Water Behind the Watts: The Hidden Risk of Powering Data Centers

This study is using water "withdrawal" figures rather than consumptive use. Hydroelectric plants withdraw tons of water, but it goes right back into the river it came from.

Edit - good information on this here: https://news.bgov.com/environment-and-energy/data-centers-us...

The "Consumptive Use vs. Withdrawals" section discusses the trouble with using withdrawal as a meaningful data point.

This just isn't how you model water use in a practical way.

- If you're going to attribute water use via electricity to data centers you need to be fair and apply it to the rest of the electricity users. And in the end, you're just raising everyone's baseline water consumption in proportion to their electricity use. Why not leave the baseline increase out and leave water use attributed at the industry level (datacenters vs coal plants vs solar plants vs ...)?

- Different kinds of electrical generation plants use different amounts of water. To get an accurate number for an electricity consumer you're backing out the mix of generation sources in your model. You've introduced complexity to your model and reduced the precision of the answers your model gives when you could just get simpler and better answers leaving it at the industry level.

- Suppose we had a place with genuinely scarce water. There is some level of water draw that's unsustainable; you can never cross that level because it'll damage the environment. If there's ever a risk of hitting that limit you're interested in all users of water (so you can correctly model how much draw happens when XYZ business or person leaves the state) pushing you back to direct attribution at the industry level. If you can't model the effects of each choice correctly how do you know the choice you made will fix the problem?

> Data centers in the states analyzed depend on about 3.4 trillion gallons of freshwater annually for electricity, around 12 times the annual water use of Los Angeles, Phoenix, and Washington D.C., combined.

> Most electricity is generated by power plants located in areas facing water stress or drought.

According to the report, the Colorado river flows through Arizona, and there are dams in Arizona that need the water to flow through them in order to produce electricity.

Remarkable insights.

What's sad about this all is that none of this is actually work that needs to be thrown out. You could absolutely use the data here in terms of informing yourself on "how at risk is the power generation capacity in these states during droughts, because it relies on water in the Colorado that might not always be available".

But instead we get bandwagon water use rage bait.

When in comes to water everything besides agriculture is a rounding error.

The only thing an article like this tells you is that the author has an ax to grind.

Coal and nuclear cooling towers use a lot of fresh water, they try to reuse it but still between 2 to 3.5 billion gallons a year evaporates from each plant. Japanese nuclear plants don't have this problem for example because they are built on the coast and consume water directly from the ocean instead of using fresh water cooling towers.

One datacenter doesn't use appreciably that much water; and we don't know what cooling paradigm they're implementing. If you want to know about water, you have to ask. So far, datacenters haven't broken the grid (cough virginia) yet. In TX planned DCs are projected to consume 5x current electricity generation.

If we use electricity as a proxy, the problem isn't any particular datacenter it is obviously scale.

Also: people lie about water use. Over a decade ago, a day of searching for The Data led to exactly one defensible conclusion: if every household in CA ate 16 less almonds a day they would solve their water shortage... at least on paper. BTDT research.

Edit: Power generation uses the most water. Color me not surprised.

This article would be much more effective by first acknowledging that all datacenters can be built and powered in different ways varying by location, climate and availability of power.

Generalizations unfortunately serve the subtle purpose of many people who are pro or anti datacenter, ironically while they use datacenters to write articles and use datacenters to post on social media, as well as consume video and nearly anything else.

For example the style and approaches datacenters built in desert like regions, which can be in the US, often misinform the overall understanding and conversation in this space.

Direct and indirect water use is an issue in many areas, its peculiar why it's being brought up when it's not comparing datacenters as a percentage of overall water use in an area.

Texas Governor Greg Abbott coming out and saying on national TV that tech companies "dug their own graves" on data center construction was a major wake up call[0]. The more research that is done on these developments, the more damning it looks.

People want to compare this to the pre-industrial rail buildouts, and sure there are some parallels, but I'd be curious to know if that period had the widespread political resentment towards to infrastructure construction that we're seeing today. Because I can also envision a scenario in which the capacity just doesn't get built since being pro-data center is becoming a political death sentence. And what happens then?

0: https://www.businessinsider.com/greg-abbott-texas-data-cente...

I want to offer up a point for reasonable debate here: Suppose you are against new data center construction as a (clumsy) policy lever to regulate the pace of AI progress and harms.

Do the facts matter? If the water issue is “fake” or overblown, but it’s resonating with voters who are packing town halls and ultimately effecting the change you want (a broad slowdown in the construction of new compute) is that a problem? Congress isn’t going to get off its ass. There is momentum here. Who actually cares if people are making misinformed choices if the outcome is correct? Isn’t this how literally all politics works?

(If you think the outcome is not correct, that’s a different debate entirely.)

Just build datacenters right ontop of fracking oil fields like North Dakota

I am not kidding

They have endless cheap or even free power from the months/years of offgassing burning off methane, it can be seen from space

* https://www.npr.org/sections/krulwich/2013/01/16/169511949/a...

And the water (and air) is already so toxically polluted from the fracking that no-one can live near or drink that water or they will get sick

It's a perfect match, again not kidding

Datacenters only need a few people to run it and they can live in highly filtered rooms like hospitals or cleanrooms, heck with bots+ai they won't even need people to run them eventually