This is why we need regulation. Without rules, the cost of the externalities is passed on entirely to communities very far from leadership decision makers or shareholders.
The problem with a purely financial accounting is it ignores environmental and social costs entirely.
They have most powerful AI systems and yet they can't solve their own power consumption nor climate change.
Everyone wants to make quick buck with AI powered solutions. The real problems like climate change will remain on back burner with this AI frenzy. There's no reward to solve climate change in short term unlike math problems.
We are going to accelerate climate change with AI.
Jensen Huang literally said that AI can help us fight climate change, but it has has to first "inflict an enormous amount of pain and suffering first."
People are pushing forward with an illogical, destructive ideology, saying that somehow it will solve our problems AFTER it hurts/destroys us.
That's because they don't like the solution. The solution is restraint and sobriety, and they don't want to restrain themselves, because they value immediate profit over the greater good.
Nuclear? They would simply consume more, just like an additional lane on a highway simply leads to more traffic.
The solution won't come from them. It might come from us, but it won't be pretty.
The solution could be to put AI-level spending into developing fusion power.
It's just more obvious that there isn't much money to be made from power-generation. With LLMs they have somehow deluded themselves into this being the next money generation opportunity.
Unlimited energy, even 100% clean fusion energy, would just lead to accelerated extraction and unnecessary production (the main thing we've been doing for the last few hundred years), since the main moderating force would be gone.
Why bother developing a more efficient solution if you can just cheaply chew up 10x more resources? Just like how software has gotten worse as computers got better.
For decades now it's been the same trend. More tech = more fossil fuels worldwide. Due to Jevons effect and due to companies and governments not doing enough about climate change.
We'll probably reach a peak in fossil fuel consumption over the next years, mostly for demographic reasons, then we'll see if we can keep doing tech the same way while consuming less fossil fuels worldwide.
Anthropic employees and CEO probably care about climate change vocally, but before the IPO they couldn't reduce their emissions, and now it might even be illegal for them to do so. Reducing CO2 emissions massively and quickly would mean destroying their business, investors wouldn't let them do that. Good alternatives take time to develop and cost more.
The reward to "solve" math problems is not much bigger than the reward to "solve" climate change on the short term. It's just that they're mathematicians so they care about maths, solving the Riemann hypothesis on paper didn't change the world much, people need to actually understand how it was done. The same way having the solution to climate change on paper won't change the world much, people need to understand and accept the solution to do it. If you have a solution to climate change tomorrow, people will congratulate you. But that's it, implementing this solution will be the hard task. And energy and climate change are very different from maths because there is a social aspect to it. Whether people accept or refuse the solution to the Riemann hypothesis doesn't matter; while if people reject your solution to climate change, they just won't implement it, which means your solution wasn't a real solution as it didn't solve the problem.
Anthropic employees (and other employees of AI companies) could surely do way more for energy and climate change. Maybe they just don't have the qualified people for that. It's a full time job to understand energy and climate change, I spent 1 year full time on it reading hundreds of papers, and that's really not enough.
> While it is certainly true that a central objective of for-profit corporations is to make money, modern corporate law does not require for-profit corporations to pursue profit at the expense of everything else, and many do not do so. For-profit corporations, with ownership approval, support a wide variety of charitable causes, and it is not at all uncommon for such corporations to further humanitarian and other altruistic objectives. Many examples come readily to mind. So long as its owners agree, a for-profit corporation may take costly pollution-control and energy-conservation measures that go beyond what the law requires. A for-profit corporation that operates facilities in other countries may exceed the requirements of local law regarding working conditions and benefits.
That isnt the solution, just like solar and wind arent the solution.
The problem is that every time we add cleaner forms of energy to the mix, we find ways to use that extra energy. On top of all the fossil energy we continue to use.
I'd say it's worse: solar and batteries are now cheaper then nuclear and could do the same job...but the US government has gone an active crusade to ensure they're also unaffordable and impossible to build.
We can't even properly store 80 barrels of spontaneous explosive for 20 years. What makes us think we can properly store 10,000 barrels of nuclear waste for 10,000 years?
By definition though, doesn't "everything" accelerate climate change in our society given it's capitalism at the core of what you're describing i.e. short term thinking?
In pre AI days some companies were at least committed to green energy. I know Google was one. They promptly dropped their commitment to clean energy as soon as AI became mainstream.
I'm not immediately sure why people feel such a strong moral injury or sense of doom from AI co2 emissions but not from zinc, copper, steel, aluminium, concrete, pulp & paper and so on.
I suppose it feels very _optional_ compared to those other things. A lot of tokens are gonna be spent optimising ads, for example.
Mobile phones have become necessity just like AI will be. I don't want to use one but I'll have to if I want to compete and earn my living in this world.
BTW I have a Nothing phone. I don't upgrade it with every new version.
Neither are iPhones, on a global scale. Apple phones have good resale value and hold well for years. If you want to go after anyone here, I'd go after low-end Android phones, or all those new upstart phones. Saying so as an Android user.
But of course the whole thing is backwards; both smartphones and LLMs have some of the highest utility per watt (including embodied power) of all goods and services worldwide. Plenty of much greater waste to go after, but it's not as sexy to complain about vacations or sportsball or fashion or home decoration or DYI arts & crafts or ... as it is about smartphones and data centers.
couldn't you have solar + wind (peaks at day vs peaks at night) plus batteries plus a small peaking gas plant for really bad stretches of weather? On a yearly basis it should be possible to average 80%+ renewable power for it, at the expense of course of using a lot of land and costing way more than coal+gas
I am hoping at some point deep geothermal will help with this as it works anywhere and would provide predictable 24/7/365 base power
People already hate data centers going in. Now you're going to expand their footprint 10x or more for the solar and wind farms? Lots of protests for that too I'd imagine.
Not we. They. They need to do it. The general public should not be on the hook for anthropic's energy needs. Make them pay for every dollar of power plants and grid upgrades.
So what? Is getting people to form friendships, even relationships with AI more important than the friggin' environment? Is it so important to have office drones sending emails written by LLMs in order to receive replies from LLMs? Is it so important that everybody can write a TODO list app?
This is why we need better smaller models. These companies will continue to make bad decisions. With better smaller models at least we don't have to participate in their bad decisions.
This is one of the myriad of reasons why investors are shedding from tech. They’re afraid with the change in US political power, there may be regulations that run counter to unfettered growth.
I thought about some leeway for temporary emergency power from generators. The flywheels won't run for long enough unfortunately and batteries would have to be huge for just emergency supply.
Thats just giving rich people the rights to shit in common garden by x bucks per shit.
I (and next generations who will be the most affected) should not have to deal with the shit because some rich people got to shit everywhere as they could pay higher taxes.
I cannot gesture at everything around hard enough to convey how bad this take is. It's not like we needed regulation for ozone restoration, lead removal, killing smoking in shared spaces, and so many other things... both customers and producers would just make the right choice for society, right? Customers are just choosing services that aren't powered by coal, right?
So where do you draw the line then? It sure sounds like this thinking leads to the understandimg that people are dumb and don't know what's best, and that the government does and should always force that view on the public.
Can we regulate this? Sure. But why regulate this specifically, especially when the public/consumers don't seem to care?
There can be lots of nuance for questions like that. But "things that can mess up humanity and the world as we know it for many generations" are extremely on the "regulate it" side and we've got a good track record so far. There's not really been anything that backfired in a way that can't be corrected.
AI data center power use is going to mess up the world for generations?
I do want to see them using less power, or not being built at all frankly. But that claim seems a bit extreme.
I put this in a different thread I believe, but China burns enough coal to run 300+ 2-gigawatt factories. We have larger fish to fry when it comes to energy use, our focus on data centers specifically doesn't correlate with their percentage of the problem.
This is a classic tragedy of the commons. That's a good place to draw the line.
The original example was overfishing. Each fishing boat has little effect on the overall fish population. Regardless of what the others do, each one ends up with more money catching as many fish as it can. But when they all do that, then the fish population collapses, which is not the outcome any of them prefer.
When they cooperate instead by regulating how many fish each can catch, they can avoid population collapse and everybody's happy.
This is why we regulate things. In the other direction it's why we pay taxes for public goods. Solving these collective action problems is one of the main purposes of government.
My argument wasn't that regulation should never be used to make meaningful gains in environmental damage.
My point was that throwing regulation at a single energy use in one area of a single industry is inefficient, won't make a meaningful impact on total energy use, and focuses regulator attention away from much larger change they could be making.
People can work on an urgent thing and a general thing at the same time and they do all the time. Temporarily putting a stronger restriction on something getting out of control, that you have more power over, doesn't mean you ignore a more general version of that problem. We still use local regulation for what to do with trash in the area, rather than saying it doesn't meaningfully impact on global garbage production and letting it collect on the streets. I don't believe anyone here is proposing dealing with DC usage and ignoring other energy sinks.
We do that already, and my point isn't that we should never regulate.
There are much bigger dish to fry if we want to regulate energy use down or require cleaner energy sources.
"Just regulate" as the OP comment is a flippant use of regulation targeting a specific use of energy in one area of a single industry. That's inefficient at best and won't make meaningful change.
Do you know how long it'd take to set up a solar array large enough for even half a typical AI data center?
I doubt Anthropic has that time right now considering the amount of stuff they're actively doing.
Whether or not it's the right thing is a different story, but just imagine the realities and the time sink of actually setting up solar for a data center powerful enough to serve stuff like a frontier model.
To play devils advocate it’s not a bad thing. It’s actually ideal and 99% of the human experience. People did not choose civilization. Civilization was forced upon people to manage higher density populations. Starvation is worse than civilization.
All life expands until it runs out of resources. Fairly sure cyanobacteria killed 99% of life on earth (anaerobic life) at some point. Trees killed 90% of life once. At least we can learn to do better. Hopefully.
Why does their urgency mean that the rest of us have to suffer?
If I decided to ignore laws/regulations when building/repairing my house, I don't think a court would be interested in "I'm in a rush, so didn't have time to get planning permission etc".
The irony is that we've been burning the planet to pursue the advancement of technology and the 'grandness of the human race' since the industrial revolution.
This is, unfortunately, just the next chapter of the same story.
More unfortunately, it feels as if the only way to change the direction of the story is for there to be real, actual, visceral consequences for a majority of the people that are powerful enough to continue pushing in the current direction. Fatally unfortunately, said powerful people will be the last to feel real, actual, visceral consequences. There is a lot of 'worthless human shield' sitting in front of them that has to be ploughed through first.
Doubly ironically, maybe it's best if AI does destroy humanity and let the Earth find a more sustainable balance of life.
It will be interesting to see, if AI does destroy humanity, will it let the Earth find a more sustainable balance of life, or will it accelerate the domination of the natural world in pursuit of its inscrutable goals? (likely to its own suicidal detriment, given it is based on learning from humans)
of course; the realities are that gas and coal are heavily subsidized, so theres never been a chance of making a fair comparison among real dollar per megawatt
No, because solar panels are notoriously hard to scale, especially in high density applications like datacenters. Otherwise it's like saying "my calculator runs off solar power, so why can't they make a car run on solar power? Clearly it's because Ford/BYD not giving a f"
The investments in AI is in the trillion of dollars, surely that's enough capital to figure this out. It might require inventing or improving existing technology.
Before you respond "it will be too slow", the urgency here is entirely artificial – there's no reason we need to be going this fast.
> urgency here is entirely artificial – there's no reason we need to be going this fast.
What makes you believe that time is irrelevant? If AI allows for breakthroughs in mathematics, science, engineering, computing, and health care—and it does seem to be delivering on this promise so far—what in your estimation is the net cost to humankind of delaying those breakthroughs by even a year? It's not zero, is it?
I mean sure, but the labs themselves are saying we need to slow down. Your position is only reasonable if the negative externalities are outweighed by the positives. Not to mention p(doom), which – unless yours is 0% – should mean we must slow down or stop immediately. We're justifying definitive downsides with speculative upsides.
More than anything though, it doesn't matter what we in tech think because we have failed to convince the general public that any of this is a good idea.
> It might require inventing or improving existing technology.
Or it might result in investing trillions of dollars and decades only to discover that there are only marginal gains to be had. There is no guarantee that throwing money at Problem X results in any kind of solution or that solutions to specific problems are even possible.
This is aside from the sibling commenters' very good point that this view assumes there are no concrete benefits or advancements to be made that depend on AI, which seems very unlikely at this point. Unfortunately, with or without AI, the problems we're facing socially, ecologically, medically, and scientifically are increasing. We can either use AI as a tool in the hopes we can at least delay the most serious consequences or just let them overtake us. Fossil fuels and climate change are just one of a million very pressing issues, despite the over representation it gets, and it seems highly likely that the benefits here exceed the costs. Addressing a downthread comment about labs saying we need to "slow down" - they are saying this for self-serving reasons and in their quest to ensure AI stays gatekept, not because they are worried about environmental impacts.
I don't see how we could invest trillions of dollars and get nothing to show for it. Consider the Apollo and Manhattan projects, both were large investments that resulted in significant scientific and engineering progress, why would this be different?
I'm not sure if the car analogy applies. Data center solar power can scale without the requirement of being portable.
The problem seems to be that they want the power online quickly and they can't get solar quickly. The article mentions the inefficiency of their generators. It is costing them quite a bit extra to generate power like that, but that's their only option of they want the power immediately.
It would be nuts to do this long term. You might argue that they don't give a f about the environment , but most people will agree that they give a f about cutting costs.
> Data center solar power can scale without the requirement of being portable.
You just have other issues like needing to get transmission and storage capacity built, which takes forever because of planning review, environmental assessments and NIMBYs.
You can, but how many cars are actually powered this way? Same goes for datacenters. It's not like electrons from solar panels physically don't work with GPUs, it's just that it's hard to make it work due to other constraints.
For electric cars it isn’t uncommon for people to put solar panels in their homes or their garages to charge them. I think the car themselves with solar panels won’t work because most people store the cars in garages usually, away from the sun - it is usually very harsh on everything that isn’t solar panels.
1. From a pure argument lawyering point of view, my original question was "why can't [Ford/BYD] make a car run on solar power?" with comparison to a calculator, which implies a car with built-in solar panel. Otherwise the statement of "they make [thing] run on solar power" becomes meaningless. eg. "Apple made a macbook run on solar power" (because you can take a regular macbook and plug it into any solar panel.
2. There probably are non-zero amount of datacenters out there that runs off solar power. It's just not commonly practical, just like with electric cars.
3. The main point of the comment is that scaling challenges exist, and you can't take a residential use case and extend it to a datacenter. The same applies for a car, even if you could do it. A solar powered calculator is easy, just slap a tiny solar strip and it just works. A solar powered car is far harder to pull off. If you don't work from home for instance, you'll need a battery system to store the energy when you're not home.
This is simply an argument against datacenters. You're pointing out a negative efficiency, and the ostensible point of concentrating data processing is to produce savings.
If you can't figure out how to get renewable power to a datacenter, but you can easily figure out how to get it to the consumers of a datacenter, and you're talking about a technology that scales down to small groups well and individuals only a bit less well (but with constantly working agents utilizing compute, better than we once imagined), what's the reasoning behind datacenters at all?
If datacenters can make a case for massive increase of demand in the future, they're simultaneously making a case that individuals will have enough demand to saturate AI located closer to them or on premises. If anything, that should be a call for p2p software to help people share or borrow excess compute with and from their neighbors in exactly the same way that home solar panels use the grid.
Datacenters are also central points of failure. Distribution is resilient.
Sounds like we shouldn't be building big data centers if there is only nonviable means to power them then?
I know meta and their ilk are trying to cut deal for nuclear power. I'm sure we can imagine they will take safety, the environment, etc very seriously too.
Yeah also they are loss making and will basically be fighting for survival now that they're public and visibly not profitable. Private investors have offloaded to the hyped public bag holders and now we're seeing how the sausage is made.
YOU make a poor argument. Most sysadmins read Slashdot all day, and complain about bloatware or Windows 27 Server security profiles. By default, is your AI research into latent atomic arguments... or monotonic parabolas? My policy lawyer sniffed glue as a child so you can't convince him. /gosh wow
Did anyone ever think these people cared about anything other then money? Spacex basically operates out of texas so they can power dcs with gas. It's part of why musk dropped half a billion on midterms.
I think all of the major ai labs are running on gas it's ridiculous
Meanwhile Chinas grid is super healthy and has a tremendous amount of renewable energy... This is what happens when you don't pay into infrastructure as a civilization or require corporations to participate or have accountability for the future of the people they serve.
China generated 5,500 TWh last year with coal. The largest AI data centers running at full power would use something on the order of 17.5 TWh.
Yes I agree these data centers should be using more renewables, but the China comparison is odd given that they burn enough coal to run 300+ 2 gigawatt data centers.
The scale of energy that AI data centres use is huge in a way that most people don't have the ability to visualize. They think of a computer that their cat likes to sit on and multiply that gentle warmth by some number in their head. This is completely unrealistic.
Here's a quote about a data centre currently being built near Edmonton, Alberta.
------
"Utah State University physicist Robert Davies was hired by SLCN to conduct a preliminary technical assessment. He said if the Wonder Valley project goes ahead it would release the equivalent of 23 atomic bombs of energy every day, though he noted the comparison has its limits."
This is not the total energy being consumed by this data centre. It's merely the waste heat it will release to the surrounding environment.
Here's the energy consumption:
------
"At full operation, it would consume energy at a continuous rate of 18 gigawatts or about 1.5 petajoules per day and emit about 30 million tonnes of carbon dioxide a year – more than four per cent of Canada’s total emissions, said Davies, whose assessment was based on publicly available information."
------
That's the equivalent of 1.66 million Canadians in carbon emissions, or 1.6 Edmontons. This is just one data centre. Many more are going up across Canada, and Canada is both late and slow to the buildout.
That's the kind of scale we're dealing with at a time when, as a world, we are trying to reduce our carbon emissions.
> at a time when, as a world, we are trying to reduce our carbon emissions.
From my perspective, that is not quite the correct way to put it. Because honestly we are not even trying.
---
I find those "Olympic swimming pool type" comparisons very tiring, to be honest. They certainly don’t help my grasp anything. "23 atomic bombs of energy" yeah okay
It sounds really dramatic. Even worse, unlike an olympic swimming pool, nuclear bombs vary drastically in terms of energy output. Like, several orders of magnitude. If they mean 18GW as mentioned later in the article, that's about the energy consumption rate (total, not just electricity) of a city like Houston.
if the prices of domestic and international flights were cheap, people would be flying a lot more, I saw a girl say stop using generative ai because it hurts the environment as a caption on her trip to Iceland, the irony is staggering.
One thing that make absolutely no fucking sense for me is that in places of Europe, for intra country (so good train network and dense population and no so long distance) train is quite more expensive that (subsidized) plane.
Trains have to deal with thicker atmosphere (closer to the ground) and have tracks that need maintenance. Planes only need a few square miles of tarmac.
Why it doesn't. Running trains is simply expensive regardless of country. Swiss have excellent network and its very expensive to use even for locals (I live here and use it for daily commute), for tourists its prohibitive. Other places have often much less quality / reliability / reach and prices are still high.
Cheap train network is a proper pipe dream, not possible. Much better to have fully self driving electric taxi cars (so not that joke with driver constantly needing to be aware if your tesla wants to steer you to standing ambulance / firefighter trucks on highway, I mean no driver setup to push prices to very low levels).
>The scale of energy that AI data centres use is huge in a way that most people don't have the ability to visualize
While I can't really visualise energy at all, this does make it seem like more than the 0.5% of world electricity consumption that it actually is.
While it is currently cheaper to build renewable energy capacity than most fossil fuels, as far as I can tell, the reason most data centers are not doing so is because they can't, All of the renewable infrastructure being produced is currently being deployed.
The question they should be being asked is when will they be switching to renewables. The height of the plot is not the important thing. It is the area underneath that matters,
0.5% of the world's energy is massive! Every other thing society does - agriculture, shipping containers across the ocean, flying planes - and 1/200th is for AI? The thing we didn't have a couple years ago and we were all fine without?
"When are you going to switch to renewables?" has proven to be a useless question. Companies will say whatever is expedient, and then they'll buy credits or whatever green washing instead of actually reducing consumption. If you let them double their energy consumption by adding renewables they'll just use double.
For starters, 0.5% of the world's electricity, not energy.
Most of the worlds energy use is not electrical.
It is a large amount. I do not deny that, I just want people to consider it in terms that are true. Every person who I have asked how much they think it is based upon the coverage they have encountered has massively overestimated.
A common response is that considering the concern reported in media they estimate data centers use 10-15% of electricity. How can we have an informed conversation about the relative importance of things if people have been misled about the relative scale.
Now it has a bit at the bottom
This story and its headline have been updated to provide further context about the preliminary nature of Robert Davies’ assessment and align with our journalistic standards and practices
I was thinking they were talking about current usage. Not guesses of how much something might use.
I'm not sure how much size matters of individual data centers apart from their contribution to the total. If they sum to the same amount, how much does it matter.
If it's true that you can build something of that size without an environmental impact report then there's something pretty wrong with the laws there.
If they centrally located that data center, the heat wouldn't need to be wasted at least six months a year.
Also AI data centers weren't responsible for decades of people burning natural gas to heat their homes without investing in insulation nor the tar sands oil which is really screwing the climate.
Who are responsible for sprawl (and thus dependence on automobiles and freeways). And the hardest part of siting new solar plants. And why North Seattle has a dearth of affordable grocery stores (... that one might be a little hyperlocal).
> He said if the Wonder Valley project goes ahead it would release the equivalent of 23 atomic bombs of energy every day, though he noted the comparison has its limits."
Saying this comparison has its limits is an understatement. Energy released by different atomic bombs can vary across different order of magnitudes.
Is this supposed to sound very impressive? It genuinely doesn't to me.
I know atomic bombs are scary, but that's because they explode very quickly in a very small space. (Radiation aside which I guess was not the point of the comparison)
If you spread that energy release throughout a whole day and a bigger area, letting the wind slowly take the energy away, then it doesn't sound scary at all.
Not to mention that the power to the data center comes from somewhere. So whatever amount of fuel, solar, or wind went in to the system is the most that can come out.
Using nuclear bombs as the measuring stick is a cheap rhetorical tactic.
The point of reducing emissions is to increase QoL. But that should not come at the cost of reducing QoL itself. That's just a religious need for degrowth
Yes but the data center hardly runs off the internal combustion engine.
The energy consumption of data centers is irrelevant: our complete failure to deploy the technology we've had for the past 40 years, or the last 5, to produce carbon-free energy is the real issue.
I'm rather tired of using energy consumption as a convenient bludgeon in the OmniCause, by people who have typically steadfastly opposed what we could've been dong this entire time and can be doing now.
Build more datacenters. Permit more solar and wind to power them. The money is flowing so I'm pretty sure we can afford some nuclear plants to backstop the baseload as well. This is all ridiculous: we should've ended carbon fuel dependence for stationary power in the early 2000s.
Given the very limited impact a small amount of latency would have, is there a reason not to build data centers in the Sahara, surrounded by solar panels with energy storage for night use and using a closed-loop water system to sink waste heat deep into the ground?
At least from the news I've seen across the ocean, I was under the impression that the USA's grid was already stretched way too thin because of their lack of investment in renewables, and many people are against data centers because electricity is already too expensive for normal people. Or was that just news cycle hyperbole?
The US people have nothing to do with that. The public US grid is indeed not sufficient, but if you read the article, you'll see that these companies now bring their own power. And since fossils (primarily gas/methane) is the only power source that can deliver huge amounts of electric power in a foreseeable time frame, companies like Anthropic are primarily interested in that. You can buy a couple hundred gas generators, stick them in containers next to an oil field in Texas (which would otherwise just burn the gas) and have a GW for your datacenter before you'll even have the permits for a conventional fossil or nuclear power plant or an equivalent solar/wind farm.
Because transportation costs money. It's the same reason why farmers occasionally dump milk or let crops rot on the field because it's not worth the money to harvest/distribute.
> USA's grid was already stretched way too thin because of their lack of investment in renewables
That doesn't follow though. If USA's grid is stretched too thin, it's because of lack of investment in power generation. Not renewables specifically. In fact, renewables are still the most roundabout/difficult way to go about addressing that. And as unfortunate as it is that gas is the economically-best option (I'd vastly prefer small nuclear), if data centers are forcing investment in new power generation, that benefits the grid.
Renewables generally have highly variable output, which necessitates overbuilding generation capacity across geographically distributed sites, as well as a much larger transmission capacity buildout to balance regional variations in generation while maintaining acceptable supply stability across the wider grid.
It's a cool idea, but even if they appear in 4-6 years, there will be many companies wanting the extra energy during those 4-6 years. And it will be a massive amount.
More than 1/2 of my energy generation at my home is grid sourced as renewable. I have a HA integration that tracks this semi-realtime. The cost is ~14c/kw ... about half of what the EU pays (on avg)
What you've described matches some places in the US. It also matches some places in Europe.
But, for counterexample, my power bill in the US is included with my water, sewer, irrigation, and trash collection. Monthly total is between less than 1% ($100) in the summer to around 3% ($300) of a median monthly household income ($10,000) in a metropolitan of about 1.5 million people. Here, the electric grid is owned by the city, not privately operated.
Generally speaking, bigger cities have a high cost of living. And the same can be said about bigger cities in Europe.
But having lived in both places, I find that life is so much harder in Europe, in terms of cost vs income, and in what your money can buy.
Is the media lying to you? It's stretching to focus on specific examples. Sensationalism is profitable. And often times media is colored by corporate or government interests.
The Australia–Asia Power Link (AAPowerLink), aka SunCable, is a proposed undersea HVDC cable to export solar power from northern Australia to Singapore via Indonesia. Due to doubts about the economic feasibility of constructing the cable, and question marks over the willingness of Indonesian and Singaporean governments to approve it, they've been exploring pivoting to building solar-powered AI data centres in northern Australia instead.
I somehow suspect that datacenters are only economically viable if you run them round the clock, right? So at night they'd still be using gas. I guess the cost of compute would have to come down a lot, in order to power down a DC and spin up one where sun is shining instead.
Although I guess for current AI workloads, it could be viable, since the additional latency of sending your (small, text only) prompt around the world for processing is little compared to the time taken to process (whereas for other stuff being close to users matters more)
> I somehow suspect that datacenters are only economically viable if you run them round the clock, right? So at night they'd still be using gas.
You could build batteries, charge the batteries during the day, run the DCs off it at night.
SunCable / AAPowerLink is likely going to be connected to the Darwin grid, which is currently not connected to the main East Australian grid; but building a connection to the main Australian grid would be a lot cheaper than undersea HVDC to Singapore. Once that happens, there would also be the potential to use pumped hydro (including Snowy 2.0) for storage.
Given this year’s events [0], even Sahara seems more likely.
But you’d still have to build it out from ≈scratch, including cooling, internet connection infrastructure, and maintenance, so I’m not sure what the time horizon on amortization would look like, even with the cheap solar power.
Aside from a dirt road from the nearest port and a few fibres to various IXPd what do you need to build in Morocco or Egypt that you don’t need elsewhere?
Obviously security is a concern which limits the number of locations, but there’s still plenty of places with stability, security, land, and the lack of taxation and tarrifs that the US puts on energy.
There have been many proposals to build solar in Africa for exports to Europe, but none of them has gone anywhere. That's because the cost of capital is too high. It's cheaper to just build in Spain.
Globally, places like Australia and India should be more competitive for solar than the Sahara, although both Australia and India have a lot of dirty power generation to replace as well.
They could run on solar in US throughout the daylight hours. Just make the tokens cost 5x after the sun sets That would reduce the overnight demand enough the data centers could run through night on batteries.
You mean recently? Western countries have not had a nuclear energy plan over the previous years, many of them wanted to phase out nuclear reactors, some countries even started to do it, and some entirely did it. It's not a very encouraging situation to build nuclear reactors quickly. Countries which made ambitious plans with nuclear reactors in the past managed to build dozens in less than 20-30 years. But that does require massive investments. It's much easier to invest less money in solar panels and wind turbines, whether that second choice allows to remove fossil fuels as fast is something we can already observe.
Long term plan is so long and so crushingly expensive it’s not a workable solution to the needs of today.
Short term plan is much faster and cheaper. Therefore it’s a much better plan.
China add an entire US grid worth (1,281GW) of renewables every 3 to 5.3 years. Why wouldn’t the US do the same and have the problem well in hand in 5 - 10 years rather than waiting 20+ for a single nuke?
There's always an option of... wait for it... datacenters in space.
*runs*
But then of course that's another front-page submission with its own long thread of commenters feeling smart for calling it stupid, despite recently becoming a worked out plan and addressing exactly the (subset of reasonable) things people are complaining about here. In fact, we had those a week or two ago.
I swear the tech space turned into thought-space NIMBYs. "I don't understand any of it, but something about it pattern-matches to something I don't like, therefore it's stupid and people behind it are evil". In this thread, and in every other thread recently, too.
I looked into it just now, and it appears SpaceX is targeting deployment of the first satellites in 2027.
To deploy 1 GW of capacity, apparently 100 Starship launches would be needed.
On earth, 23 GW are currently under construction, 100-132 GW is already active, and about 100 GW should be added by 2030.
SpaceX's CFO supposedly cited 2029 as the timeline for multiple GW of compute in operation in space. Musk's vision seems to be launching 100GW/year or 1 million ton of satellites.
This would require 10k launches of Starship per year under ideal conditions, or 5k if Starship Block 4 can double the payload capacity to 200t as planned. This would mean 14 launches per day.
It does sound like a lot. Starbase Louisiana, at a $100B buildout, is planned to have 12 launch pads that could handle 30 launches per day.
Obviously there's much more needed than some launch pads. Anyway, that's the numbers and projections I found.
And that’s the optimistic take, assuming Musk’s projections are correct. Yet they are so often wrong there’s a full Wikipedia page just for things he said would be coming (often “next year”) and didn’t.
Yes, and I am curious about the financial viability of that compute.
From a capacity perspective, if SpaceX remains successful in scaling up their Starship rocket, it seems putting AI compute into space could at least realistically happen at significant scale.
The amount of CO2 spent launching the data center to space probably offsets the lifetime saving of the solarbased energy used by the data center. That is ignoring filling atmosphere with toxing fumes on re-entry, and that after the short lifetime in space the servers, GPUs and solar panels don't find a second life.
Should SpaceX solve the problem of requiring fossil fuels to launch (doable) and the burning things on re-entry (harder) I don't oppose solar data centers. But without those, it's just externalizing the environmental cost to future generations.
> "I don't understand any of it, but it pattern-matches to something I don't like, therefore it's stupid and people behind it are evil".
This is pretty bad-faith. What is the point of your comment? You’re sowing hate and attacking a nondescript group, for what? What does that achieve apart from being hateful?
Yours is one of the few usernames I recognise on HN. I remember seeing some comment from yours (I think it involved Star Trek) which I found interesting at the time and somehow the username stuck with me. Since then, every time I bump into your comments they have gotten more hateful and generally worse and more tribalistic takes.
It genuinely saddens me. Especially as you have in your bio “I'm constantly looking for opportunities to do something actually useful for humanity”, but your words reveal the opposite.
I’m not going to be specific about past posts because frankly I no longer remember their content, only the disappointment I felt. But I can take this one as an example. In addition to every thing I mentioned, I find it a pretty bad take to claim that adding more shit to orbit improves humanity. We need less crap up there, not more. I also find it nonsensical to defend (if we can call it that; you just attacked detractors with no explanation) something that is purely theoretical (which it objectively is; it doesn’t exist yet).
Anyone who has paid attention in the last decades will see all the unkept promises. You can see them in this very submission. Blockchain and AI were full of proponents making the argument that “all this energy consumption is good, actually, because it’ll increase investment in renewables”, but obviously what we’ve seen is an even bigger spend on fossil fuels which are exacerbating the problem. Believing the next promise from proven liars who make everything worse every time is not a smart strategy.
Anyway, I’m not entirely sure why I felt I should write to you specifically. I guess I once saw the possibility of something better and am hopeful that being respectfully called out might help you to reflect. Those of us capable of introspection (which I believe includes you) sometimes need a little nudge to get back on track to the person we want to be. I know that’s true for me. Sometimes I help others, other times they help me. I hope this helps you.
I know there are smart people on HN, and I don't believe most of what (person that shouldn't be named) says. But I agree with OP, the amount of hate anytime the topic comes up is frustrating. And the arm-chair aerospace engineers are are the most annoying group.
I’m not sure you understand the meaning of irony. But I can say for sure you didn’t understand my post, because it’s exceedingly clear you missed the point. Data centres (whether in space or on Earth) and the people who oppose or defend them are very much not the focus. Fortunately, I believe the person it was meant to might understand. That comment had an audience of one.
Of course, comments and downvotes like yours were very much expected. On HN I’ve come to anticipate lack of nuance and the belief any disagreement or criticism is hateful and looking for a fight.
I can only try to counteract that one post at a time, though I can’t say I’m feeling hopeful that the bother is worth it.
1) Musk retches up another one of his Hyperloop-style dumb-ass ideas whilst patting himself vigorously on the back
2) His companies grovel hilariously to backfill the reasoning for why this is a Good and Smart idea, actually, while actual scientists put on a wtf face (“worked out plan” goes here)
3) The technorati follow the their glorious thought leader off a cliff
Just price greenhouse gas emissions (or fossil fuel extraction). These sort of weird hoops for new users to jump through fail to consider the possibility that some things are worth doing even if they increase net emissions. A price on emissions would move savings to the least valuable emitters, rather than the newest emitters.
All that means is the price of generation is bid up, making it more costly for everybody else. For example, PSE in Washington State has a mandate to be 100% renewable in ~fifteen years. They recently got outbid for a solar plant in Oregon by a compute company (Amazon? can't remember off the top of my head).
So PSE's consumer rates are scheduled to go up 30% just in the next three years. It's not going to slow down if AI is bidding up every power plant. And in states without the mandate, it just makes keeping gas plants running more cost-effective, because power is fungible, so there's still a net increase in fossil fuel generation despite said mandate.
The builders of these datacenters will tell you that those two things are the same thing. If you don't let them build datacenters without any regulation, in the cheapest possible spot, with the dirtiest possible power, then you are in fact banning them.
Yes but anti-caps focus way too much on armchair analysis, the dont's and bans and forecasts of failure. They're not completely wrong, but to move the needle they need to propose actual solutions and get them to stick - even if incremental.
I wrote a piece in the wake of the midterms about why everyone hates data centres[0]. This kind of stuff is a prime example.
The labs have failed entirely to predict this opposition and it's an existential risk to them. They have not been able to convince people of the urgency of AI build-out and certainly aren't assuaging them of concerns about the costs.
Push back. Sure solar and wind isn’t always on but you can still maximise what’s possible and invest and find new ways to make it cheaper. Same with closed loop water. Don’t like data centers being built? Good! Demand more. They should be a net win or not be built at all.
Has there been any large scale load following data centers? One would expect data processing to easily shift to where energy is currently the cheapest. At least in a functioning economy and not a semiconductor-starved one as ours is said to be, but that situation is hopefully temporary anyway. This would be well suited for renewables and other forms of cheap energy that might otherwise to go waste.
Mistral is building their data center in France, which has a 95% low carbon energy mix thanks to lots of nuclear power paired with wind and solar. Despite the U.S. tech elites talking up nuclear for decades, they've never really walked the walk and I don't believe they ever will.
I've been trying Mistral Vibe side by side with Claude over the past few days. I had very low expectations going in. But so far I've found that for regular chat questions there doesn't seem to be a meaningful difference between Opus and Mistral Large 4. Smart enough is smart enough. For deep research and coding the situation is different of course, where for some things I still need Opus. But I really don't need more intelligence for pretty much anything than Opus 5.5 gives me, and I often need less. So if Mistral could reach that level somewhere in the next year I could imagine moving over entirely.
>data center in France, which has a 95% low carbon energy mix thanks to lots of nuclear power paired with wind and solar
This is a bit of a virtual accounting, right? I mean this energy would be consumed anyway, and people who would use this green energy now have to use coal/gas. Still better than building your own gas powerplants, of course.
What type of carbon sink system are you thinking of, how much carbon a year would it sink, where would the carbon go, and what would be the energy cost of that system?
Because it takes so long and is so risky that it's not financable. The only people who can realistically find nuclear are governments or charities, but even they look at nuclear and think it's a bad RoI.
Why are we letting this happen? Why are we letting the few burn down our planet in order to further decrease reliance on (and opportunities for) the many?
I dont know, I really hate the direction this is going. Im not even that confident in the capabilities of llms, but that doesn't really matter. We're selling our futures, our environment and trading our competency to a couple (greedy) corporations in exchange to be able to cheat at our jobs.
Yes, everyone I know including myself at times, is just cheating. Doing a similar amount of work as before with less effort/time involved. There's not all that much productivity gains, maybe for a select few who practically live in a state or mania (burnout inevitable).
I dont see it as all that positive for anyone long term. Im constantly having to relearn things, never retaining anything I learn with these things, and forgetting everything I used to know. Its literally causing a cognitive diabetes of sorts. Like someone who drives their car everywhere loses muscles, gains weight and eventually gets a metabolic disease, im certain this happening to our brains. Im 100% certain of this. Its a net negative all around, there's no such thing as a free lunch.
Even if everything goes smooth with the progression of this technology, best case scenario is we have massive unemployment and people with jobs make 20% of what they did because they're now practically just a meat proxy, the equivalent of a conveyor belt operator. Suddenly find ourselves working for a dozen or so Oligarchs, tons more crime, zero uppward mobility. Thats best case scenario. Im not sure why everyone is so excited about this.
Not sure what your point is. Because it's not refuting the claim that billions want to use LLMs.
And didn't Google admit that their AI summaries are biting into their ad business (causing their recent drop), meaning a huge number of users are deriving value from it? Possibly billions.
And then there's all the white collar workers sending emails. SO many people are suddenly sending grammatically correct internal emails now, AI policy or not.
(Of course, the recipients of those emails are not as happy, since someone who used to send shitty emails now instead write shitty prompts that don't actually ask the LLM to address the real issues. But that still counts as demand)
> More people are willing to pay for netflix or youtube premium than AI
And even more people breathe air. In both cases the analogies don't work.
The point is that people use a lot of free things because it's free, not because they need it. It really isn't rocket science or up for debate.
You can't infer demand form "free users" because just from microsoft and meta shoving AI in every single crevices of their free tier software you already get like 2b recurring users even if none of them use/asked for/care about ai features.
If the next windows 11 update turned the mouse pointer into a turd would you say 1b people demanded a turd mouse pointer?
I think people intrinsically don't give a fuck either way, they just go with the flow and rationalize their behaviour so that their definition of "good" warps to fit what they happen to be doing. Otherwise lynch mobs, genocides, and Peter Thiel would not exist.
But we let millions of people fly around the world just to do stuff they could be doing in their own back yard for a few days, then fly back.
Maybe before doing absurd comparisons between LLM use and murder, let's take a look at the energy use of people's vacations, or sportsball and concerts.
At least LLM use (and murders) are productive activities.
Yeah arguably that should be cut back as well unless it's electric or hydrogen planes. Tax free fuel for airliners is a kinda insane fossil fuel subsidy when you think about it, and leaded fuel for piston aircraft should've been banned 50 years ago to drive innovation in the sector. Instead we still somehow have avgas today, it's like if we were still using freon and DDT.
so we should all abandon this new llm tech because...sustainability?
if the industrial revolution never happened or we never invented the wheel or agriculture, would the world be better off?
this anti-tech nonsense from people is hilarious. You WANT modern medicine I promise you that. If it wasn't for soap, dentistry and a bunch of other modern conveniences, we'd all be at each others throats trying to survive on what nature can 'provide'.
Nah just make them invest in solar to power them, it's not that hard and it's cheaper than it's ever been. How many people are using LLMs at night? It's a load that's entirely day driven.
Nobody's said abandon the tech. Abandon the political economy within which a handful of people get to use the tech to centralize power while burning down the material conditions for survival of humanity.
If not for the labor movement, the history of strikes and struggle and the resulting transformation of social conditions, humanity would be worse off from the industrial revolution. Ten year olds would be working twelve hour days in factories with zero attention to safety.
The tech itself is just a device for changing the social conditions. The direction of change is what the struggle is about. If we leave it solely to our oligarchy, they will make sure to use it to further entrench their interests while simultaneously reducing the bargaining power of labor.
Because public attention for most people has been successfully captured and turned towards political sink holes like immigration. We needed WW2 in Europe to understand the dangers of fascism and keep that understanding for three generations. We will need more forest fires and floods and draughts and the immigration those will actually bring before we understand that we fucked up. By then it will be too late of course. Sometimes I feel guilt towards my children. They will have to pay for our mistakes in blood.
Mass immigration has massive negative externalities for the locals: housing prices, crime increasing, infrastructure being pressured. It's not some silly political sink hole.
Minor problems that can be solved. To get a bit of a perspective on this maybe contemplate the fact that it is pure luck that you were born in a country where people want to move to rather than from.
Nah but none of these first worlders want any contemplation. All they wanna do is have their cake and eat it too.
They polluted the entire world and somehow AI equalising the equation is an issue. There's a reason developing countries are far more amenable to include AI in their processes. While these people have got so fat and thick minded from prosperity they cannot care less.
Housing getting more expensive is a consequence of supply constraints. Immigrant labor could totally help build houses; we just have zoning and lots of other hurdles that make building hard.
Just to tie it back to the start of the thread though, migration is a much bigger driver of carbon emissions than AI.
[The TL;DR for this claim is that moving to say the US or Australia from a less wealthy country is the one of the most carbon intensive things a person can do].
Of course all this is bound up in the economic growth that migration brings and how you feel about that is a values question.
It's not really within the overton window because the people who care about carbon don't want to restrict migration, and the people who want to restrict migration don't care about carbon.
No one knows, but if you believe Amodei's optimistic projections, it might delay finding cures for most major disease.
What would be the upside for the environment? There is no burning of the planet that it would prevent such as some comments here said. What precisely one could name as a realistic consequence of using more gas and coal at that scale over the next years, I am not sure.
There is also the question whether process improvements enabled by AI can offset that.
Health correlations with pollution, especially near population, is very well understood on a statistical level. That's a very clear cost - more asthma and similar breathing problems and aggravating existing such conditions.
Yes, remains to be seen. I don't know the answers to my own question I just think it's worth thinking about. I'm not very optimistic in the short term, maybe it's going to be good eventually.
Because we collectively dont care. There's more appetite electorally to oppress trans people than there is to reduce carbon footprint.
Missippi just got won on the back of an ad that suggested everyone is entitled to as much beef consumption as they want.
While there are many countries who have let the world down on climate change, America is quite the outlier. Since the gatekeepers lost control of the narrative post internet, stories that highlight American failings struggled even more than before.
I think its part lf human nature that no ruling class wants to accept responsibility and privilege and would rather find reasons to consider themselves victims. Consider the wailing over gas prices every year that America never answers with "maybe we should drive more fuel efficient vehicles".
It’s the same with the national debt. A lot of people are betting that they’ll be dead before it truly gets bad, and at that point it’s another generation’s problem. Though it does surprise me how many people with actual children and grandchildren are prepared to unapologetically saddle them with that.
It already is. Capitalism exists as a market solution to the distribution of resources. In an AI/SI-enhanced post-scarcity society, it simply ceases to exist.
This comment demonstrates a severe lack of empathy for Anthropic's shareholders. Not once in your comment did you think about how it would make the shareholders feel, you didn't give a single thought to their investment. Inexcusable.
Because the American population voted Trump into office. Twice. Now the reasons for why they did this, are probably a bit complex. But the reason why this is happening is simply because enough people voted exactly for this: no regulation of the market whatsoever (except for renewables, those need to be prevented at all costs)
Because we let mindless “environmentalists“ dictate our energy policies for the past 3 decades, and we didn’t invest in nuclear power like we should have.
I find these arguments very interesting, you know. A group of clueless ideologues decide some absurd policy somewhere in the past, and then, in the future, the same or very similar group of ideologues, claims that they should have some moral authority to impose new absurd policies to deal with the outcome of their old absurd policy.
Well, no more. It’s going to be hard for some decades, but the cycle must be broken.
So it's those environmentalists with little or no political power who made the decisions, not the fossil fuel companies with their well paid politicians?
The question is much broader than what's happening here. Like Amazon already uses more power than Ant + treats its workers like trash. So its not just an anthropic or AI thing.
In general, the answer to that is because there is no "we" that lets them. The "few" mobilize capital and influence and use it to fuck over the "many". This has been going on forever. Altman and Edison are the same bread.
Really need a bit more data to know whether or not what Anthropic will expend is bad relative what we expend today resource-wise. I'm inclined to measure it as a negative just based on what it does to people in the economy absent resource based externalities. But how much carbon are knowledge workers putting out into the environment between commuting, heating/air conditioning, and so on? Not a straightforward answer out there on the resource questions without digging much deeper.
I think the point is that data centres can be run with solar panels, so why aren't they? Why are they instead using so much natgas and other fossil fuels?
Because bureaucracy favours existing electricity consumers over new users, even if the new users are more economically valuable. Connecting to the grid is a nightmare, so every new DC is BYOG, which practically means gas turbines.
Because the sun doesn't shine 24/7, batteries aren't there yet and other than China no other country has the powerline infrastructure in place to feed the grid cross-country with the massive amount needed from one place to the other when one part of the country is cloudy and the other is not.
Yeah, electric is gradually showing up in construction, bottom up. Haven't seen heavy construction stuff myself yet, but plenty of smaller construction vehicles and heavy tools are increasingly electric where I live. A welcome development.
I don't see any environmentalists, agencies, corporate media crying about global warming and climate change. When someone in global south use a 50cc motor vehicle for his livelihood then that is a problem for planet Earth and global warming. Absolute hypocrites.
It's called an externality. It's like how people will buy a bigger car without a single thought to how much public space it's going to use up. Governments are supposed to be there to understand and compensate for externalities, but we've kinda stopped doing that.
To be fair largely the same people also said NFTs would solve global warming by driving up electricity consumption in order to give physicists an incentive to invent clean fusion power. They say a lot of stupid shit is all I'm saying.
Investors managed to just squeeze this t*d out onto the public markets just as the AI bubble is starting to implode. This was about one thing and one thing only and that was making $$$.
Given that is it any surprise that they didn’t give a $hit about the environment?
So, instead of emitting CO2 from coal they emit it from burning gas. So what exactly is the news here? Maybe they should celebrate also producing energy from "biomas", that is burning trees and corn?
Why all those companies, which have almost infinite resources are not investing in SMR reactors? That would make perfect sense for them. The technology is already here, it just needs scale to be viable? This would have at least some positive impact on the society - development of clean energy sources.
People who aren't in the energy sector don't realize how energy starved the planet is. There's literally deficits in nearly every segment due to chronic (30+ year) underinvestment. The last time a new refinery complex was constructed in the US was decades ago, natural gas power plants have more or less only been replacing ancient coal ones, and total electricity produced has been flat for around 20 years now. Global oil production has been effectively flat over the past 20 years, barely meeting GDP growth, and in fact has been decoupled from GDP since ~COVID due to a sharp spike in services driven growth.
> But renewables
Not enough, again. You could hypothetically start mass producing solar & wind turbines, but it _isn't_ enough. All that has been doing is replacing, again, ancient coal fired power plants. The world needs unprecedented investment in energy.
This is an American problem due to funneling everything into profits.
Australians have three hours of free electricity per day now. Charge your EV, charge your house battery, run your A/C at some crazy low temp, mine bitcoins. All free.
Per capita, Australia has 10x the number of residential batteries. [1]
In the last 4 years China has added the entire US grid worth of generation to their grid.
[2]
China add an entire US grid worth (1281GW) of RENEWABLES every 3 to 5.3 years.
The US could do these things if they wanted to. But they don’t. Like healthcare, education, prisons and do much more, profit is more important.
My Dad in Australia pays $0/kwh for three hours every day. That’s hard to beat.
>If its all profit taking, you would think it would be reversed?
Your logic doesn’t hold up.
The “excess” could be spent on grid upgrades, batteries, solar, etc (Australia), or it could be profits (USA).
Australia could have higher prices, but lower profits.
It’s why so much infrastructure in the USA is so dilapidated, extracting the “excess” as profit for a few decades instead of making stuff better.
The leaders of the current boom aren't even good Machiavellians. The huge irony, is that funneling everything into shortsighted profits, leads to less profits.
The strategy of the shortcuts profiting from AI + a bunch of US energy hacks, increasing energy inflation to communities, and raising a wave of protest in resistance is terrible. They could have built the datacenters, built the energy sources and had a great side hedge of energy sales to backstop temporary AI bubble burst setbacks. A high likelihood of profits no matter what over the long term - lots of options to mix the profit streams in various tax schemes too. I'm always surprised at how shallow the most publicised leaders are and I really shouldn't be at this point.
Yes. The United States managed to keep carbon emissions from growing much by wholesale switch from coal to natural gas electricity generation. But that locked us into paying for gas for decades to come, while i.e. https://en.wikipedia.org/wiki/Drake_Landing_Solar_Community could have had us carbon-free and saving money, as it is all simple materials (didn't have to wait for solar to get cheap). The "proper comparison" was over the last thirty years, when solar and wind were still scaling.
One upside is gas generation pairs well with solar, so if we cut tariffs (first boosted by Biden, not orange) we could see more solar.
Or, shit, don't even have to go that far. Over fifty percent of fossil fuel power generation is lost as heat. Could have just piped that to buildings instead of having every building in the North burning gas to heat water and air while a power plant just releases heat into the air.
Its more complex than that, and different regions are very different. Yes investment is needed, but coal use has peaked, and at least gas emits significantly less CO2 (and pollutants) than coal. Europe has 40% renewables, and China is making a lot of progress with renewables. The US is unusual in that it has large gas supplies, and a president who is trying to shut down renewables.
If demand was willing to pay more though, wouldn't we have invested in it?
One interesting thing is that data center power usage over the 2010s was flat - we kept getting more servers with more RAM, better cpus etc but the increased performance & capacity was all enabled on the same power supply - each new generation of hardware was more power efficient than the last, and demand wasn't growing like crazy.
>If demand was willing to pay more though, wouldn't we have invested in it?
The demand has always been there, but there are endless ways bad actors can stall projects (overtly and covertly) to protect their investment in the status quo. This drives the risk up, making any kind of large infrastructure project uneconomic. That and the endless red-tape that has been erected to protect established interests.
What do you mean wholesale price is dropping? We're sitting at more than double, almost triple, the wholesale price since 2020 (which had been flat for more than a decade prior). If we ignore the 2022 peak as an outlier, we're still well and truly trending upwards, with only a tiny reverse in the last year.
Projections are that price is likely to increase by 10-25%, especially as more coal is taken offline permamently. These projections look very conservative, and the footnotes say it's likely spot prices could double during peak periods (the key part that people ignore when they focus on how low the mid-day prices are, is how the peak prices have been soaring).
Excluding 2020, which I think we can all agree had unique circumstances, wholesale prices are at their lowest point in 10 years. I don't believe this is adjusting for inflation either.
>jumps from a low point in 2012 to another low point in 2020
It sure does. I used the AER report and missed a few years in that hastily assembled graph... but the point was that last year the prices were double to triple what had been the normal baseline (around $50) that had held until about 2016, and had been trending up since. I didn't have 2026 data in that report, so it's nice to see 2026 has been an exception to that trend so far. But that gigantic 20% drop mid-year looks like the result of government intervention around the fears of a fuel crisis. I wouldn't expect that to hold, and the AER certainly isn't.
Now if you want to get into the spicier discussion topic of what drove that price instability (and hence higher prices), look at what energy source appears in 2016 and starts gaining traction, per your source. Growing solar from 0% to 3% within a year was enough to drive "price" up by 50% in that same year. Astounding, and it wasn't until 2020 / 2021 with massive economic disruption, that prices came down at all... and then immediately after that disruption, prices soared even higher.
That's providing an excellent incentive to time-shift usage. If you're too lazy to do it, then you pay more. If you're not, then you pay less. I can tell you in the Pacific Northwest of the USA, time shifting doesn't get me squat, I'm certainly not getting free hours and prices are just skyrocketing.
Time shifting is definitely the rational course of action for now, and if everyone did that, it would level out the wholesale price... but it doesn't make up for a lack of investment in generation! It just kicks the can down the road a little (at great expense to every house that now needs a big battery), while still allowing massive wholesale price spikes which is what drives prices high on average.
Back to generation: In 20 years, Australia is producing about the same amount of power as it did 20 years ago. All those new renewables were simply replacing aging infrastructure. [0] And as the % of solar increases, so does the variability of day to day generation, not to mention the complete mis-match between peak solar hours and the non-solar hours when power is actually used.
We don't need as much power. Lighting uses 1/10-1/3 as much power. Insulation requirements have lowered space cooling needs. Other energy standards have lowered consumption of refrigeration, dishwashing, and clothes washing. (Edit: manufacturing has shifted to China...)
You don't need a battery to time shift. Program your A/C to run at full power during free hours. Get an all-in-one washer-dryer combo and set it to run then. Set the charge timer in your car to run then. That should be over 80% of your personal electrical load right there.
If solar is cheapest per kwh generation, then it makes sense it will continue to deploy until the fraction of free time exceeds the margin over gas. But as people take advantage of that margin, it won't grow and solar will continue to be the main source of new generation.
Well they did link a us source. I believe you are right about solar though in many places, dunno how much of the excess solar is benefiting consumers here though. My utility is only now in the last year starting to lower rates from something like 3x what it was when I moved out here >10 years ago. Of course burning down half the state and going bankrupt from the liability probably contributed to that.
narrator: you can build massive wind, solar, and battery plants and it would be more than enough. however the fossil fuel lobbies and nuclear bros would like a word with you
Makes me laugh how they’re using the thing to solve maths problems but can’t manage to put some of their trillions of investment into solar on data center roofs…
With how cheap solar power and batteries are it is utterly insane that gas and coal are not either flat illegal for new massive loads like this, or at least taxed so heavily that solar and batteries are the right financial choice
New electricity capacity is almost all renewables (90%+) : solar and wind.
Renewables are completely dominating. New coal and gas capacity is almost non-existent because it is too expensive in comparison.
Even renewables + the extra cost of battery backup are much cheaper than building a new coal or gas plant.
Coal and gas are only competitive when the plants are already built, and even for pre-existing its at the tipping point where a new-solar build is cheaper than even supplying coal/gas to an existing power plant.
How so? If you look at the total energy mix, renewables are aggressively taking over. This has more to do with (lack) of planning of infrastructure to power datacenters.
The problem with a purely financial accounting is it ignores environmental and social costs entirely.
Everyone wants to make quick buck with AI powered solutions. The real problems like climate change will remain on back burner with this AI frenzy. There's no reward to solve climate change in short term unlike math problems.
We are going to accelerate climate change with AI.
People are pushing forward with an illogical, destructive ideology, saying that somehow it will solve our problems AFTER it hurts/destroys us.
https://www.theverge.com/tech/1000140/jensen-huang-nvidia-ai...
Nuclear? They would simply consume more, just like an additional lane on a highway simply leads to more traffic.
The solution won't come from them. It might come from us, but it won't be pretty.
Even just on the energy used problem, consider how often people fly, drive, buy stuff shipped around the world sometimes multiple times, etc.
It's just more obvious that there isn't much money to be made from power-generation. With LLMs they have somehow deluded themselves into this being the next money generation opportunity.
Why bother developing a more efficient solution if you can just cheaply chew up 10x more resources? Just like how software has gotten worse as computers got better.
We'll probably reach a peak in fossil fuel consumption over the next years, mostly for demographic reasons, then we'll see if we can keep doing tech the same way while consuming less fossil fuels worldwide.
Anthropic employees and CEO probably care about climate change vocally, but before the IPO they couldn't reduce their emissions, and now it might even be illegal for them to do so. Reducing CO2 emissions massively and quickly would mean destroying their business, investors wouldn't let them do that. Good alternatives take time to develop and cost more.
The reward to "solve" math problems is not much bigger than the reward to "solve" climate change on the short term. It's just that they're mathematicians so they care about maths, solving the Riemann hypothesis on paper didn't change the world much, people need to actually understand how it was done. The same way having the solution to climate change on paper won't change the world much, people need to understand and accept the solution to do it. If you have a solution to climate change tomorrow, people will congratulate you. But that's it, implementing this solution will be the hard task. And energy and climate change are very different from maths because there is a social aspect to it. Whether people accept or refuse the solution to the Riemann hypothesis doesn't matter; while if people reject your solution to climate change, they just won't implement it, which means your solution wasn't a real solution as it didn't solve the problem.
Anthropic employees (and other employees of AI companies) could surely do way more for energy and climate change. Maybe they just don't have the qualified people for that. It's a full time job to understand energy and climate change, I spent 1 year full time on it reading hundreds of papers, and that's really not enough.
Investors have no say in Anthropic's decisions. Their governance structure ensures that.
Even public companies do not, despite popular perception, have a fiduciary duty to maximize "shareholder value" at the expense of everything else:
Burwell v. Hobby Lobby Stores, Inc. - https://www.law.cornell.edu/supremecourt/text/13-354
> While it is certainly true that a central objective of for-profit corporations is to make money, modern corporate law does not require for-profit corporations to pursue profit at the expense of everything else, and many do not do so. For-profit corporations, with ownership approval, support a wide variety of charitable causes, and it is not at all uncommon for such corporations to further humanitarian and other altruistic objectives. Many examples come readily to mind. So long as its owners agree, a for-profit corporation may take costly pollution-control and energy-conservation measures that go beyond what the law requires. A for-profit corporation that operates facilities in other countries may exceed the requirements of local law regarding working conditions and benefits.
Who killed it?
The problem is that every time we add cleaner forms of energy to the mix, we find ways to use that extra energy. On top of all the fossil energy we continue to use.
The only solution is to stop using fossil fuels.
Safe nuclear is just too expensive.
3 Mile Island, Chernobyl and Fukushima are three guys that come to mind.
You think a data center in your backyard isn't popular? Try throwing in a nuclear reactor as a package deal.
There is stuff that decelerates climate change.
It seems like they have solved their power consumption...? AI probably told them gas is the most cost efficient way.
I suppose it feels very _optional_ compared to those other things. A lot of tokens are gonna be spent optimising ads, for example.
BTW I have a Nothing phone. I don't upgrade it with every new version.
But of course the whole thing is backwards; both smartphones and LLMs have some of the highest utility per watt (including embodied power) of all goods and services worldwide. Plenty of much greater waste to go after, but it's not as sexy to complain about vacations or sportsball or fashion or home decoration or DYI arts & crafts or ... as it is about smartphones and data centers.
It’s not like they have batteries that are charged from the grid which be sourced from renewables.
I am hoping at some point deep geothermal will help with this as it works anywhere and would provide predictable 24/7/365 base power
So what? Is getting people to form friendships, even relationships with AI more important than the friggin' environment? Is it so important to have office drones sending emails written by LLMs in order to receive replies from LLMs? Is it so important that everybody can write a TODO list app?
I (and next generations who will be the most affected) should not have to deal with the shit because some rich people got to shit everywhere as they could pay higher taxes.
Can we regulate this? Sure. But why regulate this specifically, especially when the public/consumers don't seem to care?
I do want to see them using less power, or not being built at all frankly. But that claim seems a bit extreme.
I put this in a different thread I believe, but China burns enough coal to run 300+ 2-gigawatt factories. We have larger fish to fry when it comes to energy use, our focus on data centers specifically doesn't correlate with their percentage of the problem.
The original example was overfishing. Each fishing boat has little effect on the overall fish population. Regardless of what the others do, each one ends up with more money catching as many fish as it can. But when they all do that, then the fish population collapses, which is not the outcome any of them prefer.
When they cooperate instead by regulating how many fish each can catch, they can avoid population collapse and everybody's happy.
This is why we regulate things. In the other direction it's why we pay taxes for public goods. Solving these collective action problems is one of the main purposes of government.
Because everyone will be affected, informed and uninformed alike? Also, it needs to be regulated specifically because consumers don't care.
It's really not that complicated. We need to course correct when the "free market" does the wrong thing with potentially catastrophic consequences.
If you're proposing there is no choice, who is compelling those using AI to use it, and why?
Ragnarok is upon us!
My argument wasn't that regulation should never be used to make meaningful gains in environmental damage.
My point was that throwing regulation at a single energy use in one area of a single industry is inefficient, won't make a meaningful impact on total energy use, and focuses regulator attention away from much larger change they could be making.
Because you seem to be making the argument simultaneously that they should and shouldn't
There are much bigger dish to fry if we want to regulate energy use down or require cleaner energy sources.
"Just regulate" as the OP comment is a flippant use of regulation targeting a specific use of energy in one area of a single industry. That's inefficient at best and won't make meaningful change.
I doubt Anthropic has that time right now considering the amount of stuff they're actively doing.
Whether or not it's the right thing is a different story, but just imagine the realities and the time sink of actually setting up solar for a data center powerful enough to serve stuff like a frontier model.
If I decided to ignore laws/regulations when building/repairing my house, I don't think a court would be interested in "I'm in a rush, so didn't have time to get planning permission etc".
Perfect reason to slow down the frontier progress, like Dario, Sam, and Elon all spoke up and said was a good idea...
This is, unfortunately, just the next chapter of the same story.
More unfortunately, it feels as if the only way to change the direction of the story is for there to be real, actual, visceral consequences for a majority of the people that are powerful enough to continue pushing in the current direction. Fatally unfortunately, said powerful people will be the last to feel real, actual, visceral consequences. There is a lot of 'worthless human shield' sitting in front of them that has to be ploughed through first.
Doubly ironically, maybe it's best if AI does destroy humanity and let the Earth find a more sustainable balance of life.
It will be interesting to see, if AI does destroy humanity, will it let the Earth find a more sustainable balance of life, or will it accelerate the domination of the natural world in pursuit of its inscrutable goals? (likely to its own suicidal detriment, given it is based on learning from humans)
the real bill comes due in about 4-5 years :)
They go for the cheapest option, even if they've had practically infinite money until now.
No, because solar panels are notoriously hard to scale, especially in high density applications like datacenters. Otherwise it's like saying "my calculator runs off solar power, so why can't they make a car run on solar power? Clearly it's because Ford/BYD not giving a f"
Before you respond "it will be too slow", the urgency here is entirely artificial – there's no reason we need to be going this fast.
What makes you believe that time is irrelevant? If AI allows for breakthroughs in mathematics, science, engineering, computing, and health care—and it does seem to be delivering on this promise so far—what in your estimation is the net cost to humankind of delaying those breakthroughs by even a year? It's not zero, is it?
More than anything though, it doesn't matter what we in tech think because we have failed to convince the general public that any of this is a good idea.
Well done humans, well done. Slow clap.
Math, maybe. Science, where? Computing, anything measurable? Health care, where?
FYI, pre LLM techs don't count. Ain't nobody building $3tn worth of DCs for AlphaFold.
Or it might result in investing trillions of dollars and decades only to discover that there are only marginal gains to be had. There is no guarantee that throwing money at Problem X results in any kind of solution or that solutions to specific problems are even possible.
This is aside from the sibling commenters' very good point that this view assumes there are no concrete benefits or advancements to be made that depend on AI, which seems very unlikely at this point. Unfortunately, with or without AI, the problems we're facing socially, ecologically, medically, and scientifically are increasing. We can either use AI as a tool in the hopes we can at least delay the most serious consequences or just let them overtake us. Fossil fuels and climate change are just one of a million very pressing issues, despite the over representation it gets, and it seems highly likely that the benefits here exceed the costs. Addressing a downthread comment about labs saying we need to "slow down" - they are saying this for self-serving reasons and in their quest to ensure AI stays gatekept, not because they are worried about environmental impacts.
The problem seems to be that they want the power online quickly and they can't get solar quickly. The article mentions the inefficiency of their generators. It is costing them quite a bit extra to generate power like that, but that's their only option of they want the power immediately.
It would be nuts to do this long term. You might argue that they don't give a f about the environment , but most people will agree that they give a f about cutting costs.
You just have other issues like needing to get transmission and storage capacity built, which takes forever because of planning review, environmental assessments and NIMBYs.
Some. Hence "why can't they make a car run on solar power" isn't the obviously dumb question you make it out to be.
2. There probably are non-zero amount of datacenters out there that runs off solar power. It's just not commonly practical, just like with electric cars.
3. The main point of the comment is that scaling challenges exist, and you can't take a residential use case and extend it to a datacenter. The same applies for a car, even if you could do it. A solar powered calculator is easy, just slap a tiny solar strip and it just works. A solar powered car is far harder to pull off. If you don't work from home for instance, you'll need a battery system to store the energy when you're not home.
If you can't figure out how to get renewable power to a datacenter, but you can easily figure out how to get it to the consumers of a datacenter, and you're talking about a technology that scales down to small groups well and individuals only a bit less well (but with constantly working agents utilizing compute, better than we once imagined), what's the reasoning behind datacenters at all?
If datacenters can make a case for massive increase of demand in the future, they're simultaneously making a case that individuals will have enough demand to saturate AI located closer to them or on premises. If anything, that should be a call for p2p software to help people share or borrow excess compute with and from their neighbors in exactly the same way that home solar panels use the grid.
Datacenters are also central points of failure. Distribution is resilient.
I know meta and their ilk are trying to cut deal for nuclear power. I'm sure we can imagine they will take safety, the environment, etc very seriously too.
Are you high?
I think all of the major ai labs are running on gas it's ridiculous
Meanwhile Chinas grid is super healthy and has a tremendous amount of renewable energy... This is what happens when you don't pay into infrastructure as a civilization or require corporations to participate or have accountability for the future of the people they serve.
Yes I agree these data centers should be using more renewables, but the China comparison is odd given that they burn enough coal to run 300+ 2 gigawatt data centers.
Texas is also the #1 state in solar and wind production as well as growth rate.
Here's a quote about a data centre currently being built near Edmonton, Alberta.
------
"Utah State University physicist Robert Davies was hired by SLCN to conduct a preliminary technical assessment. He said if the Wonder Valley project goes ahead it would release the equivalent of 23 atomic bombs of energy every day, though he noted the comparison has its limits."
Source: https://www.cbc.ca/news/canada/edmonton/wonder-valley-slcn-c...
------
This is not the total energy being consumed by this data centre. It's merely the waste heat it will release to the surrounding environment.
Here's the energy consumption:
------
"At full operation, it would consume energy at a continuous rate of 18 gigawatts or about 1.5 petajoules per day and emit about 30 million tonnes of carbon dioxide a year – more than four per cent of Canada’s total emissions, said Davies, whose assessment was based on publicly available information."
------
That's the equivalent of 1.66 million Canadians in carbon emissions, or 1.6 Edmontons. This is just one data centre. Many more are going up across Canada, and Canada is both late and slow to the buildout.
That's the kind of scale we're dealing with at a time when, as a world, we are trying to reduce our carbon emissions.
From my perspective, that is not quite the correct way to put it. Because honestly we are not even trying.
---
I find those "Olympic swimming pool type" comparisons very tiring, to be honest. They certainly don’t help my grasp anything. "23 atomic bombs of energy" yeah okay
Cheap train network is a proper pipe dream, not possible. Much better to have fully self driving electric taxi cars (so not that joke with driver constantly needing to be aware if your tesla wants to steer you to standing ambulance / firefighter trucks on highway, I mean no driver setup to push prices to very low levels).
While I can't really visualise energy at all, this does make it seem like more than the 0.5% of world electricity consumption that it actually is.
While it is currently cheaper to build renewable energy capacity than most fossil fuels, as far as I can tell, the reason most data centers are not doing so is because they can't, All of the renewable infrastructure being produced is currently being deployed.
The question they should be being asked is when will they be switching to renewables. The height of the plot is not the important thing. It is the area underneath that matters,
"When are you going to switch to renewables?" has proven to be a useless question. Companies will say whatever is expedient, and then they'll buy credits or whatever green washing instead of actually reducing consumption. If you let them double their energy consumption by adding renewables they'll just use double.
Most of the worlds energy use is not electrical.
It is a large amount. I do not deny that, I just want people to consider it in terms that are true. Every person who I have asked how much they think it is based upon the coverage they have encountered has massively overestimated.
A common response is that considering the concern reported in media they estimate data centers use 10-15% of electricity. How can we have an informed conversation about the relative importance of things if people have been misled about the relative scale.
Is there a difference? Where else would the energy go?
Is it not possible to use some of that for basically free local district/water heating?
Is that feasible at all?
What yield?
Little Boy (the type of bomb dropped on Hiroshima) was 63TJ
The IEA is projecting 500TWh (57GW ) by 2030 so it might be consistent with their figures.
Now it has a bit at the bottom This story and its headline have been updated to provide further context about the preliminary nature of Robert Davies’ assessment and align with our journalistic standards and practices
I was thinking they were talking about current usage. Not guesses of how much something might use.
The only other source I found was https://majorprojects.alberta.ca/details/Wonder-Valley-AI-Da...
Indicating the initial build would be 1.4GW.
I'm not sure how much size matters of individual data centers apart from their contribution to the total. If they sum to the same amount, how much does it matter.
If it's true that you can build something of that size without an environmental impact report then there's something pretty wrong with the laws there.
Meanwhile codebases get exponentially more bloaty and buggy. The internet gets filled with more and more disinformation. Students get worse and worse.
And the growth everyone is depending on is also so cleary destructive. This is not sustainable and it will not be the solution.
Also AI data centers weren't responsible for decades of people burning natural gas to heat their homes without investing in insulation nor the tar sands oil which is really screwing the climate.
Good luck getting that past the NIMBYs
We could build The Great Pyramid from TNT byt we don't as its a stupid idea.
Saying this comparison has its limits is an understatement. Energy released by different atomic bombs can vary across different order of magnitudes.
Is this supposed to sound very impressive? It genuinely doesn't to me.
I know atomic bombs are scary, but that's because they explode very quickly in a very small space. (Radiation aside which I guess was not the point of the comparison)
If you spread that energy release throughout a whole day and a bigger area, letting the wind slowly take the energy away, then it doesn't sound scary at all.
Using nuclear bombs as the measuring stick is a cheap rhetorical tactic.
This is wrong.
The energy consumption of data centers is irrelevant: our complete failure to deploy the technology we've had for the past 40 years, or the last 5, to produce carbon-free energy is the real issue.
I'm rather tired of using energy consumption as a convenient bludgeon in the OmniCause, by people who have typically steadfastly opposed what we could've been dong this entire time and can be doing now.
Build more datacenters. Permit more solar and wind to power them. The money is flowing so I'm pretty sure we can afford some nuclear plants to backstop the baseload as well. This is all ridiculous: we should've ended carbon fuel dependence for stationary power in the early 2000s.
At least from the news I've seen across the ocean, I was under the impression that the USA's grid was already stretched way too thin because of their lack of investment in renewables, and many people are against data centers because electricity is already too expensive for normal people. Or was that just news cycle hyperbole?
> and many people are against data centers because electricity is already too expensive for normal people
Yeah, exactly. That's why Anthropic is generating more power for itself...
That doesn't follow though. If USA's grid is stretched too thin, it's because of lack of investment in power generation. Not renewables specifically. In fact, renewables are still the most roundabout/difficult way to go about addressing that. And as unfortunate as it is that gas is the economically-best option (I'd vastly prefer small nuclear), if data centers are forcing investment in new power generation, that benefits the grid.
Yeah I’d prefer various fantasies that don’t exist too.
Not fantasy.
That's indeed a way to put the active dismantlement that happened.
More than 1/2 of my energy generation at my home is grid sourced as renewable. I have a HA integration that tracks this semi-realtime. The cost is ~14c/kw ... about half of what the EU pays (on avg)
But, for counterexample, my power bill in the US is included with my water, sewer, irrigation, and trash collection. Monthly total is between less than 1% ($100) in the summer to around 3% ($300) of a median monthly household income ($10,000) in a metropolitan of about 1.5 million people. Here, the electric grid is owned by the city, not privately operated.
Generally speaking, bigger cities have a high cost of living. And the same can be said about bigger cities in Europe.
But having lived in both places, I find that life is so much harder in Europe, in terms of cost vs income, and in what your money can buy.
Is the media lying to you? It's stretching to focus on specific examples. Sensationalism is profitable. And often times media is colored by corporate or government interests.
Hope for solar datacenters in the Middle East?
Although I guess for current AI workloads, it could be viable, since the additional latency of sending your (small, text only) prompt around the world for processing is little compared to the time taken to process (whereas for other stuff being close to users matters more)
You could build batteries, charge the batteries during the day, run the DCs off it at night.
SunCable / AAPowerLink is likely going to be connected to the Darwin grid, which is currently not connected to the main East Australian grid; but building a connection to the main Australian grid would be a lot cheaper than undersea HVDC to Singapore. Once that happens, there would also be the potential to use pumped hydro (including Snowy 2.0) for storage.
Given this year’s events [0], even Sahara seems more likely.
But you’d still have to build it out from ≈scratch, including cooling, internet connection infrastructure, and maintenance, so I’m not sure what the time horizon on amortization would look like, even with the cheap solar power.
[0] https://www.reuters.com/world/middle-east/amazon-cloud-unit-...
Obviously security is a concern which limits the number of locations, but there’s still plenty of places with stability, security, land, and the lack of taxation and tarrifs that the US puts on energy.
Globally, places like Australia and India should be more competitive for solar than the Sahara, although both Australia and India have a lot of dirty power generation to replace as well.
I find the concept of agrivoltaics really exciting for climates such as found in Spain.
https://www.google.com/search?q=agrovoltaics+in+spain+resear...
That is simply not workable, and the suggestion needs to die because it sucks motivation from solutions that can work in a year or two.
Short term plan is much faster and cheaper. Therefore it’s a much better plan.
China add an entire US grid worth (1,281GW) of renewables every 3 to 5.3 years. Why wouldn’t the US do the same and have the problem well in hand in 5 - 10 years rather than waiting 20+ for a single nuke?
*runs*
But then of course that's another front-page submission with its own long thread of commenters feeling smart for calling it stupid, despite recently becoming a worked out plan and addressing exactly the (subset of reasonable) things people are complaining about here. In fact, we had those a week or two ago.
I swear the tech space turned into thought-space NIMBYs. "I don't understand any of it, but something about it pattern-matches to something I don't like, therefore it's stupid and people behind it are evil". In this thread, and in every other thread recently, too.
To deploy 1 GW of capacity, apparently 100 Starship launches would be needed.
On earth, 23 GW are currently under construction, 100-132 GW is already active, and about 100 GW should be added by 2030.
SpaceX's CFO supposedly cited 2029 as the timeline for multiple GW of compute in operation in space. Musk's vision seems to be launching 100GW/year or 1 million ton of satellites.
This would require 10k launches of Starship per year under ideal conditions, or 5k if Starship Block 4 can double the payload capacity to 200t as planned. This would mean 14 launches per day.
It does sound like a lot. Starbase Louisiana, at a $100B buildout, is planned to have 12 launch pads that could handle 30 launches per day.
Obviously there's much more needed than some launch pads. Anyway, that's the numbers and projections I found.
From a capacity perspective, if SpaceX remains successful in scaling up their Starship rocket, it seems putting AI compute into space could at least realistically happen at significant scale.
Should SpaceX solve the problem of requiring fossil fuels to launch (doable) and the burning things on re-entry (harder) I don't oppose solar data centers. But without those, it's just externalizing the environmental cost to future generations.
This is pretty bad-faith. What is the point of your comment? You’re sowing hate and attacking a nondescript group, for what? What does that achieve apart from being hateful?
Yours is one of the few usernames I recognise on HN. I remember seeing some comment from yours (I think it involved Star Trek) which I found interesting at the time and somehow the username stuck with me. Since then, every time I bump into your comments they have gotten more hateful and generally worse and more tribalistic takes.
It genuinely saddens me. Especially as you have in your bio “I'm constantly looking for opportunities to do something actually useful for humanity”, but your words reveal the opposite.
I’m not going to be specific about past posts because frankly I no longer remember their content, only the disappointment I felt. But I can take this one as an example. In addition to every thing I mentioned, I find it a pretty bad take to claim that adding more shit to orbit improves humanity. We need less crap up there, not more. I also find it nonsensical to defend (if we can call it that; you just attacked detractors with no explanation) something that is purely theoretical (which it objectively is; it doesn’t exist yet).
Anyone who has paid attention in the last decades will see all the unkept promises. You can see them in this very submission. Blockchain and AI were full of proponents making the argument that “all this energy consumption is good, actually, because it’ll increase investment in renewables”, but obviously what we’ve seen is an even bigger spend on fossil fuels which are exacerbating the problem. Believing the next promise from proven liars who make everything worse every time is not a smart strategy.
Anyway, I’m not entirely sure why I felt I should write to you specifically. I guess I once saw the possibility of something better and am hopeful that being respectfully called out might help you to reflect. Those of us capable of introspection (which I believe includes you) sometimes need a little nudge to get back on track to the person we want to be. I know that’s true for me. Sometimes I help others, other times they help me. I hope this helps you.
I know there are smart people on HN, and I don't believe most of what (person that shouldn't be named) says. But I agree with OP, the amount of hate anytime the topic comes up is frustrating. And the arm-chair aerospace engineers are are the most annoying group.
Of course, comments and downvotes like yours were very much expected. On HN I’ve come to anticipate lack of nuance and the belief any disagreement or criticism is hateful and looking for a fight.
I can only try to counteract that one post at a time, though I can’t say I’m feeling hopeful that the bother is worth it.
1) Musk retches up another one of his Hyperloop-style dumb-ass ideas whilst patting himself vigorously on the back
2) His companies grovel hilariously to backfill the reasoning for why this is a Good and Smart idea, actually, while actual scientists put on a wtf face (“worked out plan” goes here)
3) The technorati follow the their glorious thought leader off a cliff
So PSE's consumer rates are scheduled to go up 30% just in the next three years. It's not going to slow down if AI is bidding up every power plant. And in states without the mandate, it just makes keeping gas plants running more cost-effective, because power is fungible, so there's still a net increase in fossil fuel generation despite said mandate.
This is pretty standard capitalist propaganda.
- "But I don't want this toy here!"
- "It must go here, because over there is a big crocodile that will eat my toy."
- "Oh okay then..."
Seriously, your "standard capitalist propaganda" as stated is a 3-year-old argument. Any reasonable adult will just say "then don't build it".
The labs have failed entirely to predict this opposition and it's an existential risk to them. They have not been able to convince people of the urgency of AI build-out and certainly aren't assuaging them of concerns about the costs.
0: https://hugotunius.se/2026/11/14/of-course-people-hate-data-...
I've been trying Mistral Vibe side by side with Claude over the past few days. I had very low expectations going in. But so far I've found that for regular chat questions there doesn't seem to be a meaningful difference between Opus and Mistral Large 4. Smart enough is smart enough. For deep research and coding the situation is different of course, where for some things I still need Opus. But I really don't need more intelligence for pretty much anything than Opus 5.5 gives me, and I often need less. So if Mistral could reach that level somewhere in the next year I could imagine moving over entirely.
This is a bit of a virtual accounting, right? I mean this energy would be consumed anyway, and people who would use this green energy now have to use coal/gas. Still better than building your own gas powerplants, of course.
Then at least wind farms or solar can muscle in and displace those gas generators if the economics work out
How many years does it generally take to build a nuclear power plant?
Yes, everyone I know including myself at times, is just cheating. Doing a similar amount of work as before with less effort/time involved. There's not all that much productivity gains, maybe for a select few who practically live in a state or mania (burnout inevitable).
I dont see it as all that positive for anyone long term. Im constantly having to relearn things, never retaining anything I learn with these things, and forgetting everything I used to know. Its literally causing a cognitive diabetes of sorts. Like someone who drives their car everywhere loses muscles, gains weight and eventually gets a metabolic disease, im certain this happening to our brains. Im 100% certain of this. Its a net negative all around, there's no such thing as a free lunch.
Even if everything goes smooth with the progression of this technology, best case scenario is we have massive unemployment and people with jobs make 20% of what they did because they're now practically just a meat proxy, the equivalent of a conveyor belt operator. Suddenly find ourselves working for a dozen or so Oligarchs, tons more crime, zero uppward mobility. Thats best case scenario. Im not sure why everyone is so excited about this.
2. They said "want to", not "use". And of course even less "directly pay to use"
People take free shit even when they don't really need it, it doesn't measure demand.
More people are willing to pay for netflix or youtube premium than AI
And didn't Google admit that their AI summaries are biting into their ad business (causing their recent drop), meaning a huge number of users are deriving value from it? Possibly billions.
And then there's all the white collar workers sending emails. SO many people are suddenly sending grammatically correct internal emails now, AI policy or not.
(Of course, the recipients of those emails are not as happy, since someone who used to send shitty emails now instead write shitty prompts that don't actually ask the LLM to address the real issues. But that still counts as demand)
> More people are willing to pay for netflix or youtube premium than AI
And even more people breathe air. In both cases the analogies don't work.
You can't infer demand form "free users" because just from microsoft and meta shoving AI in every single crevices of their free tier software you already get like 2b recurring users even if none of them use/asked for/care about ai features.
If the next windows 11 update turned the mouse pointer into a turd would you say 1b people demanded a turd mouse pointer?
I think people intrinsically don't give a fuck either way, they just go with the flow and rationalize their behaviour so that their definition of "good" warps to fit what they happen to be doing. Otherwise lynch mobs, genocides, and Peter Thiel would not exist.
Maybe before doing absurd comparisons between LLM use and murder, let's take a look at the energy use of people's vacations, or sportsball and concerts.
At least LLM use (and murders) are productive activities.
if the industrial revolution never happened or we never invented the wheel or agriculture, would the world be better off?
this anti-tech nonsense from people is hilarious. You WANT modern medicine I promise you that. If it wasn't for soap, dentistry and a bunch of other modern conveniences, we'd all be at each others throats trying to survive on what nature can 'provide'.
If not for the labor movement, the history of strikes and struggle and the resulting transformation of social conditions, humanity would be worse off from the industrial revolution. Ten year olds would be working twelve hour days in factories with zero attention to safety.
The tech itself is just a device for changing the social conditions. The direction of change is what the struggle is about. If we leave it solely to our oligarchy, they will make sure to use it to further entrench their interests while simultaneously reducing the bargaining power of labor.
I've elaborated this critique from some other angles here if you're interested: https://news.ycombinator.com/item?id=50577833
They polluted the entire world and somehow AI equalising the equation is an issue. There's a reason developing countries are far more amenable to include AI in their processes. While these people have got so fat and thick minded from prosperity they cannot care less.
Housing getting more expensive is a consequence of supply constraints. Immigrant labor could totally help build houses; we just have zoning and lots of other hurdles that make building hard.
^ reporting on the evidence
[The TL;DR for this claim is that moving to say the US or Australia from a less wealthy country is the one of the most carbon intensive things a person can do].
Of course all this is bound up in the economic growth that migration brings and how you feel about that is a values question.
It's not really within the overton window because the people who care about carbon don't want to restrict migration, and the people who want to restrict migration don't care about carbon.
It's a you get what you deserve kind of thing and the people against immigration just want to close the border and shirk responsibility.
What would be the upside for the environment? There is no burning of the planet that it would prevent such as some comments here said. What precisely one could name as a realistic consequence of using more gas and coal at that scale over the next years, I am not sure.
There is also the question whether process improvements enabled by AI can offset that.
While there are many countries who have let the world down on climate change, America is quite the outlier. Since the gatekeepers lost control of the narrative post internet, stories that highlight American failings struggled even more than before.
I think its part lf human nature that no ruling class wants to accept responsibility and privilege and would rather find reasons to consider themselves victims. Consider the wailing over gas prices every year that America never answers with "maybe we should drive more fuel efficient vehicles".
I find these arguments very interesting, you know. A group of clueless ideologues decide some absurd policy somewhere in the past, and then, in the future, the same or very similar group of ideologues, claims that they should have some moral authority to impose new absurd policies to deal with the outcome of their old absurd policy.
Well, no more. It’s going to be hard for some decades, but the cycle must be broken.
In general, the answer to that is because there is no "we" that lets them. The "few" mobilize capital and influence and use it to fuck over the "many". This has been going on forever. Altman and Edison are the same bread.
And that is not including the efficiency gains yet to come.
Really need a bit more data to know whether or not what Anthropic will expend is bad relative what we expend today resource-wise. I'm inclined to measure it as a negative just based on what it does to people in the economy absent resource based externalities. But how much carbon are knowledge workers putting out into the environment between commuting, heating/air conditioning, and so on? Not a straightforward answer out there on the resource questions without digging much deeper.
Why are we letting you use the internet? Better to stick to farming.
Why are we letting you use lightbulbs? Go to sleep at sundown!
I rarely see farm equipment, but I have noticed electric construction machinery in Scandinavia. (Excavators and so on.)
In US I think over 50% is fossil fuels, if all the GPUs were in Europe the situation would be a bit better I believe.
This is going to quickly become similar to worrying about how many trainloads of coal it takes to run everyone's WiFi router.
It costs less -> we use much more -> total resource usage increases
It doesn't excuse anthropic one bit, esp considering Dario's flowery message about helping the world with Ai.
Given that is it any surprise that they didn’t give a $hit about the environment?
Why all those companies, which have almost infinite resources are not investing in SMR reactors? That would make perfect sense for them. The technology is already here, it just needs scale to be viable? This would have at least some positive impact on the society - development of clean energy sources.
> But renewables
Not enough, again. You could hypothetically start mass producing solar & wind turbines, but it _isn't_ enough. All that has been doing is replacing, again, ancient coal fired power plants. The world needs unprecedented investment in energy.
Source: https://www.eia.gov/electricity/data/browser/
Australians have three hours of free electricity per day now. Charge your EV, charge your house battery, run your A/C at some crazy low temp, mine bitcoins. All free. Per capita, Australia has 10x the number of residential batteries. [1]
In the last 4 years China has added the entire US grid worth of generation to their grid. [2]
China add an entire US grid worth (1281GW) of RENEWABLES every 3 to 5.3 years.
The US could do these things if they wanted to. But they don’t. Like healthcare, education, prisons and do much more, profit is more important.
[1] https://e360.yale.edu/digest/australia-home-batteries
[2] https://www.bloomberg.com/news/articles/2026-01-28/china-s-f...
How much do you pay for your grid electricity?
>If its all profit taking, you would think it would be reversed? Your logic doesn’t hold up. The “excess” could be spent on grid upgrades, batteries, solar, etc (Australia), or it could be profits (USA).
Australia could have higher prices, but lower profits.
It’s why so much infrastructure in the USA is so dilapidated, extracting the “excess” as profit for a few decades instead of making stuff better.
The strategy of the shortcuts profiting from AI + a bunch of US energy hacks, increasing energy inflation to communities, and raising a wave of protest in resistance is terrible. They could have built the datacenters, built the energy sources and had a great side hedge of energy sales to backstop temporary AI bubble burst setbacks. A high likelihood of profits no matter what over the long term - lots of options to mix the profit streams in various tax schemes too. I'm always surprised at how shallow the most publicised leaders are and I really shouldn't be at this point.
https://www.iea.org/reports/energy-and-ai/energy-supply-for-...
People don't get how cheap American gas is. In Germany, Russian gas was twice as expensive as German coal. That was considered cheap:
https://www.ffe.de/en/publications/merit-order-shifts-and-th...
In the US, gas is cheaper than coal. There's no political will to stop this.
One upside is gas generation pairs well with solar, so if we cut tariffs (first boosted by Biden, not orange) we could see more solar.
One interesting thing is that data center power usage over the 2010s was flat - we kept getting more servers with more RAM, better cpus etc but the increased performance & capacity was all enabled on the same power supply - each new generation of hardware was more power efficient than the last, and demand wasn't growing like crazy.
AI has broken that trend.
The demand has always been there, but there are endless ways bad actors can stall projects (overtly and covertly) to protect their investment in the status quo. This drives the risk up, making any kind of large infrastructure project uneconomic. That and the endless red-tape that has been erected to protect established interests.
Wholesale prices are dropping, my provider just sent an updated price sheet with a lower price per kWh.
Is this a US problem only?
Projections are that price is likely to increase by 10-25%, especially as more coal is taken offline permamently. These projections look very conservative, and the footnotes say it's likely spot prices could double during peak periods (the key part that people ignore when they focus on how low the mid-day prices are, is how the peak prices have been soaring).
[0] https://i.imgur.com/zT2o5ui.png https://www.aer.gov.au/publications/reports/performance/whol...
https://openelectricity.org.au/tracker/timeline?range=all&ta...
Excluding 2020, which I think we can all agree had unique circumstances, wholesale prices are at their lowest point in 10 years. I don't believe this is adjusting for inflation either.
It sure does. I used the AER report and missed a few years in that hastily assembled graph... but the point was that last year the prices were double to triple what had been the normal baseline (around $50) that had held until about 2016, and had been trending up since. I didn't have 2026 data in that report, so it's nice to see 2026 has been an exception to that trend so far. But that gigantic 20% drop mid-year looks like the result of government intervention around the fears of a fuel crisis. I wouldn't expect that to hold, and the AER certainly isn't.
Now if you want to get into the spicier discussion topic of what drove that price instability (and hence higher prices), look at what energy source appears in 2016 and starts gaining traction, per your source. Growing solar from 0% to 3% within a year was enough to drive "price" up by 50% in that same year. Astounding, and it wasn't until 2020 / 2021 with massive economic disruption, that prices came down at all... and then immediately after that disruption, prices soared even higher.
If you graph prices in constant dollars, it looks... fine: https://chatgpt.com/share/6b02e01f-cf94-83ec-bc92-9599d588fc...
The price today is roughly the same as what it was in the year 2000 and has only ever been cheaper for short periods of time.
If trends continue, electricity will be at an all-time lowest cost within a year.
Back to generation: In 20 years, Australia is producing about the same amount of power as it did 20 years ago. All those new renewables were simply replacing aging infrastructure. [0] And as the % of solar increases, so does the variability of day to day generation, not to mention the complete mis-match between peak solar hours and the non-solar hours when power is actually used.
[0] https://openelectricity.org.au/tracker/timeline?range=all&ta...
You don't need a battery to time shift. Program your A/C to run at full power during free hours. Get an all-in-one washer-dryer combo and set it to run then. Set the charge timer in your car to run then. That should be over 80% of your personal electrical load right there.
If solar is cheapest per kwh generation, then it makes sense it will continue to deploy until the fraction of free time exceeds the margin over gas. But as people take advantage of that margin, it won't grow and solar will continue to be the main source of new generation.
> with that free energy, such as mining bitcoins?
Things to compute with are not free.
Renewables are completely dominating. New coal and gas capacity is almost non-existent because it is too expensive in comparison.
Even renewables + the extra cost of battery backup are much cheaper than building a new coal or gas plant.
Coal and gas are only competitive when the plants are already built, and even for pre-existing its at the tipping point where a new-solar build is cheaper than even supplying coal/gas to an existing power plant.
But sure, I guess they just hate money and want to burn cash as fast as they can. They'd totally ignore allegedly cheaper energy sources.
Are they? So far it seems like we use as much fossil as 10 years ago, we just pilled up renewables on top.