Data centers are water and power hogs, but does putting them in the ocean help?:
Data centers like those used to train and run AI models have this irksome tendency to drain the local water supply for the purpose of cooling through heat exchange, sometimes worsening water scarcity in an area. They also suck down so much energy that they drive up demand, and it appears we may be paying for it with higher bills.
Maybe the solution is right under our noses: submerge the data centers in the ocean, and power them with wind.
In Shanghai's Lin-gang Special Area, a new project that cost the equivalent of $226 million has proven that such a project can at least get through the early phase of construction. In theory, this will be a sort of free lunch for compute once it's completed: water ceases to be an issue, as does the data center's carbon footprint. But is it actually a good idea?
Reports about the project have been published in a few places, including Wired. The facility, Wired's story notes, currently has "a total power capacity of 24 megawatts." That's like a normal, pre-AI data center, according to a report by McKinsey, which notes that data centers "that averaged tens of megawatts before 2020 will be expected to accommodate at the gigawatt scale" in the coming years.
That story also notes that over 95 percent of the center's energy "comes from offshore wind turbines," so it sounds as if the energy comes from wind that is then wired in, rather than having a wind power generating station installed right there at the data center.
But as Wired also pointed out in a story last year about a smaller, but similar, project in the US, this might not be a great idea. In part, that's because while it may sound green, the heat exchange from all those GPUs would at least to some degree heat up the ocean—one of the main things climate hawks are trying to avoid.
The founders of a startup called NetworkOcean said they would "dunk a small capsule filled with GPU servers into San Francisco Bay," but did so "without having sought, much less received, any permits from key regulators," Wired's Paresh Dave and Reece Rogers note. Dave and Rogers sought out commentary from multiple scientists, learning that even minor temperature changes in the bay "could trigger toxic algae blooms and harm wildlife." And a data center doesn't have to be huge to cause problems. "Any increase" in temperature is a potential problem, as it could "incubate harmful algae and attract invasive species."
A 2022 paper on underwater data centers further speculated that unpredictable events like ocean heatwaves near such data centers would result in animals essentially suffocating in de-oxygenated water.
In the Wired story on NetworkOcean, fear of regulatory pushback eventually appears to drive the company to consider other jurisdictions beyond the U.S., although it claims it still wants to operate in San Francisco Bay. NetworkOcean might be a great company, and I'm not in any way picking on it. I'm bringing it up as a reminder of a truism: Here in the U.S., big, disruptive tech ideas sometimes meet with regulatory pushback—and sometimes that's because more information about what could go wrong really is needed.
By contrast, the Chinese project appears to have obeyed local regulators, according to Scientific American's piece on the underwater data center. The project received an assessment from the China Academy of Information and Communications Technology, which is under the aegis of a Chinese government ministry.
But China has big time ambitions around driving down the energy use of its data centers. According to one report, the power usage effectiveness (PUE) for data centers globally has fallen to about 1.56 on average and essentially plateaued. A press release on a Chinese government website last year stated that by the end of 2025, China will drive down its own average PUE to 1.5.
It would be an understatement to say China and the U.S. are two contrasting business and regulatory environments. But the ocean is a big interconnected resource that we all share. Lots of data centers are about to be built. Here's hoping that submerging them to meet ambitious environmental goals is something that happens, if it turns out to be a good idea.
(Score: 3, Insightful) by jasassin on Friday October 31, @07:01AM (5 children)
I'm sure you've all heard it before. The answer to any news article posting that ends in a ? is a "No." answer.
Like the oceans don't have enough problems with sun tan lotion and garbage islands we need some AI nobody asked for killing what's left of the phytoplankton.
I hate AI.
jasassin@gmail.com GPG Key ID: 0xE6462C68A9A3DB5A
(Score: 4, Interesting) by turgid on Friday October 31, @09:58AM (4 children)
The fish love the warm water. There's more food and they grow bigger. Anglers used to love our cooling water outlet. It was 9C warmer than the rest of the surrounding sea.
I refuse to engage in a battle of wits with an unarmed opponent [wikipedia.org].
(Score: 3, Touché) by c0lo on Friday October 31, @11:12PM (2 children)
You forgot: in some place of Scotland.
I'm not so sure they do, say, in the Caraibes.
https://www.youtube.com/@ProfSteveKeen https://soylentnews.org/~MichaelDavidCrawford
(Score: 0) by Anonymous Coward on Saturday November 01, @03:00AM (1 child)
Anyway might be useful for aquaculture.
(Score: 2) by c0lo on Saturday November 01, @07:17AM
You think it matters?
Shanghai has a humid subtropical climate (Köppen: Cfa), with an average annual temperature of 17.5 °C (63.5 °F) for downtown areas and 16.2–17.2 °C (61.2–63.0 °F) for suburbs [wikipedia.org]
The climate of the area is tropical, varying from tropical rainforest in some areas to tropical monsoon and tropical savanna in others. [wikipedia.org]
https://www.youtube.com/@ProfSteveKeen https://soylentnews.org/~MichaelDavidCrawford
(Score: 2) by mcgrew on Saturday November 01, @12:15AM
I would ask "What ocean?" If it's in the far north, the arctic, or off the coast of Antarctica it sounds like a great idea. Not that I'd want to work there. Off the coast of Ecuador? Not so much.
We have a president who posted a fake video of himself shitting on America
(Score: 4, Insightful) by spiraldancing on Friday October 31, @09:38AM (4 children)
Just as a matter of foundational curiosity ... what happens to the water after a "traditional" data center uses it to cool the engines? Is the heated water recooled and reused? Or is it just dumped? Or what? And ... you know ... heated water shouldn't be a waste product so much as a commodity. For many applications, people actually prefer the water to be warm/hot -- I realize there are transportation issues and what-not, but in principle, at least, hot water should be a net positive, no?
Unless the data centers are also contaminating the water?
I recall some old story (stories?) of entrepreneurial bitcoin miners relying on the machines to heat their home/apt in the winter....
Lets go exploring.
(Score: 5, Informative) by turgid on Friday October 31, @09:50AM
Commercial power stations use cleaned-up (demineralised) fresh water for cooling. It's in a nominally closed loop. Very little is wasted. It does need the odd top-up now and then but it's relatively small. Secondary cooling is done using an open loop of e.g. sea water that has had macroscopic objects filtered out (old boots, shopping trolleys, planks of wood, dead mafioso). There's no reason data centres can't be the same.
I refuse to engage in a battle of wits with an unarmed opponent [wikipedia.org].
(Score: 2) by VLM on Friday October 31, @12:16PM (1 child)
If you live in an area next to a soft water river or soft water wells that is ALSO in a low dew point area, its a capex vs opex balance where you can spray water against the air conditioner condenser coil and that'll chill it down to the local dew point. So now instead of rejecting heat at air temp maybe 95 on a summer day, you can reject heat to cold coils at maybe 70 at worst where I live (clearly not tropical south LOL) This means the coils are smaller.
Basically its an elaborate lower humidity "swamp cooler".
You get to pay for water and sprayer pumps and maintenance to the condensers as some crap will build up on them even in "soft" water areas.
Where I live the well water is legendarily hard, the river has strong environmental protection (tourist state, we'd rather have fishermen tourists than cheaper cooling) and paying for water softening of hard well water makes it a lot cheaper to install "somewhat larger condenser coils".
Other "fun" results such as massive black mold growth on the condensers etc.
SOME places have this for surge cooling on a couple of the hottest afternoons of the year. For the expense of setting up the gear it would be cheaper to hire a day laborer with a garden hose and sprayer... Theoretically I imagine you could fill an olympic sized swimming pool with rain (we get 3 to 4 feet/year) for the hottest days of the year and it would be water-neutral.
You can slug the compressor if you do something really dumb that floods the system with entirely liquid refrigerant if you get it too cold... compressors usually don't like a slug of liquid in the intake very much. Also the optimum pressures of the system are somewhat fine tuned so its not as simple as "just spray water on the condenser" if you're trying to squeeze out every % of efficiency, you have to design the whole system to an optimum point.
I remember at uni having to read Schrader's Thermodynamics for physics and it was a much better text than whatever they made us read in the chem eng thermodynamics class that I don't even remember although physics thermodynamics books are way more general and the book we had to read for chem eng focused pretty tightly on process heat and refrigeration. Kind of like you always remember Strang for linear algebra or Knuth for computer science. I would imagine "HVAC technician" level books explain what to do but not why to do it.
(Score: 2) by VLM on Friday October 31, @04:55PM
I found my copy on my bookshelf and its Schroeder's Introduction to Thermal Physics
Its a classic, been "the cool textbook" since before Y2K until today AFAIK.
I think its too new, and I'm misremembering "Heat and Thermodynamics" by Zemansky.
One of them starts macro and goes micro and the other starts micro and goes macro; in my infinite spare time I'll get around to re-reading both. I think its Zemansky that starts big and goes smaller. Probably I think you get the best education by reading both.
(Score: 3, Interesting) by stormwyrm on Saturday November 01, @07:02AM
Numquam ponenda est pluralitas sine necessitate.
(Score: 4, Interesting) by VLM on Friday October 31, @12:02PM (2 children)
What will they do with these things in the future:
1) The bubble will pop: They'll be bankrupt, abandoned, flooded, and all the metals and plastics in them will corrode and pour into the ocean.
2) Moore's Law the usual techno-optimism: So you'll be able to do the same work in 2035 with 2.4 MW instead of the 24 MW in the article. Again, now what?
Its kind of a temporary economic condition like harvesting all the nice trees in the UK to make clipper sailing ship sail spars. If you asked an economist in the 1800s what we'd be doing post Y2K they'd have a long lecture about genetically engineering test tube grown sailing ship spars for the era of wooden ships and iron men.
A lot of decisions boil down to "lets just get past this nonsense without doing anything too stupid"
If you want a more serious answer if I were the Red Army I'd put up solar powered coastal drone charger stations. They're "for AI". They're really to defend against ancient F-16s and attack submarines with drone swarms if the Taiwan thing ever heated up. Perfect location, reasonable power levels, good cover story...
(Score: 3, Informative) by c0lo on Friday October 31, @11:20PM (1 child)
Then I'll do 10 times the same work for the same energy as now.
Not that it will be more socially efficient/beneficial than now, but your premise is "you will be able".
https://www.youtube.com/@ProfSteveKeen https://soylentnews.org/~MichaelDavidCrawford
(Score: 2) by VLM on Saturday November 01, @01:16PM
Possibly correct.
I invest in energy as a hobby and electrical demand in the USA is interesting because from WWII to Y2K its a smooth-ish very linear increase, then at Y2K it flatlines. Theoretically its interesting to think about a flat line in computation use.
(Score: 3, Interesting) by ElizabethGreene on Friday October 31, @03:12PM (7 children)
Can someone check my math and show me where I went wrong?
Pessimistically*, the SF bay is about 400 square miles and about 10m depth, so that's 1E10 cubic meters.
Multiply that by 1000 to get the mass of a water in kilograms and 1000 again to get the mass of the water in grams = 1E16 grams of water
The specific heat of water, the energy required to heat one gram by one degree, is 4 joules, so to heat the entire bay up by 1 degree would require 4E16=40,000,000,000,000,000 joules of energy.
24 megawatts = 2.4E7 Joule seconds, multiply that by 365*86400 to get joule-years = 7.6E14 joules
You could dump an entire year's heat output of the data center into the bay and warm it less than 0.02 degrees. IIRC, we can only measure down to millikelvins, 1E-3, so spread across an entire year it's literally too small to measure.
* Internet search estimates for the Bay surface area were 400-1700 square miles and an average depth of 10-17m.
I'm reasonably confident I did the above math right.
But what about growth?
California's total electricity generation is on the order of 2.5E8 megawatt hours [Source: https://www.eia.gov/electricity/state/California/ [eia.gov] ]
Multiply that times 3.6E9 joules per megawatt = 9E17 joules. Putting ALL of that into the bay at once would increase the temperature dramatically, by 9E17/4E16 =22.5 degrees C =~41F. (Is this right? We could dump the entire electrical output of California **for a year** into the bay with a giant space laser and only get it up to bathtub warm?)
Presuming I did that math correctly, I am unconcerned about submerged datacenter thermal output until it increases by several orders of magnitude.
(Score: 4, Informative) by OrugTor on Friday October 31, @04:49PM (1 child)
The assumption in the math is that the whole bay gets heated. In practice there is a temperature gradient from where the heat is dumped (typically in one place, at the facility) to the rest of the bay. There will be a rise in temperature around the facility that will encourage blooms etc. It's like the cooling water coming out of a nuclear power station on the shore. The heat does not rapidly disperse into the entire ocean. Instead the temperature is higher locally and the biome is quite different.
(Score: 2) by c0lo on Friday October 31, @11:25PM
The same happens near underwater volcanic vents. Warm water doesn't stay stagnant when there's a thermal gradient.
https://www.youtube.com/@ProfSteveKeen https://soylentnews.org/~MichaelDavidCrawford
(Score: 2) by VLM on Friday October 31, @05:13PM (1 child)
Google thinks the bay might hold 100 trillion liters
It takes 1500 watts for two minutes to boil my 1L tea kettle (to one sig fig).
For your example you used 2.5E8 MWh and 41F.
One ratio is obvious, the bay has 100 trillion times more water than my 1L kettle.
1500 W for 2 mins is 1500/1e6 megawatts / 30 (two minute intervals in an hour) is what like 0.00005 megawatt hours? So you got 5 trillion times more energy in the bay than in my tea kettle?
Your 41F rise is about 25% of the way from cold tap water to 212F in my tea kettle, so ideally my calculated number should be around 25% also.
So 5 trillion times more energy / 100 trillion times more water = the bay should warm 5% of the way to boiling. Thats about a fifth of your estimate of 25%.
Probably the figure I trust the least is the volume of the bay. Doesn't it depend on tides? Theres a LOT of rounding going on both your mass analysis and my volume analysis, so being within a factor of five is actually pretty good.
Overall your math is probably correct although both our estimates feeding into the math are pretty "meh".
The bigger picture is the conclusions are unchanged regardless who's analysis you trust. Using the entire bay as a well mixed heatsink would work pretty well.
Large bodies of water are rarely well mixed, however.
(Score: 2) by VLM on Friday October 31, @05:21PM
Oh I think I missed something obvious.
So if a mL is a gram (back of envelope engineering estimate) thats 1000 or 1e3 grams per liter
So your estimate of the capacity of the bay is 1e16 / 1e3 = 10e12 liters.
My google estimate of volume was some WAG of 100 trillion liters and yours is 10 trillion liters so whatever.
Anyway close enough the "bigger picture" does not change, you can't do a fish boil dinner in the Bay using current levels of electricity generation. You'd need many orders of magnitude more electricity to boil the bay.
(Score: 2) by c0lo on Friday October 31, @11:30PM (2 children)
If you're so keen on math, compute how much energy (or power) is coming from the Sun over the bay surface, taking the solar constant at about 1kW/sqm. Assume some reasonable absorption coefficient for the bay water.
https://www.youtube.com/@ProfSteveKeen https://soylentnews.org/~MichaelDavidCrawford
(Score: 2) by ElizabethGreene on Saturday November 01, @03:42AM (1 child)
That's a good comparison, though I can never remember if the 1kW/m^2 is on top of or under the atmosphere. 1E9 square meters, so spitball at 1E12 joules, half that because of nighttime, so 5E11 watts vs 24 megawatts = 2.4E7 watts .. 4 orders of magnitude more than the data center?
Is that right? I'd expect a 20 mile per side square of solar panels to power more than 2,000 data centers, but I don't really have mental framework for the sizes they build at today.
(Score: 2) by c0lo on Saturday November 01, @05:09AM
Not that it matters for order of magnitude comparison, but about 1kW/sqm is sea level, the solar constant at top atmosphere is about 1.3kW/sqm.
https://www.youtube.com/@ProfSteveKeen https://soylentnews.org/~MichaelDavidCrawford
(Score: 3, Informative) by pTamok on Friday October 31, @03:42PM (2 children)
All About Industries: 2024-06-28: Underwater data center: Microsoft Abandons the Concept of Reliable Underwater Data Centers [all-about-industries.com]
Project Natick
The costs of building an entire data-centre underwater that can be maintained by humans would be immense. Far cheaper to build on the shore and run cooling pipes into the sea.
(Score: 2) by turgid on Friday October 31, @06:34PM (1 child)
I was going to make a facetious comment about it being Microsoft and reliable. The location is irrelevant.
I refuse to engage in a battle of wits with an unarmed opponent [wikipedia.org].
(Score: 4, Funny) by mcgrew on Saturday November 01, @12:25AM
A guy is searching under a streetlight. A cop walks up. "lose something?"
"Yeah, my car keys."
The cop looks around, the nearest car is parked between the lights. "Why did you have your keys out here?"
"I didn't, I dropped them by the car,"
"Why are you looking here?"
"There's no light by my car."
The cop says "You must work for Microsoft."
We have a president who posted a fake video of himself shitting on America