Why can't data centers recycle their water?

This is not the argument that the “Peak Oil” proponents were writing about in the mid 1990’s to 2000’s. Their thesis was that global production would peak due to reserve depletion, causing an economic crisis in the fairly near future (ranges of 2005-2015) were named. If you’d like to tell the authors of all the peak oil literature that they’ve misunderstood their own writing, be my guest.

As far as your point that nothing physical lasts forever, the best anyone can say about that is “duh”, of course every extraction curve will have a beginning, peak, and end. Of course oil supply will peak someday (assuming of course that we’re still trying to get it), it’s so obvious that it borders on being tautological. The important point is that the peak has neither the cause (reserve depletion) nor the significance (price spike) nor the timeline (21st century) that the best known Peak Oil writers suggested.

To be clear, we do need to get off oil due to the other overwhelming negative externalities (climate, war, pollution, inequality, etc). But everybody needs to stop pretending that geological scarcity is going to cause a price shock that will serve as a wake-up call. Obviously there will be other types of supply shocks, war is a big one that comes to mind right now, but no shock is going to happen because there’s not enough oil in the ground to be had.

Since we are going down the rabbit hole. Peak Oil (proper noun) was most certainly all about imminent depletion of reserves and shortages. Back when it was a big deal I worked in oil and gas exploration geophysics services. I attended a conference (in Bahrain I think) where the head of exploration of Saudi Aramco gave a little talk about peak oil and how silly it was. He outlined just how much oil there is actually under the ground. The oil producing nations expect that demand will dwindle before supply runs out as world economies shift away from oil. (Maybe, given current global politics.)

The core problem with any resource estimates is that the idea of a reserve has at least three meanings. And this matters. You have proven reserves, probable reserves, and possible reserves. Moreover, this only covers stuff you have an enough data for to be counted as a reserve. You then get to resources, which are seperate again.

Proven reserves are reserves that you have a very high confidence the oil or gas that is there, and you can recover with current economic conditions. That is oil you can take to the bank. This is often the one referenced.

Probable reserves are a reserve you are pretty sure is recoverable. With a 50% probability. Reserves that you have good data for, at least some wells, understanding of the geology and recovery. You can’t necessarily take this to the bank, but you could certainly take it to the sharemarket.

Possible reserves are less certain again. You probably don’t know enough about the geology and recovery options to be sure of economic recovery, or whether there are hidden problems. But you know enough that there is almost certainly oil there.

Resources are where you still in the discovery phase. Contingent Resources are those where exploration has reached the point where there is a good understanding that there is a deposit there, but there isn’t enough knowledge or economic drivers to push operations further. Finally Prospective Resources are where you have good feeling about a province, and have probably done some initial exploration. These are resources you keep in your back pocket.

The problem with conversations about reserves is the structural mechanism that keeps resources and reserves in these bands. Oil companies almost never have open access to new regions. Countries open up areas in a controlled manner, and companies bid for exploration rights. If they find something they will need to negotiate production rights. If they have exploration rights they will be required to actually perform the exploration within a fixed time, or lose the rights. Once production commences they are on the hook for production rights payments. All this places a drag on taking resources to reserves to production. Companies only ever have enough proven (and producing) reserves as they need right now. It takes a long time to develop a field, so the pipeline is always progressing, but only as fast as projections for demand dictate. There are no upsides to bringing reserves up early.

The same effect applies to minerals and mining in general. Whenever there is handwringing about the world running out of X, be careful that X is not rate limited by these processes. Nothing is infinite out there, but there is usually more than we think. (The only one that really worries me is helium. Economic helium could be depleted within our lifetime in a very unfortunate manner. Our profligate use may come back to bite us.)

Moderator Note

A comparison to peak oil is ok, but let’s keep in mind that the topic of this thread is data centers and not peak oil. If you want to discuss peak oil further, please take that discussion to another thread.

For this thread, let’s keep the focus on data centers and water use, and keep further discussions of peak oil out of the thread unless the discussion is directly related to data centers and water use.

Peak oil was predicated on available oil sources (“reserves”). Fracking - extracting oil from oil shale, which is plentiful across the US midwest - added a few decades to the calculations. It didn’t hurt that the “gold rush” to do fracking resulted in a glut of oil, meaning many fracking companies built on the assumption of higher oil prices went bankrupt. Thus, the infrastructure they left behind was bought up cheap, and can now produce with no capital cost baggage.

back to the topic at hand…

Let’s say your typical AI Datacenter uses 150MW. 1 MWh is 806,000kCal. To boil 1L of water from 20°C (68F) (to steam) takes 620kcal. 150MW therefore will boil about 200,000L/hr. An Olympic swimming pool is about 2.5M litres. So you need to boil off almost 2 swimming pools a day.

Now cooling/evaporation is not boiling, so likely will need even more cooling water.

You might use those clever tricks as mentioned, like heating sidewalks. What do you do when the sun is hot enough to fry an egg on the asphalt? Where does the heat go?

You can dump the heat in the local lake or river, but there are ecogical consequences which would discourabge this, unless you have a govrnment who doesn’t care about the environment.

Forced air, like a car radiator? That’s going to be a huge (YUGE!!) radiator and a lot of noisy fans.

Quick calculation if I still know my math & physics…

An Olympic pool is 2.5M litres. To heat one from 20°C to 60°C (68°F to 140°F a bit hotter than a typical household hot water tank) takes 40Kcal/L so about 10M Kcal.

1 MWh = 806,000Kcal. To dump 150MWh into the lake each hour, would take heating 12 swimming pools an hour and likely kill any fish near the outflow.

The short answer is still - disposing of waste heat will either do an ecological catastrophe, or consume mass quantities of water, or generate a huge amount of noise. This being a typical corporate endeavour, why not all three at once?

It’s treated for cooling, to avoid corrosion, so it’s not really suitable for drinking water, and everybody else doesn’t really want hot water, any more than the data centre does.

It it’s been used for evaporative cooling, it’s also saltier and harder than people want for any use: if it was still usable, the data centre wouldn’t be dumping it.

There is one data centre I am aware of that is planning to use hot exit water for district heating, but normally data centres are in places that don’t want district heating, and even if they were, the distribution costs are very high.

I can remember that whenever peak oil predictions were made, somebody else would say that if prices went up, so would reserves. I didn’t give that any attention, because “Hubert’s Pimple” described what repeatedly happened to fish stocks, and fisherman repeatedly refused to believe that they were running out of fish.

But it turned out I was wrong: oil and mineral exploration people really did have some idea about non-reserve resources.

Doh!! 100MKcal to heat a pool, so 1.2 swimming pools an hour. 14.4 a day. Not as bad, but still likely lethal to any fish nearby.

It’s not infinite, of course. But we’ll never actually “run out” of oil. As supplies dwindle and/or the EROI ratio for oil decreases below a certain threshold (2:1?), the price will increase to the point where other enery sources become competitive, and heance demand for oil will decrease.

A bit OT, but when it comes to finite resources, I am more concerned about helium. Unlike oil, I don’t think there’s a readily-available substitute for helium that’s used for industrial purposes.

Google doesn’t seem to know what that is. Did you misspell it? Can you provide a link explaining it?

You use geothermal cooling.

I don’t understand why they can’t setup the cooling same as STEPGS ( nuclear power station in Bay City Texas… no cooling tower just a 7000 acre (2800 ha) reservoir that gets topped off as needed by the Colorado River ( one east of the Rockies, in TX) ?

Poking around on google, it seems that you have just added between $10 million and $150 million to the project just in land costs (varies quite a bit depending on where the land is located).

The cost to construct a data center also varies (depending on how big of a data center you want), but the numbers I am seeing on google are roughly in the same ballpark as your land costs for your cooling reservoir.

The cost of building a nuclear power plant is several orders of magnitude larger than that, so it doesn’t impact the overall cost anywhere near as much.

Adding a $20 million cost to a $50 million data center is increasing the overall cost by a whopping 40 percent. Adding $20 million to a $20 billion nuke plant is a 0.1 percent increase in cost.

So some parvenu tech bros are going to have to go with out their whores and Lamborghinis for a year or two :person_shrugging:

Give the size of the STPNOC reservoir they could provide cooling for several data centers. This would be more cost efficient and more inline with Venture Capital ROI targets.

Regarding the link from @HMS_Irruncible: Nordic homes are being warmed by waste heat from massive data centers, I didn’t realize some urban areas had District Heating – a system of distributing heat across a network of water pipes. If the data center is in an urban area, it can sink the heat as low-cost heating for the area.

District heating is broadly classified as multiple generations. The fifth generation has the ability to provide both heating and cooling when both are needed at the same time.

The challenge in the US (aside from the lack of district heating), is data centers are frequently built in low-density areas based on land prices, energy costs, and municipal cooperation. There are some industrial uses for the latent heat. The pulp and paper manufacturing industry often re-uses their low-temperature waste heat for adjacent greenhouses.

There are some advantages to data center’s low-temperature waste heat compared to other industries – it is clean and continuous. With some planning, it should be possible to harness the waste energy, but the US doesn’t have a mechanism to create and apply a plan.

Coal and nuclear plants all have cooling towers and return some warmed water to the environment.

There are many, many ways to use waste heat. A few US municipalities are set up for it, most aren’t. As you mentioned, pulp mills have some amazing recovery capabilities (not just for waste water but for waste industrial products).

All of this relies on infrastructure and incentives that simply aren’t there for datacenters. They don’t generate toxic waste, so there’s little need for recovery boilers and whatnot. The warm water could be put to good public use, but most places don’t have the infrastructure, and the datacenters are being built in places where it wouldn’t really make sense anyway (too distributed, agriculture doesn’t really need it).

We’re kind of mixing complaints here. The presumption in the OP is that data centers use too much water and need to recycle it. One of the obstacles to that is heat. We’ve drifted off topic into ways that the heat could be used, but dumping it in the river doesn’t address the (supposed) problem of excessive water use.

To me this kind of demonstrates my proposition that people dislike datacenters for reasons that are really nothing to do with their water use. The underlying presumption is that they create no economic value, and hence should be regulated more aggressively than other uncontroversial industries that use equal or greater amounts of water, which muddles 2 tractable arguments to create a sense of urgency.

I like that idea. Warmer water to houses means less electricity or gas to heat it, and a lot of the water use in a home is hot water. Warmer water wouldn’t hurt toilet flushing, car washing, grass/bush watering, etc. Even the chilled water and ice from your refrigerator’s door dispenser would still work, the fridge would just have to cool the water a bit more. I think that energy trade-off would be minimal compared to the larger usage of hot water in a home for bathing/showering, dishwashing (manual or machine), and washing clothes (which wouldn’t be harmed by warmer water).
Hmmm, maybe “they” will think of it?

No, I think what they’re saying is, don’t evaporative cool the water at all: use it, then pump the warmed water back into the municipal water system it came from. You’d have to figure out upstream/downstream, etc, but assuming downstream users use more water than the data center needs, then the warm water would have a place to go. I’m a mechanical/nuclear engineer, and it seems like it would work. Of course, there’d be energy usage pumping the water back in against 50-100 psi.