Amazon Joins Push For Nuclear Power To Meet Data Center Demand (reuters.com) 47
Amazon said on Wednesday it has signed three agreements on developing the nuclear power technology called small modular reactors, becoming the latest big tech company to push for new sources to meet surging electricity demand from data centers. From a report: Amazon said it will fund a feasibility study for an SMR project near a Northwest Energy site in Washington state. The SMR is planned to be developed by X-Energy. Financial details were not disclosed. Under the agreement, Amazon will have the right to purchase electricity from four modules. Energy Northwest, a consortium of state public utilities, will have the option to add up to eight 80 MW modules, resulting in a total capacity up to 960 MWs, or enough to power the equivalent of more than 770,000 U.S. homes. The additional power would be available to Amazon and utilities to power homes and businesses. "Our agreements will encourage the construction of new nuclear technologies that will generate energy for decades to come," said Matt Garman, CEO of Amazon Web Services. SMRs will have their components built in a factory to reduce construction costs. [...]
Amazon said it is also leading a funding round for $500 million to support X-Energy's development of SMRs. Amazon and X-Energy aim to bring more than 5 gigawatts online in the United States by 2039, which the companies call the largest commercial deployment target of SMRs yet. Amazon also signed an agreement with Dominion Energy, opens new tab to explore the development of an SMR project near the utility's existing power station in Virginia. The about 300 megawatt project would help meet power needs in a region where demand is expected to jump 85% in 15 years, Dominion said.
Amazon said it is also leading a funding round for $500 million to support X-Energy's development of SMRs. Amazon and X-Energy aim to bring more than 5 gigawatts online in the United States by 2039, which the companies call the largest commercial deployment target of SMRs yet. Amazon also signed an agreement with Dominion Energy, opens new tab to explore the development of an SMR project near the utility's existing power station in Virginia. The about 300 megawatt project would help meet power needs in a region where demand is expected to jump 85% in 15 years, Dominion said.
Re: (Score:1)
Re: (Score:3)
I will say that both sides tend to have naive arguments that rely more on oversimplification than real-world facts. The pro-nuclear people aren't able to admit that nuclear problem has any problems, the anti-nuclear people aren't willing to admit that problems can have solutions.
Re: (Score:2, Interesting)
That's a crazy oversimplification.
There is a range of anti-nuclear people. Some are exactly what you say. Others are pragmatists that look at the bottom line - nuclear is too expensive in comparison to other zero-carbon generation options unless some other factor comes in, such as using a nuclear plant that already exists (uprating, license extension) which you cannot do forever, or you can somehow monetize the spectacular amount of waste heat generated - I don't think that's going to be a thing in a data
Re:Popcorn ready (Score:4, Informative)
OK, I will add you to the list "anti-nuclear people who aren't willing to admit that problems can have solutions."
However, go ahead and scroll down to my other post in this topic, where I point out that nuclear power is not cheap, and go argue with the guy who says that they are.
Re: (Score:2)
He stated a problem (cost). He implied that since it hasn't been solved, it can't be solved. That puts him in the list of people who aren't willing to admit that problems can have solutions.
What problem do you have with that?
The 1980's called (Score:2)
Crap that's taking over the world, enjoy the decline.
They want their expert systems back.
AI is on par with fusion. It'll be great when it finally gets here. Whenever that is. In the meantime, startup ventures have perfected the art of squeezing funds out of suck- .. investors to keep their latest fad alive.
No thanks. I don't need to be in on the ground floor. I'll wait until the technology matures. That would be an "AI engine" that consumes roughly 20 Watts, not megawatts.
Re: The 1980's called (Score:4, Insightful)
I'll wait until the technology matures.
As a buyer of AI you might wait.
As a consumer in general, you will get AI instead of customer service whether it's ready or not.
Tech people with nuclear reactors (Score:2)
Re: (Score:2)
"Well that reactor melted down. Write off the loss and let's move on to the next idea!"
Re: (Score:3, Funny)
You mean we can't just reboot it?
Re: (Score:2)
The chances are pretty good that the AI bubble will burst, or someone will find a more efficient way to do it, or both, before any SMRs even reach prototype stage.
Re: (Score:3)
NuScale already has an NRC-licensed design.
Of course it's too expensive for anyone to actually want to build in the US, but they are licensed. That's beyond prototype stage.
They were going to build a production plant in Utah until they couldn't find any buyers for the energy because it was going to be too expensive.
Re: (Score:2)
The licence is to build a prototype.
Re: (Score:2)
Old and conservative is the opposite of brute force, or, as known today, "artificial intelligence"
Re: (Score:2)
I think there is enough regulation that they will have to resort to good old (but modern) engineering. I am more concerned by what they will run on while the power plants are being built and what happens when the AI bubble bursts.
I am afraid that nuclear is being used as a carbon-free promise that they won't keep anyway.
Re: (Score:3)
This is someplace I'd prefer old conservative engineering and not a move-fast tech-person approach.
I remember when this exact argument was used for having NASA run every aspect of space development. Human lives ride on and around those rockets, after all.
Re: (Score:1)
So much for conservation (Score:2, Insightful)
I like how we've been pushed to green energy sources, closing nuke and coal plants, etc just in time for these massive data centers and bitcoin mines to suck up all of the savings.
Re:So much for conservation (Score:5, Insightful)
Re: (Score:3, Informative)
Green apologia just keeps repeating itself, regardless of how much the lie is debunked.
1. Closure of nukes is one of the primary Green goals. In many cases, it's THE cause around which the Green movement started (example: Germany). As a result there are no causes more anti-nuclear power than Green.
2. Nuclear is the cheapest power source we have. The problem is bureaucracy that came in the wake of Green activism, goal of which was making it impossibly expensive to build. This is why power plants built in les
Expensive power, but arguably worth it (Score:3, Insightful)
...2. Nuclear is the cheapest power source we have.
Unfortunately, no it's not. Real-world experience is that nuclear power plants are pretty much the most expensive power source we have [duckduckgo.com].
Pro nuke people keep asserting that it could be cheap, but the actual cost of real-world nuclear power plants continues to be expensive.
The argument is sometimes made that, even though it's expensive, the benefit of low-carbon-dioxide-emissions power means we need it anyway.
Re: (Score:3)
Notice the inversion of reality. People who observe reality before and after Greens did their jobs in bureaucracy are "asserting", and people who know that their movement has done everything in their power to maximize costs cut those numbers off and go with "but these latest ones where our bureaucracy has inserted all the poison pills into the system to make them exceedingly expensive, they are!" are the ones who need not prove anything because...
"Cost to build a nuclear power plant" now, after Green poison
Re: (Score:1)
I didn't at any point claim that "nuclear power will be expensive always and forever." I said that real-world experience shows that nuclear power plants are expensive. These two statements are not the same.
If you scroll up to my other post on this topic, I said "The pro-nuclear people aren't able to admit that nuclear power has any problems, the anti-nuclear people aren't willing to admit that problems can have solutions." I stand by that statement, and add you to my list of "pro-nuclear people who aren't a
Re: (Score:3)
It could be cheap if done in a particular way. You can't use the real-world cost of nuclear power plants that weren't built in that way to argue that way won't work; you need to look at similar examples in other fields where they did do it that way.
If you're on /. then you should be familiar with what happens when you try to do an IT project as a Big Project. For another example, look at NASA and SpaceX. NASA keeps building things as one-hit-wonders, whereas SpaceX is going for the factory assembly line app
Re: (Score:2)
Yes, everybody in favor of nuclear power has an explanation of why it's so expensive, and how their proposed approach would solve the problem. It's reasonable to expect that of them may be right. It has yet to be proven which.
My comment didn't say nuclear power can't be cheap. It said that, as of today, it isn't cheap.
Re: (Score:3)
Why can't data centers do their database indexing, AI training and other power-hungry tasks when the sun is shining and generating cheap electricity?
Why SMRs (Score:2)
A datacenter with a >200MW base load seems to make a better anchor customer of a standard reactor. Are these just [neon]greenwashing investments?
Re: (Score:2)
Re: (Score:3)
Usually broken into ~50-100MW halls, but I think Switch is well over 350MW by now. The AI compute facilities can easily hit the 200MW threshold in a single hall. (Hall being a self-contained unit of infrastructure rather than a room typically.)
Re: (Score:2)
Well, I think the logic goes like this:
"We could spend $20B on building a 1000MW rated PWR and 20 years of legal wrangling and construction overruns, or we could build a facility that can house several SMRs that get us what we need without costing $20B and taking 20 years."
Example: NuScale has pre-designed facilities [nuscalepower.com] that can house 12 of their SMRs, which would be around 960MWe output. And because the reactors are modular, you could build the facility to accept 12 reactor modules, but only start with 6 or
If Amazon read Sladhdot... (Score:1)
...they might think to invest in Fervo. Perhaps they could even pump their server waste heat back into Fervo's generators.
Data center power question (Score:3)
I assumed that a data center, particularly an AI data center, would need a lot of power but in a constant flow, particularly for LLM ones that are being used fairly constantly. This would make nuclear power very good, because nuclear power's main drawback is not radiation but rather how it doesn't scale very easily with the daily fluctuations in demand. Economically nuclear is best when you're operating at 90%+ capacity. If a data center had consistent power needs, nuclear would make an excellent option.
The person I spoke to said that constant power demand is actually not a thing in data centers, particularly these AI ones with quite a large number of GPUs. While you might have a lot of ongoing work, you actually have surges in power needs particularly because the GPUs they use for these things spin up and spin down a lot, so you get a lot of spiking of power needs.
If that's the case, then nuclear is actually not a very good option at all. Is that true? Do data centers and AI specific data centers tend to have a variable and peak-y power demand?
Re: (Score:1)
...If that's the case, then nuclear is actually not a very good option at all. Is that true? Do data centers and AI specific data centers tend to have a variable and peak-y power demand?
What are you smoking? Like you can make the wind blow at the same time AI demand spikes? Good grief...
Re: (Score:2)
The person I spoke to said that constant power demand is actually not a thing in data centers, particularly these AI ones with quite a large number of GPUs. While you might have a lot of ongoing work, you actually have surges in power needs particularly because the GPUs they use for these things spin up and spin down a lot, so you get a lot of spiking of power needs.
You may be using different definitions of the time frame of constant power. You were talking about "daily fluctuations in demand." Your friend was talking about spikes in data processing, which would be fluctuations on the scale of many seconds, possibly several minutes. Different thing. For nuclear source, you can easily use capacitors, or batteries, to deal with spikes in power on the "minutes" time scale. It's hours or days of storage that's the problem,.
Four, by my count (Score:3)
Oracle, Microsoft, Google, and now Amazon, all mouthing stuff about nuclear in the last 30 days.
We need a SpaceX type of company for nuclear power (Score:2)
One that is able to reduce costs as they've done, without compromising safety.
https://x.com/WholeMarsBlog/st... [x.com]
This is staggering.
Re: (Score:3)
One that is able to reduce costs as they've done, without compromising safety. https://x.com/WholeMarsBlog/st... [x.com]
Since the SpaceX approach is to blow up two or three before getting one to work, this may not be a good approach.
Or maybe it is. Maybe we worry too much about a few nuclear power disasters, and should just accept them as part of the cost of maintaining our civilization.
Re: (Score:1)
As a teenager I was very surprised to learn that the USA had been testing nukes in the Pacific before they blew up Japan. Multiple casual nuclear explosions, just for data? And no ice age? We're alive? The planet did not explode? Godzilla did not rise from Bikini Atoll? No giant asteroid summoned by Gaia?
Nowadays I am smarter than in my teenage years. I can tell that civilization has indeed regressed a lot due to all that radiation from US nuclear detonation. For example over the last 100 years we have basi
US homes are down to... (Score:1)
1250 watts per house. That;s fantastic progress.
Missing from the models? (Score:2)
I added the question mark because I honestly don't know the answer. Nuclear power is better than burning carbon, because the only greenhouse gases associated with it come from mining, refining, maintenance efforts, etc. But at the end of the exercise we're still creating heat.
I read lots about CO2 and methane emissions, but I never hear anything about all the heat we're putting into our atmosphere - heat which is also trapped here by greenhouse gases. Does the sun's heat input so dwarf the heat we humans cr
Am I the only one who thinks this is insane? (Score:2)
I't all just pants-on-head stupid.