So when do we start putting big ice cubes in the ocean?
(Really this at least makes more sense then land slop centers, still silly)
Soma vibes intensify
1.15 PUE is considered normal for water cooled datacenters. 1.5 is for air cooling. It might be more tightly packed, or conservative to not very deep water temperatures.
I hope it fucken sinks
Why?
AI is why.
It has its uses. The way it’s being pushed is a different issue entirely.
Let’s boil the ocean everyone.
Instead of drilling into the core we’re just going to get boiled alive. Pretty poetic for a garbage system like capalitism.
This makes me wonder what is better - underwater DCs heating the oceans, or above water ones with all the pollution creating and water sucking cooling instead. Part of me thinks the underwater one might be better.
The issue with climate change was never with “heat production”. It’s always been the generation of heat trapping chemicals. The sun sends a stupid amount of energy our way. Generally the earth radiates almost the same amount back out into space, with a minor amount captured by various things, like photosynthesis.
Pollution alters that equation and causes more energy from the sun to get trapped in the atmosphere. That’s the problem. We could never generate as much energy as the sun (even the tiny amount that hits the earth), but we can definitely alter the atmosphere to trap more and more of that heat.
Also, the ocean is a MASSIVE heat sink. I saw someone work out the calculations recently, I don’t remember the numbers, but the conclusion was that we’d never measure a notable increase in ocean temps if we housed every datacentre in existence in the ocean. The sun hitting the ocean every day dumps more energy into the ocean directly than we’d ever be able to manage.
It all comes down to pollution.
I wonder how many sq km of data centers it would take to increase the temp of the ocean by 1 degree.
This page says the ocean is about 352,670,000,000,000,000,000 gallons, which is about 1.3 x 10^21 liters, and each liter is a kg of water (yeah, yeah, the dissolved salt adds some mass but I don’t think it adds sufficient thermal mass to make a difference). It takes 4.184 kilojoules to raise 1kg of liquid water 1°C, and 1 joule is 2.778 x 10^-4 wh.
So that’s 1.55 x 10^18 watt hours, or 1,550,000 TWh.
Global electricity consumption is about 30,000 TWh per year, so if you use the entire world’s electricity consumption for 51 years you’d raise the oceans’ temperature by 1°C.
Or if you take global data center power capacity of about 125 GW, and ran them at full power 24/7, you’d be producing about 10.8 TWh per day or 3944 TWh per year. It’d take about 393 years of the world’s data centers to raise the ocean by 1°C.
Just goes to show that much more of the energy heating up our world and our oceans is coming from the sun heating up the planet and the planet failing to radiate it out past our greenhouse blanket, not from the actual heating of our atmosphere from our own energy sources.
Parts of the ocean are colder and several species are having issues locating new spawning grounds.
I remember hearing of a corodile species or something that recovered due to a new power plant discharging warm water.
Overall ocean temps rising is a problem, but the real problem is becoming more uniform temps.
Cold spots are getting warmer. But warm spots are getting colder too. And especially for fish and reptiles. They need warm spots to spawn.
Ecologically speaking this is likely to be a good thing and within a couple years this could be a very important habitat that people are talking about and acting shocked about.
Even tho logically it’s obvious
The world-first part must be the wind-power thing.
We’ve had small offshore data centers for years, passively cooled but powered by nuclear energy.
(And if you’re still not getting the joke, let’s discuss how a nuke sub would cool its massive computing power. Big boats are like floating data centers; submarines even more advanced. )
:Yawn: Worlds first? Microsoft already did this.
https://news.microsoft.com/source/features/sustainability/project-natick-underwater-datacenter/
Good for you for reading the article!
Finally we can
piratefish for real RAM!China says
China, you have lied to me too many times.
How many AI datacenters will it take to boil the ocean?
Around (4 to 6) * 10^(26 to 27) J total
1 gigawatt is 10^9 J/s (so around 130 billion years to reach the above.) For a terawatt that’s 130 million years. For a petawatt 130,000 years. For an exawatt about 130 years…
Note: the sun bathes Earth with around 170,000 TW (0.17 exawatts) of energy. That’s about 700-800 years if you could make the oceans sink all that sun energy. Again, this isn’t the total output of the Sun but just what impacts Earth directly.
It would probably take more energy than we can harvest on earth, considering the sunlight and geothermal energy doesn’t boil it currently.
I could see it affecting the temperature on local scales, such as the area immediately around the data center.
I don’t think people mean literally boil the ocean. Just increasing it by few Celsius degrees can be world ending.
If every data centre was passively cooled in the ocean it wouldn’t change temps by even 0.01 degrees. The Sun blasts an entire half of the planet with an absurd amount of energy every day. Energy, which technically originated from the sun, is just converted and being utilized to do work.
That’s true, but I still don’t think we can raise ocean temperatures through direct cooling and renewable sources the way that the greenhouse effect can. Water can absorb a lot of heat energy without changing temperature, and that is why regions close to oceans have a more temperate climate.
While I don’t have enough knowledge in this field to be making any definitive statements, my logic is as follows:
- outside of nuclear fission/fusion reactions, heat energy on the earth’s surface comes from either the sun or molten rock in the core
- that energy is responsible for everything that happens on earth, including wind energy
So we would need to get energy from off planet, use nuclear fission/fusion, or cover enough of the land area in wind and solar farms in order to redirect the sun’s energy over to the oceans.
I think the bigger concern, when it comes to heating the ocean, is that manufacturing, construction, and transport related to the data centers still releases a lot of greenhouse gases. Those gases trap the sun’s energy within our atmosphere and that WILL heat up the earth. Way more than direct cooling using ocean water.
There are a number of 6-8GWe nuclear plants that dump 15+GW into the nearby sea (or in the case of Bruce, Lake Huron). I don’t see it being much of an issue. Better than virtually any other cooling option.
The issues are maintenance, energy source, and equipment supply.
The plants on the lakes so monitor the water temp so they don’t affect the ecosystem during the warmer seasons still.
But I doubt the one in NB had to worry about that when more water flows by it than all the rivers in the world combined.
But yes, much better source of cooling at the cost of maintenance and equipment. Just like tidal power but with fewer moving parts.
Good point, although on the local scale you mention, wildlife could still be impacted. Hopefully, the overall impact on the ecosystem will be monitored and studied before expanding these data centers more broadly.
Well, if the energy comes from solar on the thingy, then it’s probably going to cool the ocean, could be similar with wind.
That’s a good point. Maybe not cool, but it would warm the water less.
(I’m guessing solar cells reflect less energy back into space than water, since they’re specifically designed not to.)I don’t know how to objectively figure this out, but solar panels only convert energy from radiation down to far infrared of 1100nm. Water can absorb longer wavelengths, but solar output has less and less energy output at these wavelengths. However, the mystery is whether or not the panels themselves absorb or reflect such far infrared energy. I’m torn between “it might be the same” and “I’m wrong”
40…2…ob…viously…
sudo systemctl poweroffOH FUCK I was in an ssh session!
*Puts on scuba gearHopefully that server has IPMI.
Deep sea
welderadmin.Most stressful job imaginable
At least you won’t have to deal with end-users on site.
China: “We will use the oceans water”. usa: “We will use the citizens drinking water”.
Oh don’t worry. Chinese citizens already can’t drink their water.
That’s from the 90’s, the more modern study shows a pretty big but gradual improvement
https://www.sciencedirect.com/science/article/abs/pii/S0048969720374544
If only the US had a coastline.
China will use the drinking water, tell everyone they’re using the oceans water, and then arrest anyone who tries to report on their crimes.
This is pretty impressive. If only China had a good human rights record. But then again there’s only a few countries that do and none of them are superpowers.
To be fair, there’s a huge difference between some countries human rights records and China. China is IMO easily a bottom 25%.
So there’s a non-zero chance we will find out later that it’s just a bathysphere full of children doing math.
But then again there’s only a few countries that do and none of them are superpowers.
Really? Which ones?
Personally I’m a big fan of Antarctica.
Never seen their people or government do stupid shit.
Wakanda and Narnia.
Time to invest in teabag stock.
I’d really like to know how they handle all the small-scale HW issues. As a DC tech, I’m kept quite busy with those
Larger DCs don’t replace individual components, they wait until a percentage of servers on a rack have failed, then replace the rack or servers.
They will likely adopt this same model
I bet they duplicate everything and just switch off faulty units. Every year or so, they would emerge the whole thing and replace what they need at a large scale.
Sounds expensive. I’m betting they just abandon it and sink a new one with new, faster hardware.
I’m thinking they replace the module every so often when it fails
There was an Intel experiment a while back where they left a bunch of racks in the parking lot. They found that the failure rate wasn’t much higher than inside, and not needing a data center building saved money. Maybe this project just accepts the eventual failure of components.
Sure, if there is zero weather a building wouldn’t be needed.
They did get rained on. I am having trouble finding an article about it now.
Computers can’t get wet from the rain if they’re underwater in the ocean.
👉😏
Probably have diver it techs… boy that’s kind of cool
They’re probably stacks of 8x NPU Huawei servers all cooperatively serving the same few models.
As an older example, I believe Deepseek V3 was most optimally served with ~384 GPUs in a single cluster, before they switched to Chinese NPUs. So they’d have some software that ties all these together as one “server” and maybe multiple of those all serving API requests for one endpoint.
But it doesn’t actually need all 384 in each server. Many models will fit in a single 8-GPU/NPU server, but the software pools more just to try and utilize the hardware better.
If one server fails, the system would return a few requests as empty and have to restart the serving software, but… that’s fine. All the data is ephemeral. Even if the whole 24MW unit fails, they can just route API requests somewhere else, and a few failed generations isn’t a big deal.
