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!"
"Discussions about expanding electricity supply to power the future often become debates about which source is most suitable: gas, nuclear, solar, or something else. But these are a distraction. Far more fundamental is ensuring power can be efficiently delivered where needed."
This is the reason why data centers are not run only on cheap solar power.
In the Innovation Adoption Curve, we are absolutely beyond the Early Adopters phase and possibly the Early Majority. The growth rates necessarily have to start trending down because there’s no one left to sell to.
I don't think Anthropic are doing that because they don't have enough compute capacity. If we had 100* more datacentres the message would still be the same - they're focusing on selling Fable access to enterprise users because that's what makes them more money.
Either the cost to serve this larger model is so high that they cannot offer any reasonable usage quota for it at subscription prices, or they really do not have the capacity.
I think that it might well be true. The Opus models had capacity issues on many occasions too. Can the larger model even be served on all of the hardware they have, or only a subset?
It would not surprise me if growing enterprise demand threatens capacity, making it impossible for Anthropic to offer the model in subscriptions at this time, even though they could do so at a profit.
We do not know whether subscription plans are unprofitable at all.
Some estimates suggest that this is the case only for the heaviest users.
Many seem to confuse API prices with the actual cost to serve the models, and thus reach the conclusion that subscriptions must be deeply unprofitable.
Anthropic is officially citing capacity constraints with the intent to bring the Fable model back to subscriptions plans as soon as capacity allows.
> We do not know whether subscription plans are unprofitable at all.
That's pretty much certain. It's sort of cute when people like to pretend otherwise.
> Many seem to confuse API prices with the actual cost to serve the models, and thus reach the conclusion that subscriptions must be deeply unprofitable.
I don't make that mistake. I actually suspect that the actual costs may be higher than the API prices. I think those may still be subsidized.
> Anthropic is officially citing capacity constraints with the intent to bring the Fable model back to subscriptions plans as soon as capacity allows.
Yeah, I don't think they are being truthful at all.
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?
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.
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.
You don’t even need to mention the long-term sustainability benefits of renewable energy. It is simply the dominant option economically. Dollars in, watts produced, fossil energy can’t touch it.
Politics is merely a downstream effect of the root of the problem: corruption and regulatory capture. Regression into the authoritarian petrostate pattern.
Who is opposing renewable energy? Texas is projected to top California as the state with the largest amount of solar before the end of the Trump administration.
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.
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?
Unit economics for renewables coupled with storage are excellent. I agree we should reform nuclear regulation to allow new nuclear plants to pencil out. I disagree that we should discount the value of renewables.
Solar and wind take up a bunch of space and generate a bunch of waste after the panels are decommissioned, plus the wind generators are ruining every single landscape we have. With 5 nuclear stations per country you could cleanup so much of Europe.
The environmental damage caused by "clean" power sources is done mostly in countries which are far from Europe, so it's not much discussed in Europe.
Like:
Copper. "As the world shifts to wind energy and electric cars, demand for the conductive metal has increased. But mining copper brings its own environmental hazards"
It's been regulated "to death" because it's responsible for some of the worst man-made catastrophes of all time and has made large swaths of land uninhabitable for ~forever.
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.
I'd like to see a well reasoned plan to install small modular nuclear power at peoples houses that prevents the mentally ill, criminally reckless or terrorist minded people from cracking them open and obtaining access to the isotopes.
From a strictly red team threat analysis perspective, if you have an extremely safe working reactor that can't be made to melt down, no, the reactor can't be used to hurt much that is in the same location as the reactor. If you are able to get the isotopes out and start spreading them around or making a dirty bomb type thing (where the explosion just serves the purpose of throwing the isotopes around), that could be pretty catastrophic.
have you never visited a rural neighborhood? or an affluent neighborhood?
there are still many neighborhoods where people leave their doors unlocked because it is actually that safe. not every location is rife with criminal activity, and many are well protected.
Serious question, do you actually think that if you distribute millions of small nuclear reactors to homeowners geographically spread out around all of North America, 0.000% of them will be dangerously mentally ill, criminally reckless or inclined to terrorist like activity? Based on the frequency and number of mass shooter type incidents, (or like, David Koresh and the Branch Davidians) this would be a very naive view.
It doesn't require a criminally minded 3rd party coming onto someone's "safe" property to do something horrible with a sawzall and/or oxy-acetylene cutting torch.
That's a very optimistic view of humanity that I can't say I share. If you give a sufficiently motivated person enough isolation and time, they can cut into just about anything. And possibly deal with cleaning up the results of any internal tamperproof countermeasures. In a world that contains people like the Las Vegas mass shooter or those who conducted the 2015 attacks in Paris, handing out isotopes to the ordinary person seems like a recipe for disaster.
We live in a world where multiple people are killed every year by tipping vending machines over onto themselves and you propose to make nuclear reactors a mass market consumer good that goes in everyone's garage?
> The worst thing an evildoer can do is blow up your own house and the few around it.
No, they can take the isotopes out and dump it into your local water supply. Or if they're suicidal and the isotopes have been encapsulated in some sort of tamperproof system, grid the whole thing down to granulated powder using less than $20,000 of power tools (disregarding their own health and the entire nearby area, of course) and then dump it into the local water supply.
If someone evil has access to the water supply, is radioactive material the worst they could do? that'll, what, give some people cancer which is really bad but it'll take a while to get them. If people wanted to be shitheads, they could already dump arsenic or LSD into the water supply, or any number of others things. that are already available to them right now! Have you personally tested your taps chlorine or flouride or lead levels recently?
You're seriously proposing that a country with regular mass-shootings should give everyone a device that can cause a radioactive meltdown or a small explosion?
We already let people have cars and those things are crazy dangerous! No one should be allowed anything, ever, until we bubble wrap everything in the world to be perfectly safe!
Well, we already have strict background checks, licensing and regulations for a scenario such as if a person wants start a home based business manufacturing and storing C4, Semtex or similar at their rural property. If the idea is to start handing out nuclear reactors for peoples' houses, the possible damage that could be done is far greater. No matter how well packaged it is or designed to be consumer friendly.
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!"
The most important sentences of the article are:
"Discussions about expanding electricity supply to power the future often become debates about which source is most suitable: gas, nuclear, solar, or something else. But these are a distraction. Far more fundamental is ensuring power can be efficiently delivered where needed."
This is the reason why data centers are not run only on cheap solar power.
In the Innovation Adoption Curve, we are absolutely beyond the Early Adopters phase and possibly the Early Majority. The growth rates necessarily have to start trending down because there’s no one left to sell to.
I cannot currently access Fable class models, I am out of usage credits.
Anthropic is removing these larger models from personal plans at the end of the week to focus on selling it to enterprise users.
Putting these more intelligent new models into the hands of more people seems very worthwhile to me.
I don't think Anthropic are doing that because they don't have enough compute capacity. If we had 100* more datacentres the message would still be the same - they're focusing on selling Fable access to enterprise users because that's what makes them more money.
Either the cost to serve this larger model is so high that they cannot offer any reasonable usage quota for it at subscription prices, or they really do not have the capacity.
I think that it might well be true. The Opus models had capacity issues on many occasions too. Can the larger model even be served on all of the hardware they have, or only a subset?
It would not surprise me if growing enterprise demand threatens capacity, making it impossible for Anthropic to offer the model in subscriptions at this time, even though they could do so at a profit.
Do you pay API prices? Or are you on some of those plans that are deeply unprofitable?
I am not even sure if API prices are actually profitable, but they certainly aren't as unprofitable as the subscription plan users.
Either way, that's why you don't have access. It has nothing to do with capacity constraints.
We do not know whether subscription plans are unprofitable at all.
Some estimates suggest that this is the case only for the heaviest users.
Many seem to confuse API prices with the actual cost to serve the models, and thus reach the conclusion that subscriptions must be deeply unprofitable.
Anthropic is officially citing capacity constraints with the intent to bring the Fable model back to subscriptions plans as soon as capacity allows.
> We do not know whether subscription plans are unprofitable at all.
That's pretty much certain. It's sort of cute when people like to pretend otherwise.
> Many seem to confuse API prices with the actual cost to serve the models, and thus reach the conclusion that subscriptions must be deeply unprofitable.
I don't make that mistake. I actually suspect that the actual costs may be higher than the API prices. I think those may still be subsidized.
> Anthropic is officially citing capacity constraints with the intent to bring the Fable model back to subscriptions plans as soon as capacity allows.
Yeah, I don't think they are being truthful at all.
1. People fighting back due to never ending effects caused by these massive data centers
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.
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?
I think it’s more likely we’ll raise prices until demand lowers enough to match supply.
Open models enforce a floor on prices, unless overall compute is so constrained that those prices rise also.
Project Kilby is probably the most intelligent approach to this problem so far.
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.
The ocean tidal power data center idea was a neat one.
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.
Crusoe bypasses the grid completely by leveraging old batteries and wasted resources to power compute.
‘the abundance of [AI] will be limited by the abundance of energy’.
And the reason current US policy opposes clean, renewable energy is --- purely political.
You don’t even need to mention the long-term sustainability benefits of renewable energy. It is simply the dominant option economically. Dollars in, watts produced, fossil energy can’t touch it.
Politics is merely a downstream effect of the root of the problem: corruption and regulatory capture. Regression into the authoritarian petrostate pattern.
> And the reason current US policy opposes clean, renewable energy is --- purely political.
What isn't "purely political?"
Everything is political in the world, unfortunately.
Who is opposing renewable energy? Texas is projected to top California as the state with the largest amount of solar before the end of the Trump administration.
> Who is opposing renewable energy?
The current US admin:
Trump’s Multi-Pronged Attack on Renewable Energy https://scholarship.law.columbia.edu/cgi/viewcontent.cgi?par...
but they are slowly being beaten back:
Trump admin abandons fight against wind energy as clean energy output surges - https://arstechnica.com/science/2026/06/trump-admin-abandons...
The Trumpian adventures in Iran have wrong footed his fossil fuel buddies and slowed his Don Quixote roll against giant scary windmills for now.
He hates wind turbines because he simply doesn't like how they look, and couldn't stop them next to one of his golf courses.
https://www.bbc.com/news/articles/c15l3knp4xyo
https://blogs.edf.org/climate411/2025/07/31/trump-admin-bloc...
https://www.pewresearch.org/science/2025/06/05/views-on-trum...
Maybe it's the same thing I heard it was six months ago when it was determined that more hardware was sold than twice the nation's supply of electric.
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.
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?
Unit economics for renewables coupled with storage are excellent. I agree we should reform nuclear regulation to allow new nuclear plants to pencil out. I disagree that we should discount the value of renewables.
Solar and wind take up a bunch of space and generate a bunch of waste after the panels are decommissioned, plus the wind generators are ruining every single landscape we have. With 5 nuclear stations per country you could cleanup so much of Europe.
The environmental damage caused by "clean" power sources is done mostly in countries which are far from Europe, so it's not much discussed in Europe.
Like:
Copper. "As the world shifts to wind energy and electric cars, demand for the conductive metal has increased. But mining copper brings its own environmental hazards"
https://www.theguardian.com/us-news/2021/nov/09/copper-minin...
Impacts of lithium mining on water stressed regions in Chile
https://www.theguardian.com/global-development/2026/jan/01/c...
Impacts of rare earth refining in China.
https://www.nytimes.com/2025/07/05/business/china-rare-earth...
silicon tetrachloride from solar production
https://www.azocleantech.com/article.aspx?ArticleID=831
Europe should not outsource it's ecological impact to developing countries.
And the person you replied to probably doesn’t realize spent nuclear fuel has really no where to go other than long term storage.
I don't see how expensive energy helps us "upgrade civilization".
Its only expensive because its been regulated to death. In terms of potential energy output in a given area its tops.
It's been regulated "to death" because it's responsible for some of the worst man-made catastrophes of all time and has made large swaths of land uninhabitable for ~forever.
Move fast and break things this is not.
Unlimited energy for 200 years is a good step, no?
It's hardly unlimited.
And not if plants can't get cooling water because of drought.
https://www.euronews.com/2026/06/25/france-takes-nuclear-rea...
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.
I'd like to see a well reasoned plan to install small modular nuclear power at peoples houses that prevents the mentally ill, criminally reckless or terrorist minded people from cracking them open and obtaining access to the isotopes.
There isn’t much need to extract isotopes if you have an actual working reactor, is there? Just use that as-is to cause whatever damage.
From a strictly red team threat analysis perspective, if you have an extremely safe working reactor that can't be made to melt down, no, the reactor can't be used to hurt much that is in the same location as the reactor. If you are able to get the isotopes out and start spreading them around or making a dirty bomb type thing (where the explosion just serves the purpose of throwing the isotopes around), that could be pretty catastrophic.
have you never visited a rural neighborhood? or an affluent neighborhood?
there are still many neighborhoods where people leave their doors unlocked because it is actually that safe. not every location is rife with criminal activity, and many are well protected.
Serious question, do you actually think that if you distribute millions of small nuclear reactors to homeowners geographically spread out around all of North America, 0.000% of them will be dangerously mentally ill, criminally reckless or inclined to terrorist like activity? Based on the frequency and number of mass shooter type incidents, (or like, David Koresh and the Branch Davidians) this would be a very naive view.
It doesn't require a criminally minded 3rd party coming onto someone's "safe" property to do something horrible with a sawzall and/or oxy-acetylene cutting torch.
have you never heard of tamper-proof containers? mess with it, it’s useless and you go to jail for a long time?
Plus, they’re ubiquitous, you don’t know who has one, max damage is minimal even worst case — go fish!
We have microwave ovens. They’re pretty safe. Imagine something as safe as that. We can do it.
But I can’t disagree that it’s more exciting to imagine terror dreams.
That's a very optimistic view of humanity that I can't say I share. If you give a sufficiently motivated person enough isolation and time, they can cut into just about anything. And possibly deal with cleaning up the results of any internal tamperproof countermeasures. In a world that contains people like the Las Vegas mass shooter or those who conducted the 2015 attacks in Paris, handing out isotopes to the ordinary person seems like a recipe for disaster.
We live in a world where multiple people are killed every year by tipping vending machines over onto themselves and you propose to make nuclear reactors a mass market consumer good that goes in everyone's garage?
While that is true, if there is something worthwhile enough in a far and safe location - it won't stay that way
It’s everywhere. It’s in the shed in your backyard. Nobody knows it’s there. Lots of people have them. It’s an appliance.
The worst thing an evildoer can do is blow up your own house and the few around it. and no one does that because you go to jail forever.
The worst thing you can do is let it melt down, which means it quietly shuts itself down.
> The worst thing an evildoer can do is blow up your own house and the few around it.
No, they can take the isotopes out and dump it into your local water supply. Or if they're suicidal and the isotopes have been encapsulated in some sort of tamperproof system, grid the whole thing down to granulated powder using less than $20,000 of power tools (disregarding their own health and the entire nearby area, of course) and then dump it into the local water supply.
If someone evil has access to the water supply, is radioactive material the worst they could do? that'll, what, give some people cancer which is really bad but it'll take a while to get them. If people wanted to be shitheads, they could already dump arsenic or LSD into the water supply, or any number of others things. that are already available to them right now! Have you personally tested your taps chlorine or flouride or lead levels recently?
You're seriously proposing that a country with regular mass-shootings should give everyone a device that can cause a radioactive meltdown or a small explosion?
We already let people have cars and those things are crazy dangerous! No one should be allowed anything, ever, until we bubble wrap everything in the world to be perfectly safe!
Well, we already have strict background checks, licensing and regulations for a scenario such as if a person wants start a home based business manufacturing and storing C4, Semtex or similar at their rural property. If the idea is to start handing out nuclear reactors for peoples' houses, the possible damage that could be done is far greater. No matter how well packaged it is or designed to be consumer friendly.
Why would I need “home nuclear” when I’m already self sufficient in power in the winter with 15k of solar and battery, let alone the summer.