will bite AI capex projections years before any anti-AI candidate does.
Good. Fuck 'em. Pretty soon the media will admit that it's not really data centers people are against, it's AI, but the media has a hype interest in hiding that obvious fact.
Not just AI, but also the data centers, which are mammoth sized consuming mammoth sized amounts of water as well as mammoth sized amounts of (non) renewable energy. All of that against a background of a 24/7 60 decibel hum.
There is of course also the issue of water and energy affordablity for the locals near te mammoth data center.
The political effect may be felt well before the 2028 presidential election. The WA state legislature meets each year between January and April with a couple thousand bills to consider. Until this year that included maybe 1 or 2 AI bills and 0 or 1 data center bills. In 2026, there were 8 AI bills and 8 data center bills. For the first time, I noticed someone representing a major tech company testify in a public hearing, rather than an industry lobbyist. (Tech bills generally have 4 or more public hearings in committees before passing, all webcast live and recorded.) In 2026, most bills did not pass and others were watered down with amendments. 2027 could be different, and 2028 legislation will also be long before the election. My only question about this post is "AI could have been a great thing." What would that great thing have been and how could it have happened in the real world?
I'm aware. It's gonna be a rough ride ahead. Our president is terrified and losing control over his phony narrative, which renders him desperate and extremely dangerous.
For starters, tech fascists and clueless investors could have listened to Gary years ago, when he demonstrated with knowledge and logic that LLMs would hit a wall and never achieve AGI. Then research money and resources could have been more wisely spent on finding a way to develop specific and *reliable* AI tools for limited tasks, instead of trying to sell them to the public as human beings spewing forth authoritative BS.
Wasn't this already happening? Research money and resources being spent on protein-folding and other reliable AI tools for limited tasks? Are you saying AI would have been great if everyone ignored transformers? Or used them in a different way? "Achieving AGI" was never the goal, though it became a marketing slogan. The only AGI definition I saw was one that Gary and some friends produced that was not adequate or followed up on, as far as I know -- it focused on cognitive intelligence, whereas social intelligence is both more difficult and more important. Almost everyone uses emotions to channel our limited cognitive capabilities or avoid consequences for their failures. The goal was to make money and accumulate power. A wall there was less clear than for undefined AGI. It still isn't crystal clear. There may be places where it will make money that needed LLMs to discover, such as coding, better video games or porn, etc. For now the money to be made is selling tools to the prospectors -- platform use, cloud computing. Microsoft Copilot is not doing that well, but Microsoft is.
Achieving AGI mos def WAS and still IS the goal of the major players currently competing in the field, who keep changing their definition of AGI as their overhyped predictions fail to materialize.
Have you read Karen Hao's book "Empire of AI"? It would answer the majority of your questions. My point is that focusing almost *exclusively* on LLMs to achieve AGI has sucked money, attention, and resources from research that could be less expensive and focused on more limited and specific goals that do not require hoovering up all the data on the internet with the accompanying need for huge data centers, more power and water, and an obscene amount of money.
I think you missed quite a few of Gary's articles judging from your summary above. LLMs detect patterns very well (which is why they are good at protein folds), but suck at math, novelty, world maps, symbolic thinking, reasoning, and maps/charts/diagrams. In no way will a computer ever become socially intelligent because it cannot FEEL either human emotions or physical sensations without having a human body.
Yes, that's just one reason they will never "align" with human concerns. Our nerves are not all located in the brain. According to Sleepy Science, 80% of the vagus nerve—the main nerve of our parasympathetic nervous system—carries signals *from* our heart and digestive system *to* the brain, not the other way around. And all the signals sent from those organs profoundly affect our human perceptions, needs, and feelings.
Oaktown--I read an article on that recently (the vagus nerve)--but it doesn't really take much thought these days to step away from what goes on with AI and towards the massive differences between that and human beings.
It's just that the ones who are promoting that idea are speaking from their own specialty and LACK of even a generalist's knowledge of humans and history, and not from a horizon that would include knowledge that has even reached the general public from the fields of the human sciences, history, the humanities, philosophy, and even theology. Shallow cannot think beyond its own shallowness because it defines **everything** in the world through the lenses of its own shallow viewpoint.
They will not find the answers to their questions (and remedy for their problems) by looking in their bank accounts and wallets.
Public opinion won't matter once they're cocooned in military bases, as the President recently declared, for national security purposes, of course. Public opinion only matters, and is used to justify, when it supports what the public/private partnership blob wants.
There is an argument that what people are actually objecting to is tech-bros. Since the tech-bros have proven themselves to be either liars, nazis or psychopaths - or most commonly all three - and since their rapacious fortunes rely on more data center constructon, it follows that nothing said in favour of the latter can be trusted.
Those 65 and older had a sizeable shift. Support was cut in half (33 to 17), and opposition up by about 40% (41% increased to 57%). Source: topline report, page 5.
How do you see this evolving as AI becomes more ubiquitous in the standard American’s life over the next 12-18 months and how will politicians navigate the fear (rational or not) of their constituents vs. the existential national security threat of not maintaining the pace of the frontier ahead of other nations?
Are current genAI systems really that critical for national security? Are they more critical that energy and water supply security? Are all the risks caused by global warming worth the benefits delivered by AI … and no, I don’t think LLMs are going to “solve” global warming or cure cancer. There is absolutely no reason to think they can.
So we already know that local datacenters either drive up electric rates or pollute the air using their own generators. They use up local water, depriving locals of their water and driving up water rates. Noise is also a factor.
Internationally, countries are still ceding their AI sovereignty. Look at the fuss being raised over Palantir accessing medical and military data. It is not unlike who really has control of nuclear weapons when the US uses a foreign land as a base.
There is definitely a FOMO when AI companies want to build datacenters in other nations, as well as local cash-strapped city and county governments in the US.
The local costs you name are real. Rates, water, backup generators, noise. I wasn’t arguing those fears are irrational.
I also hear the sovereignty point differently than “fear of the plant next door.” Your nuclear-basing / Palantir frame is about ceding power to the AI buildout: data, medical, military, host-nation dependence. That’s a real deferred cost too.
My original push was narrower. “The one issue unifying the public” starts to unravel once you look at the basis underneath the shared unease. Some are defending the town. Some are defending who sets the frontier. Same headline word, different root.
On the energy side, you can argue that blocking nuclear in the 70s and 80s didn’t erase demand. It moved the bill onto fossil plants, fracking, and hydro, with the climate and water costs that came with that path. The opposition had reasons. The deferred cost was still real.
So I don’t think the question is whether there’s an issue. There is. It’s which bill we’re less willing to pay: the local and sovereignty costs of building, or the cost of not remaining the pacemaker and letting other nations set the outcomes, good and bad.
The following is extracted from the latest post by Ed Zitron. It indicates that the latest deal is only an MOU, and may not even happen based on a prior, similar announcement. Zitron's wider point that he has been banging on for years is that these hyperscalers are not profitable now, nor can they ever be. IOW, these huge datacenters are an economic drag with negative ROI, and apart from the immediate damage, are pushing out other, likely more profitable investments. AI might provide some benefits, but with a cost that outweighs the realizable value. This is not unlike the chip industry that Warren Buffett famously said he wouldn't invest in because they never seemed to generate free cash flow for returns. It was all invested in the next generation of fabs.
--------------------------------------From Zitron post today-------------------------
Yet when you actually open the press release (which I found on my Terminal but cannot for the life of me get a link to), there’s one glaring detail everybody left out, emphasis mine:
>Memorandums of understanding signed with six of the world’s premier financial institutions to create these partnerships aim to establish the first compute financing platforms of their kind at global scale to enable the AI infrastructure buildout across NVIDIA’s ecosystem, including leading frontier AI labs, enterprises and AI clouds. Under these strategic partnerships, NVIDIA will work with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to create dedicated pools of capital at significant scale at attractive rates for NVIDIA customers.<
That’s right folks. There is no deal! It’s an MOU! It’s fucking theoretical! And I have not seen this fact reflected in a single god damn story about this god damn “deal” in one god damn place! It’s from the press release from the fucking companies!
I think you are implying that AI will continue to require these huge datacenters to support a cloud-based architecture. But is this true?
Firstly, once training is over, distillation and pruning the size of the model is possible. The Chinese have shown that it works.
Secondly, there is a demand for smaller, domain-specific models that can work on local/edge hardware, avoiding the need to pay for access to the hyperscalers.
Thirdly, there is a new thrust to get useful, neuromorphic chips to handle both neural and symbolic code. These chips will vastly decrease the computing resources needed. They may obsolete the need for these hyperscaled facilities.
Think of the changes that electrification brought to factories needing motive power. The single steam engine produced all teh power, and this was distributed by beltes across teh facory floor. When electric motors arrived, a large motor just replaced teh steam engine. But as small motors could be built, it waas realized that the better solution was to distribute electric power around teh factories and drive small motors at each point. These motors then became integrated into the machines and tools, the way we have had it for the last century.
I see the same logic for AI. There will still be a need for some large facilities for training and inference of "frontier" models. But most will run locally on much smaller hardware. Can the datacenters be repurposed? Maybe. However, I suspect they will become obsolete white elephants within a decade or two. Just shutting them down will relieve the pressure on physical resources.
So the costs that you mention might be obviated by waiting a while for better technology. It may be like the transition from steam to electricity, or the transition from railroads to tarmac roads with buses and cars and trucks. In teh 1960s, the railroad building frenzy in England was dismantled due to their poor economics, and the routes were drastically cut in favor of different, more flexible, transport modes.
BTW, I note O'Reilly's Radar for August notes the push to smaller AI models, which supports my POV that AI will move to smaller models that run locally.
Lastly, I note that as yet, there is no proof that AI is useful beyond a few domains. Does healthcare delivery work better with big AI models? There is evidence that doctors at teh UK NHS hospitals prefer their own systems to Palantir's. Has AI improved the effectiveness of the military? Yet to be demonstrated. Even writing code, there is no evidence that this increases the speed of the development cycle when tested.
Maybe there is no lost opportunity cost from not bowing to FOMO and pressure to be at the forefront of AI, the new "Abundance". etc., etc.
Your first question is right, and I don’t know. I was treating “stay at the frontier” as though it meant hyperscale ubiquity forever. It may not.
The AI stack is clearly broadening: smaller models, local inference, routing, compression, and domain specialists. O’Reilly’s August Radar shows that. It also describes trillion-parameter releases. This is bifurcation, not simple replacement.
Frontier compute still looks concentrated upstream for now. Distillation needs a teacher, though not every local model is distilled, and pruning or quantization do not require a new one. Small models can be trained directly, but “small enough to run locally” does not mean cheap to create. Cheaper inference may create more demand through agents and always-on work, but that is a pattern, not a law. Some campuses will matter. Others will become white elephants.
On your electricity analogy, I think it is a useful rhyme, not proof. Edge models may be motors in the tools; hyperscalers may be part of the upstream layer. But electricity must flow continuously, while trained weights can be copied. The analogy supports patience about form factor as much as urgency about access.
I have more intellectual interest here than a position to defend. The useful question is not simply build or wait: how much compute, owned by whom, built where, at whose expense, and compared with which flexible alternatives? A country, a town, and a firm will answer differently. That is a portfolio decision under uncertainty, not a contest over who predicts the final architecture.
Alex and James: It's further disheartening to hear (this week, two people, one was Blanche the other, Trump's banker Cheshire cat) that (paraphrased) there's always going to be whiners and people who reject on principle what the people in power say and want. "So what else is new?"
If that's the standard thought these days, then very powerful people have lost any connection they might have had with legitimate dialogue--a situation which is so evident in Trump's MO: no matter what the "democrats" say, it's just more noise. Any truth that others may have is dead and buried under such a view.
To turn this around: "there's always going to be whiners" Who in this case sit in their offices in Sillicon Valley and asking themselves why people really hate them personally.
Patrick Boyle had a nice abstraction on how Altmann, Amodei and the rest of these people think: Every problem is a software problem which can be overcome by patching the bug. So they are whining about why their patching does not seem to work, and is making the "problem" worse.
"However, I suspect they will become obsolete white elephants within a decade or two. Just shutting them down will relieve the pressure on physical resources." I'm not sure how you are seeing this, given the land is now locked up with huge, unused buildings, often covering farmland, and possibly land suitable for housing. Who is going to have to pay to dismantle them? I am sure there are those who could imagine repurposing them along the lines of detention centers, given the utilities are already connected, and who would want to let good potential penal colonies go to waste?
My mental image is of the abandoned, rotting factories in the Rust Belt". But as you say, these datacenters are huge facilities. One proposal was for a facility that covered as much area as Manhattan! (Presumably in separate buildings.) I would hope we have ended detention centers by then, but you aren't wrong, given the proposals to convert unused warehouses.
In California, it is very hard to convert and rezone agricultural land for factories and housing, so I expect that datacenter construction will have to be on low-value land. I have read that elsewhere, this often means on land with little water, further worsening the local ecosystem and for anyone unfortunate to live there.
I think that is the local version of the question I was trying to get at. A project can be strategically useful somewhere and still be a bad local deal if the upside is private while the grid, water, and exit costs are public. The test is not only whether it gets built, but whether the approval names who pays to remediate or repurpose it if the demand forecast is wrong. Cheers!
“AI has been mostly awFUL and the AI companies have engaged in mostly bad behavior”
Fixed it for Bindu Reddy
Engaged in mostly bad behavior whilst telling themselves and everyone else that they are saving humanity 🙈
On a silver platter…for dinner
"To Serve Man," as told by Rod Sterling.
Let's call it SAIvation.
Good news from your friends at Skynet: we're building a toxic-waste dump and playground next to the data center in your community!
Inadvertently the opposition to AI and AI data centers may save the tech industry. It will stop them from spending insane amounts on AI, when there is no prospect for any substantial revenue.
Good luck convincing them of that; they live in self-deluded, juvenile SciFi fantasies.
I don't expect anyone to convince them, but the opposition to data centers may save them, despite themselves.
I see it the way you do, just don't think for a minute they'll ever agree with us. They seem to see themselves as victims whenever they encounter any obstacle to their goal of total domination over the planet.
Oaktown: Yes, and "the people" who just want to live their lives in their homes and communities without "interruption" from the Arrogant Gaslighting Idiots (AGI) among us, are probably thought of as just so many backward neanderthals.
That's perfect: AGI = Arrogant Gaslighting Idiots
All in this thread--see my post above (or somewhere here) where the quotations are from FORTUNE 500 Digest today--showing several of the BIG companies already investing in data centers and "infrastructure." Of course, they COULD be finding ways to self-energize and actually help the people around those centers (shall we laugh now or just wait till later?)
But it seems to me, and I speculate here, that the power brokers are just going ahead with "WHAT I WANT" and to hell with the "local" protest. Let them scream. And apparently the U. S. Government, such as it is, is not on the side of "the people."
The problem, as I see it, the "power brokers" assume that the stuff they are building - AI and data centers - will make them money. They do not do now, and likely they won't in the future.
So, if locals stop them, they will save the "power brokers" money.
Richard Bielak: I saw all that big money/corporations as listed in that Fortune 500 Digest investing in data centers as a clue about the influence of local protests: None.
Just let a few billionaires rule and surprisingly they just do what they want.
Franz-Xaver Huber: What's that saying? "Power corrupts. Absolute Power Corrupts Absolutely"? And there is this REALLY STUPID IDEA that the more money and power you have, the more intelligent and even wise you are. Even when some of these "entrepreneurs" have a streak of what is obviously DUMB LUCK, they tend to think it's "something about me."
They will always say they are talented, intelligent and working hard to justify their wealth and income/profits. That's part of the narrative. They don't want to share anything.
I just read that the Kushner's are in a deal to purchase the L.A. Lakers for an obscene amount of money.
It's only a matter of time before these data centers start mysteriously burning down around the nation.
AI is the latest revolution in capitalist production, renewing (they hope) an economy of overproduction. It follows on from previous revolutions; computers, communications, aircraft, the automobile, the train and so on, until we run out of planet.
There is always Mars.
If the Martians are smart, they are already planning a defense.
Although with Elon’s Starship, which can’t even get to low earth orbit, that probably won’t be necessary
Mars is a hellhole infinitely more hellish than the hellishest hellholes on earth we havent managed to successfully colonise.
So good luck with that.
If humans can’t even get along in a bountiful paradise for which they are uniquely adapted by aeons of evolution, there is no way in Hell they will survive in any significant numbers on Mars.
If they can’t get along in the Garden of Eden, they won’t last more than a brief time in a place like Mars where everything needed for survival is scarce.
“Tribes” will be raiding and fighting one another (to the death) for those resources pretty much from day one.
It is said that Von Neumann, Teller, Ulam et al were from Mars.
If that’s the case, humans won’t stand a chance
Him and H.G. Wells
Maybe add Orson, as well.
Good one.
The bipartisan angle is the real signal here -- polarized issues get litigated in courts and campaign ads for years without changing much, but bipartisan opposition moves zoning boards and utility commissions fast, and that's where data center buildout actually lives or dies. Worth watching state-level power-purchase-agreement fights over the next two quarters, not the 2028 election -- local utility rate cases will bite AI capex projections years before any anti-AI candidate does.
Good. Fuck 'em. Pretty soon the media will admit that it's not really data centers people are against, it's AI, but the media has a hype interest in hiding that obvious fact.
Not just AI, but also the data centers, which are mammoth sized consuming mammoth sized amounts of water as well as mammoth sized amounts of (non) renewable energy. All of that against a background of a 24/7 60 decibel hum.
There is of course also the issue of water and energy affordablity for the locals near te mammoth data center.
The political effect may be felt well before the 2028 presidential election. The WA state legislature meets each year between January and April with a couple thousand bills to consider. Until this year that included maybe 1 or 2 AI bills and 0 or 1 data center bills. In 2026, there were 8 AI bills and 8 data center bills. For the first time, I noticed someone representing a major tech company testify in a public hearing, rather than an industry lobbyist. (Tech bills generally have 4 or more public hearings in committees before passing, all webcast live and recorded.) In 2026, most bills did not pass and others were watered down with amendments. 2027 could be different, and 2028 legislation will also be long before the election. My only question about this post is "AI could have been a great thing." What would that great thing have been and how could it have happened in the real world?
I've maintained for a while now that I expect the midterms this year to be chaotic, so we could start reaping what we've sown in November
I'm aware. It's gonna be a rough ride ahead. Our president is terrified and losing control over his phony narrative, which renders him desperate and extremely dangerous.
For starters, tech fascists and clueless investors could have listened to Gary years ago, when he demonstrated with knowledge and logic that LLMs would hit a wall and never achieve AGI. Then research money and resources could have been more wisely spent on finding a way to develop specific and *reliable* AI tools for limited tasks, instead of trying to sell them to the public as human beings spewing forth authoritative BS.
Wasn't this already happening? Research money and resources being spent on protein-folding and other reliable AI tools for limited tasks? Are you saying AI would have been great if everyone ignored transformers? Or used them in a different way? "Achieving AGI" was never the goal, though it became a marketing slogan. The only AGI definition I saw was one that Gary and some friends produced that was not adequate or followed up on, as far as I know -- it focused on cognitive intelligence, whereas social intelligence is both more difficult and more important. Almost everyone uses emotions to channel our limited cognitive capabilities or avoid consequences for their failures. The goal was to make money and accumulate power. A wall there was less clear than for undefined AGI. It still isn't crystal clear. There may be places where it will make money that needed LLMs to discover, such as coding, better video games or porn, etc. For now the money to be made is selling tools to the prospectors -- platform use, cloud computing. Microsoft Copilot is not doing that well, but Microsoft is.
Achieving AGI mos def WAS and still IS the goal of the major players currently competing in the field, who keep changing their definition of AGI as their overhyped predictions fail to materialize.
Have you read Karen Hao's book "Empire of AI"? It would answer the majority of your questions. My point is that focusing almost *exclusively* on LLMs to achieve AGI has sucked money, attention, and resources from research that could be less expensive and focused on more limited and specific goals that do not require hoovering up all the data on the internet with the accompanying need for huge data centers, more power and water, and an obscene amount of money.
I think you missed quite a few of Gary's articles judging from your summary above. LLMs detect patterns very well (which is why they are good at protein folds), but suck at math, novelty, world maps, symbolic thinking, reasoning, and maps/charts/diagrams. In no way will a computer ever become socially intelligent because it cannot FEEL either human emotions or physical sensations without having a human body.
Oaktown--also, AI won't die, at least not like humans?
Yes, that's just one reason they will never "align" with human concerns. Our nerves are not all located in the brain. According to Sleepy Science, 80% of the vagus nerve—the main nerve of our parasympathetic nervous system—carries signals *from* our heart and digestive system *to* the brain, not the other way around. And all the signals sent from those organs profoundly affect our human perceptions, needs, and feelings.
Oaktown--I read an article on that recently (the vagus nerve)--but it doesn't really take much thought these days to step away from what goes on with AI and towards the massive differences between that and human beings.
It's just that the ones who are promoting that idea are speaking from their own specialty and LACK of even a generalist's knowledge of humans and history, and not from a horizon that would include knowledge that has even reached the general public from the fields of the human sciences, history, the humanities, philosophy, and even theology. Shallow cannot think beyond its own shallowness because it defines **everything** in the world through the lenses of its own shallow viewpoint.
They will not find the answers to their questions (and remedy for their problems) by looking in their bank accounts and wallets.
Google has disbanded its protein folding team. Arrrgh.
Sheesh. What reason did they give for doing that?
You can find the story in Scientific American and on
https://tech.yahoo.com/science/articles/why-google-deepmind-broke-alphafold-130000502.html
If you are interested here is the open source site on protein folding :
https://openfold.io/index.html
Thank you!
Hi Gary? For a good reason - people are starting to see these as grave factories - that hypermanufacture caskets - including theirs.
Public opinion won't matter once they're cocooned in military bases, as the President recently declared, for national security purposes, of course. Public opinion only matters, and is used to justify, when it supports what the public/private partnership blob wants.
AI is an environmental disaster threatening our most finite resource WATER
There is an argument that what people are actually objecting to is tech-bros. Since the tech-bros have proven themselves to be either liars, nazis or psychopaths - or most commonly all three - and since their rapacious fortunes rely on more data center constructon, it follows that nothing said in favour of the latter can be trusted.
And it us the same AI slop that keeps appearing again and again and again.
Those 65 and older had a sizeable shift. Support was cut in half (33 to 17), and opposition up by about 40% (41% increased to 57%). Source: topline report, page 5.
https://www.annenbergpublicpolicycenter.org/wp-content/uploads/AI_W38_Topline.pdf
How do you see this evolving as AI becomes more ubiquitous in the standard American’s life over the next 12-18 months and how will politicians navigate the fear (rational or not) of their constituents vs. the existential national security threat of not maintaining the pace of the frontier ahead of other nations?
Are current genAI systems really that critical for national security? Are they more critical that energy and water supply security? Are all the risks caused by global warming worth the benefits delivered by AI … and no, I don’t think LLMs are going to “solve” global warming or cure cancer. There is absolutely no reason to think they can.
Easy one: No, No and No. :-)
So we already know that local datacenters either drive up electric rates or pollute the air using their own generators. They use up local water, depriving locals of their water and driving up water rates. Noise is also a factor.
Internationally, countries are still ceding their AI sovereignty. Look at the fuss being raised over Palantir accessing medical and military data. It is not unlike who really has control of nuclear weapons when the US uses a foreign land as a base.
There is definitely a FOMO when AI companies want to build datacenters in other nations, as well as local cash-strapped city and county governments in the US.
The local costs you name are real. Rates, water, backup generators, noise. I wasn’t arguing those fears are irrational.
I also hear the sovereignty point differently than “fear of the plant next door.” Your nuclear-basing / Palantir frame is about ceding power to the AI buildout: data, medical, military, host-nation dependence. That’s a real deferred cost too.
My original push was narrower. “The one issue unifying the public” starts to unravel once you look at the basis underneath the shared unease. Some are defending the town. Some are defending who sets the frontier. Same headline word, different root.
On the energy side, you can argue that blocking nuclear in the 70s and 80s didn’t erase demand. It moved the bill onto fossil plants, fracking, and hydro, with the climate and water costs that came with that path. The opposition had reasons. The deferred cost was still real.
So I don’t think the question is whether there’s an issue. There is. It’s which bill we’re less willing to pay: the local and sovereignty costs of building, or the cost of not remaining the pacemaker and letting other nations set the outcomes, good and bad.
The following is extracted from the latest post by Ed Zitron. It indicates that the latest deal is only an MOU, and may not even happen based on a prior, similar announcement. Zitron's wider point that he has been banging on for years is that these hyperscalers are not profitable now, nor can they ever be. IOW, these huge datacenters are an economic drag with negative ROI, and apart from the immediate damage, are pushing out other, likely more profitable investments. AI might provide some benefits, but with a cost that outweighs the realizable value. This is not unlike the chip industry that Warren Buffett famously said he wouldn't invest in because they never seemed to generate free cash flow for returns. It was all invested in the next generation of fabs.
--------------------------------------From Zitron post today-------------------------
Yet when you actually open the press release (which I found on my Terminal but cannot for the life of me get a link to), there’s one glaring detail everybody left out, emphasis mine:
>Memorandums of understanding signed with six of the world’s premier financial institutions to create these partnerships aim to establish the first compute financing platforms of their kind at global scale to enable the AI infrastructure buildout across NVIDIA’s ecosystem, including leading frontier AI labs, enterprises and AI clouds. Under these strategic partnerships, NVIDIA will work with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to create dedicated pools of capital at significant scale at attractive rates for NVIDIA customers.<
That’s right folks. There is no deal! It’s an MOU! It’s fucking theoretical! And I have not seen this fact reflected in a single god damn story about this god damn “deal” in one god damn place! It’s from the press release from the fucking companies!
source: https://www.wheresyoured.at/dont-look-up/?ref=ed-zitrons-wheres-your-ed-at-newsletter
I think you are implying that AI will continue to require these huge datacenters to support a cloud-based architecture. But is this true?
Firstly, once training is over, distillation and pruning the size of the model is possible. The Chinese have shown that it works.
Secondly, there is a demand for smaller, domain-specific models that can work on local/edge hardware, avoiding the need to pay for access to the hyperscalers.
Thirdly, there is a new thrust to get useful, neuromorphic chips to handle both neural and symbolic code. These chips will vastly decrease the computing resources needed. They may obsolete the need for these hyperscaled facilities.
Think of the changes that electrification brought to factories needing motive power. The single steam engine produced all teh power, and this was distributed by beltes across teh facory floor. When electric motors arrived, a large motor just replaced teh steam engine. But as small motors could be built, it waas realized that the better solution was to distribute electric power around teh factories and drive small motors at each point. These motors then became integrated into the machines and tools, the way we have had it for the last century.
I see the same logic for AI. There will still be a need for some large facilities for training and inference of "frontier" models. But most will run locally on much smaller hardware. Can the datacenters be repurposed? Maybe. However, I suspect they will become obsolete white elephants within a decade or two. Just shutting them down will relieve the pressure on physical resources.
So the costs that you mention might be obviated by waiting a while for better technology. It may be like the transition from steam to electricity, or the transition from railroads to tarmac roads with buses and cars and trucks. In teh 1960s, the railroad building frenzy in England was dismantled due to their poor economics, and the routes were drastically cut in favor of different, more flexible, transport modes.
BTW, I note O'Reilly's Radar for August notes the push to smaller AI models, which supports my POV that AI will move to smaller models that run locally.
Lastly, I note that as yet, there is no proof that AI is useful beyond a few domains. Does healthcare delivery work better with big AI models? There is evidence that doctors at teh UK NHS hospitals prefer their own systems to Palantir's. Has AI improved the effectiveness of the military? Yet to be demonstrated. Even writing code, there is no evidence that this increases the speed of the development cycle when tested.
Maybe there is no lost opportunity cost from not bowing to FOMO and pressure to be at the forefront of AI, the new "Abundance". etc., etc.
Your first question is right, and I don’t know. I was treating “stay at the frontier” as though it meant hyperscale ubiquity forever. It may not.
The AI stack is clearly broadening: smaller models, local inference, routing, compression, and domain specialists. O’Reilly’s August Radar shows that. It also describes trillion-parameter releases. This is bifurcation, not simple replacement.
Frontier compute still looks concentrated upstream for now. Distillation needs a teacher, though not every local model is distilled, and pruning or quantization do not require a new one. Small models can be trained directly, but “small enough to run locally” does not mean cheap to create. Cheaper inference may create more demand through agents and always-on work, but that is a pattern, not a law. Some campuses will matter. Others will become white elephants.
On your electricity analogy, I think it is a useful rhyme, not proof. Edge models may be motors in the tools; hyperscalers may be part of the upstream layer. But electricity must flow continuously, while trained weights can be copied. The analogy supports patience about form factor as much as urgency about access.
I have more intellectual interest here than a position to defend. The useful question is not simply build or wait: how much compute, owned by whom, built where, at whose expense, and compared with which flexible alternatives? A country, a town, and a firm will answer differently. That is a portfolio decision under uncertainty, not a contest over who predicts the final architecture.
Alex and James: It's further disheartening to hear (this week, two people, one was Blanche the other, Trump's banker Cheshire cat) that (paraphrased) there's always going to be whiners and people who reject on principle what the people in power say and want. "So what else is new?"
If that's the standard thought these days, then very powerful people have lost any connection they might have had with legitimate dialogue--a situation which is so evident in Trump's MO: no matter what the "democrats" say, it's just more noise. Any truth that others may have is dead and buried under such a view.
To turn this around: "there's always going to be whiners" Who in this case sit in their offices in Sillicon Valley and asking themselves why people really hate them personally.
Patrick Boyle had a nice abstraction on how Altmann, Amodei and the rest of these people think: Every problem is a software problem which can be overcome by patching the bug. So they are whining about why their patching does not seem to work, and is making the "problem" worse.
"However, I suspect they will become obsolete white elephants within a decade or two. Just shutting them down will relieve the pressure on physical resources." I'm not sure how you are seeing this, given the land is now locked up with huge, unused buildings, often covering farmland, and possibly land suitable for housing. Who is going to have to pay to dismantle them? I am sure there are those who could imagine repurposing them along the lines of detention centers, given the utilities are already connected, and who would want to let good potential penal colonies go to waste?
My mental image is of the abandoned, rotting factories in the Rust Belt". But as you say, these datacenters are huge facilities. One proposal was for a facility that covered as much area as Manhattan! (Presumably in separate buildings.) I would hope we have ended detention centers by then, but you aren't wrong, given the proposals to convert unused warehouses.
In California, it is very hard to convert and rezone agricultural land for factories and housing, so I expect that datacenter construction will have to be on low-value land. I have read that elsewhere, this often means on land with little water, further worsening the local ecosystem and for anyone unfortunate to live there.
I think that is the local version of the question I was trying to get at. A project can be strategically useful somewhere and still be a bad local deal if the upside is private while the grid, water, and exit costs are public. The test is not only whether it gets built, but whether the approval names who pays to remediate or repurpose it if the demand forecast is wrong. Cheers!