Sam Altman compares AI energy use to the cost of "training" humans, says water-usage concerns are "fake"

midian182

Posts: 11,620   +175
Staff member
WTF?! It seems OpenAI CEO Sam Altman has grown tired of defending AI's consumption of our planet's rapidly depleting natural resources. In a recent interview, he said water-usage concerns over data centers were "fake," before highlighting the shocking amount of energy it takes to "train a human."

Altman launched his tirade at the India AI Impact summit during an interview with The Indian Express.

When asked about reports of how much water ChatGPT uses every time it responds to a query, Altman said that claims it consumes "17 gallons of water for each query or whatever," are "completely untrue, totally insane, no connection to reality." He then added that the water issue had been more of a concern when OpenAI used evaporative cooling in data centers, though the CEO never revealed exactly how much energy the company now uses.

Most data centers continue to use huge amounts of water for cooling purposes, but some are turning to technologies that reduce this amount. In August 2024, Microsoft launched a new data center design that consumes no water for cooling.

But even with these efficiency gains, water companies have warned that water drawn for data center cooling is expected to triple over the next 25 years as compute demand increases.

Altman acknowledged that concerns over the energy used by data centers were "fair" as so much of the world is now using AI. As such, he suggested that the energy sector moves toward nuclear, wind, or solar power very quickly.

Altman was also asked about Bill Gates' comments about whether a single ChatGPT query currently uses the equivalent of 1.5 iPhone battery charges, to which Altman replied, "There's no way it's anything close to that much."

The OpenAI boss then complained about the negativity around AI, highlighting how much energy we humans use. "One of the things that is always unfair in this comparison is people talk about how much energy it takes to train an AI model," he said. "But it also takes a lot of energy to train a human."

"It takes like 20 years of life, and all the food you eat before that time, before you get smart. And not only that, it took like the very widespread evolution of the hundred billion people that have ever lived and learned not to get eaten by predators and learned how to figure out science and whatever to produce you, and then you took whatever you took."

Altman insisted that the fair comparison is "if you ask ChatGPT a question, how much energy does it take once a model is trained to answer that question, versus a human, and probably AI has already caught up on an energy efficiency basis, measured that way."

Unsurprisingly, Altman comparison of AI's energy use to that of a human has not gone down too well. It feels like an especially contentious thing to say at a time when more AI executives are warning about the enormous job losses the technology will cause. Altman himself once said AI could wipe out entire job categories.

Globally, data centers consume around 415 TWh of electricity, about 1.5% of the world's total. The IEA projects global data center electricity use could roughly double by 2030 to around 945 TWh (just under 3% of global electricity), and a large share of this growth will stem from AI specifically.

In the United States, a US DOE report estimates that data center energy usage could hit as much as 580 TWh by 2028, or around 12% of the nation's total. If that concerns you, just remember Altman's justification: humans use up a lot of energy as well.

Permalink to story:

 
The electricity and water use are seriously the least impactful aspect of the AI buildout. Personally I'm not all that bothered by it. There are much more serious issues:

a.) They're sucking up all investment money leaving nothing for companies not doing AI. Hiring in tech has completely stalled.
b.) They're sucking up all hardware types leaving shortages and price hikes for most consumer electronics. Worse is all these chips will be garbage in 5-10 years.
c.) They have no plan to profitability from all this. The way AI is done today is hugely inefficient. It requires significant compute cycles for each prompt. It is not sustainable and there's no way it's covered by the $20/month they're charging.
 
CEO Sam Altman has grown tired of defending AI's consumption of our planet's rapidly depleting natural resources
I'm tired of the ridiculous hysteria about "rapidly depleting natural resources".
Which resources exactly are rapidly depleting???? Because I can't recall any.
 
a.) They're sucking up all investment money leaving nothing for companies not doing AI.
Venture capital is not a fixed-size fund. Good ideas are still being capitalized.

Hiring in tech has completely stalled.
The tech industry doesn't exist to employ programmers; it exists to generate products and services. AI is helping to provide more of those at lower cost.

b.) They're sucking up all hardware types leaving shortages and price hikes for most consumer electronics.
Short-term, sure. Long-term (and even medium-term) AI is funding the R&D for the next generation of processors and memory chips: advances that will benefit the entire industry. It's been a VERY long time since videogamers were spending enough to fund GPU development.

c.) They have no plan to profitability from all this ... It is not sustainable and there's no way it's covered by the $20/month they're charging.
You've confused public LLM chatbots with all of AI, when they are only one tiny sliver of the entire machine learning market.
 
Venture capital is not a fixed-size fund. Good ideas are still being capitalized.


The tech industry doesn't exist to employ programmers; it exists to generate products and services. AI is helping to provide more of those at lower cost.


Short-term, sure. Long-term (and even medium-term) AI is funding the R&D for the next generation of processors and memory chips: advances that will benefit the entire industry. It's been a VERY long time since videogamers were spending enough to fund GPU development.


You've confused public LLM chatbots with all of AI, when they are only one tiny sliver of the entire machine learning market.
Guys....I think we found Altman's Alt.
I'm tired of the ridiculous hysteria about "rapidly depleting natural resources".
Which resources exactly are rapidly depleting???? Because I can't recall any.
Clean fresh water for one. Where do you think it is coming from? Especially the western US has been pumping aquifers dry for decades, those dont refill quickly, or ever. The snowpacks in Colorado cant keep up with demand and snowfall keeps reducing over time.

The fossil fuels used to generate all this electricity for another. I mean does that even need explanation?
 
Guys....I think we found Altman's Alt.
Clean fresh water for one. Where do you think it is coming from? Especially the western US has been pumping aquifers dry for decades, those dont refill quickly, or ever. The snowpacks in Colorado cant keep up with demand and snowfall keeps reducing over time.

The fossil fuels used to generate all this electricity for another. I mean does that even need explanation?
70%+ of earth is covered in water, this is the least scarce thing ever.
Closed loop cooling systems consume virtually no water.
In many cities, more than 50% of treated water is lost in pipe leaks, but I haven't seen a single hysterical campaign about that.
Pipe leaks in a single major city waste far more water than the combined consumption of all datacenters.

Fossil fuels are also nowhere near scarce, there's more than enough of each kind. We'll stop using fossil fuels long before they become scarce.
Resource scarcity is nonsense.
 
I'm tired of the ridiculous hysteria about "rapidly depleting natural resources".
Which resources exactly are rapidly depleting???? Because I can't recall any.
The one that leads to ever higher taxes for the poor and more tax scams for the rich, siphoning money to them, just like the imminent ice age, global warming and finally climate change scam.
 
Altman insisted that the fair comparison is "if you ask ChatGPT a question, how much energy does it take once a model is trained to answer that question, versus a human, and probably AI has already caught up on an energy efficiency basis, measured that way."
That's fair. But I reckon it's more beneficial for humanity to train the human.
 
70%+ of earth is covered in water, this is the least scarce thing ever.
Closed loop cooling systems consume virtually no water.
In many cities, more than 50% of treated water is lost in pipe leaks, but I haven't seen a single hysterical campaign about that.
Pipe leaks in a single major city waste far more water than the combined consumption of all datacenters.

Fossil fuels are also nowhere near scarce, there's more than enough of each kind. We'll stop using fossil fuels long before they become scarce.
Resource scarcity is nonsense.

Most of that water is salt water.....which requires massive investment and electricity to desalinate.

As for fossil fuels, the abundance and/or scarcity of those are at the whims of the political class. For every four years the prices drop and supply increases, there are 4 years where the prices skyrocket and supplies are scarce.

I'm all for 100% wind and solar being the exclusive power supplies for AI. That would solve all our problems. The wealthy tech elite were pushing these on the general public for years. Now it's time to show us how much they really believe these sources can solve all of OUR energy problems and decrease our reliance on fossil fuels. They just need to put their money where their mouths are and use them for this AI future.

I bet just like the objections to fossil fuels that have evaporated, we will now start hearing about how wind and solar "are not suitable" for what THEY are doing. It's just fine for us plebs.

Altman is only concerned with his status, legacy and bank account. He could care less what happens to everyone else.
 
That's fair. But I reckon it's more beneficial for humanity to train the human.
If we replace all the jobs with AI then noone will have money to spend to justify the cost of AI. The mental gymnastics required to believe AI is going to he beneficial to society and justify the spending is flabbergasting. It's a cool technology, but there is no business model that makes sense....or money....

But I guess the money furnaces aren't going to load themselves, maybe we will make an AI to load them for us.
 
70%+ of earth is covered in water, this is the least scarce thing ever.
Closed loop cooling systems consume virtually no water.
In many cities, more than 50% of treated water is lost in pipe leaks, but I haven't seen a single hysterical campaign about that.
Pipe leaks in a single major city waste far more water than the combined consumption of all datacenters.

Fossil fuels are also nowhere near scarce, there's more than enough of each kind. We'll stop using fossil fuels long before they become scarce.
Resource scarcity is nonsense.
The oceans you mean, the really salty ones??
Have you been drinking a lot of it lately??

Sometimes I think this forum has been invaded by the Body Snatchers.
 
Most of that water is salt water.....which requires massive investment and electricity to desalinate.

As for fossil fuels, the abundance and/or scarcity of those are at the whims of the political class. For every four years the prices drop and supply increases, there are 4 years where the prices skyrocket and supplies are scarce.

I'm all for 100% wind and solar being the exclusive power supplies for AI. That would solve all our problems. The wealthy tech elite were pushing these on the general public for years. Now it's time to show us how much they really believe these sources can solve all of OUR energy problems and decrease our reliance on fossil fuels. They just need to put their money where their mouths are and use them for this AI future.

I bet just like the objections to fossil fuels that have evaporated, we will now start hearing about how wind and solar "are not suitable" for what THEY are doing. It's just fine for us plebs.

Altman is only concerned with his status, legacy and bank account. He could care less what happens to everyone else.
There is no need whatsoever to desalinate water. Closed cooling systems practically do not consume water. Zillions of things consume way more water than datacenters. The total consumption of all datacenters is less than the pipe leakage losses of one big city.
The whole water 'problem' is entirely made up.

For fossil fuels, you're right about the whims of the political class.
But there is no need to decrease reliance on fossil fuels. Why would anyone normal do that, decrease reliance on what's reliable and abundant??? We should generate as much energy as possible, regardless of the source. Wind and solar are as good as coal, oil and gas.
 
c.) They have no plan to profitability from all this. The way AI is done today is hugely inefficient. It requires significant compute cycles for each prompt. It is not sustainable and there's no way it's covered by the $20/month they're charging.
Altman himself has said that even the $300/month tier isn't enough to make their money back so I have no clue how the people funding them can hear this and still continue to fund them. It's insane to me.
 
The tech industry doesn't exist to employ programmers; it exists to generate products and services. AI is helping to provide more of those at lower cost.
Who do you think creates those products and services, God? It's programmers lol. All AI does is piggyback off of things that physical beings have already done. AI has not created a single product or service on it's own that has been deemed profitable. Absolutely nothing. Name something.
 
70%+ of earth is covered in water, this is the least scarce thing ever.
That's 97% salt water
Closed loop cooling systems consume virtually no water.
Which are not used in data centers. They use evaporative cooling, which is 100% loss.
In many cities, more than 50% of treated water is lost in pipe leaks, but I haven't seen a single hysterical campaign about that.
Pipe leaks in a single major city waste far more water than the combined consumption of all datacenters.
Going to need a citation for those insane number claims.
Fossil fuels are also nowhere near scarce, there's more than enough of each kind. We'll stop using fossil fuels long before they become scarce.
Resource scarcity is nonsense.
Someone hasnt been paying attention to the fossil fuel extraction industry. Fracking bailed us out of the last energy crisis in the 2000s. The oil wells of Texas have long passed their heyday. The easy to extract sweet crude is being pumped dry of what is left, the nearby offshore reservoirs are gone. Newer methods, like fracking, are very expensive by comparison to regular pumps and pose significant environmental issues. Undersea drilling has to move increasingly deeper to find new sources, driving up prices, and increasing the risk of a spill.

Actual cheap, usable fossil fuels are becoming scarce. Fossil fuels themselves are not, but what is left will become prohibitively expensive as time marches on. Smart people look at that system and decide to fix it BEFORE we end up with $10 gasoline again.
There is no need whatsoever to desalinate water. Closed cooling systems practically do not consume water. Zillions of things consume way more water than datacenters. The total consumption of all datacenters is less than the pipe leakage losses of one big city.
The whole water 'problem' is entirely made up.

For fossil fuels, you're right about the whims of the political class.
But there is no need to decrease reliance on fossil fuels. Why would anyone normal do that, decrease reliance on what's reliable and abundant??? We should generate as much energy as possible, regardless of the source. Wind and solar are as good as coal, oil and gas.
This is what convinced me you must be trolling. Do you not understand how corrosion works? Putting salt water into a closed cooling system is an AWFUL idea.
 
I wonder if Altman can go one day without spouting a load of crap.

Of course Altman would say this. If he doesn't it might negatively affect his precious AI business.
 
The oceans you mean, the really salty ones??
Have you been drinking a lot of it lately??
You understand that desalinating water is an incredibly simple process, right? And even without desalination, each year - each day, in fact -- countless trillions of unused gallons of fresh water pour directly into the earth's oceans.

Far less than 1% of the fresh water in the US is actually used for drinking. The water shortages in the Southwest are primarily because (a) we persist in growing high-water requirement crops like alfalfa in the deserts of California and Nevada, and (b) environmentalists refuse to permit either the canals or desalination plants that would bring in additional fresh water from other sources.
 
Who do you think creates those products and services, God? It's programmers lol.
Not for much longer. Do you see much thread being spun and cloth woven by hand any more?

AI has not created a single product or service on it's [sic] own that has been deemed profitable. Absolutely nothing. Name something.
Just last year a radiologist reviewed some imaging of mine, and missed an issue despite being told specifically to look for it. An AI algorithm caught it, potentially saving my life. The radiologist charged $5,000 and took three days to perform his mistaken review ... the AI based tool caught it in under one second, at a net cost in electricity of about ten cents.

You may not like being displaced by AI ... but your thoughts and feelings here are as relevant as the Luddite weavers who found themselves no longer able to hand produce men's stockings.
 
The oceans you mean, the really salty ones??
Have you been drinking a lot of it lately??

Sometimes I think this forum has been invaded by the Body Snatchers.
That's a good way to put it. Only it turns out the problem isn't the Pod People but the Podcast People.
 
The tech industry doesn't exist to employ programmers; it exists to generate products and services. AI is helping to provide more of those at lower cost.

I don't know about you but personally I've yet to have a situation where I'm getting more for less. If anything almost every service I use from github to netflix, to health insurance, to groceries is getting more expensive. We hope that in the future AI will give us more for less, but we're definitely not there yet. Though AI does provide efficiencies.

Short-term, sure. Long-term (and even medium-term) AI is funding the R&D for the next generation of processors and memory chips: advances that will benefit the entire industry. It's been a VERY long time since videogamers were spending enough to fund GPU development.

Up until 2023 50% of nvidia's revenue was from consumer cards. For AMD it was an even larger share. It has not been a "VERY" long time. Right now it's insignificant compared to their data center revenue. Nvidia wins in this, but many other business like console makers are hurt.

You've confused public LLM chatbots with all of AI, when they are only one tiny sliver of the entire machine learning market.

Chatgpt gets 2.5 billion prompts/day. These are for the latest gen models which are also the most compute intensive to run.

I am a programmer and use AI daily for both personal work assistance, and in several workflows related to our products. For my personal use I go back and forth between chatgpt and claude. For our business use cases we use smaller more efficient models which we self host. I am intimately familiar with these tools.

I do like them. They are the future. But what's happening right now is not sustainable for many reasons, some of which I mentioned.
 
You understand that desalinating water is an incredibly simple process, right? And even without desalination, each year - each day, in fact -- countless trillions of unused gallons of fresh water pour directly into the earth's oceans.

Far less than 1% of the fresh water in the US is actually used for drinking. The water shortages in the Southwest are primarily because (a) we persist in growing high-water requirement crops like alfalfa in the deserts of California and Nevada, and (b) environmentalists refuse to permit either the canals or desalination plants that would bring in additional fresh water from other sources.
Desalination plants are also extremely energy intensive to operate, which is a major problem given that the current grid is heavily reliant on fossil fuels to operate.

The largest desalination plant in the world, Ras Al-Khair plant in Saudi Arabia, produces roughly 1 million cubic meters of water per day. That plant uses 200 MW to operate. You would need 13 of these just for residential use in california alone. If you include total water use, you would need 144 of them, using 28.8 GW of power. California's grid today, which struggles with demand and brownouts, produces between 20-40GW depending on load. So you are looking at a 50-100% increase in power generation needed to meet one states water needs, not getting into the infrastructure or environmental concerns this would create.

Then there is the concern of other states. Great, you somehow cna afford to do this in california...now how about nevada? Arizona? New Mexico? Where are they going to put their desalination plants? The colorado river flows the wrong way.
 
Up until 2023 50% of nvidia's revenue was from consumer cards.
Yes, but revenue is not profit -- and profits drive R&D, not total revenues. Even in 2022, NVidia was earning nearly 3X the profit from data center/visualization as it was from consumer cards.

Chatgpt gets 2.5 billion prompts/day. These are for the latest gen models which are also the most compute intensive to run.
Again: chatbots are only a tiny fraction of overall AI usage. I already gave an example from the medical field, where an AI agent corrected a human doctor's error, and did so in less than 1/100,000 the time and cost. Here's another: a friend of mine from college owns a logistics firm. He recently replaced his routing software with an AI-based algorithm, which is not only saving him nearly one million gallons of diesel fuel per year, but results in faster shipping times as well.
 
Back