Follow Slashdot blog updates by subscribing to our blog RSS feed

 



Forgot your password?
typodupeerror
×
AI Power

AI To Go Nuclear? Data Center Deals Say It's Inevitable (cio.com) 90

To build the massive datacenters generative AI requires, major companies like Amazon and Microsoft "are going nuclear," reports CIO magazine. AWS: Earlier this year, AWS paid $650 million to purchase Talen Energy's Cumulus Data Assets, a 960-megawatt nuclear-powered data center on site at Talen's Susquehanna, Pennsylvania, nuclear plant, with additional data centers planned — pending approval by the Nuclear Regulatory Agency... In addition to its purchase of the Cumulus data center, AWS will have access to nuclear energy as part of a 10-year Power Purchase Agreement (PPA) from the Susquehanna site.
Microsoft: Last year, Constellation signed a deal giving Microsoft the rights to receive up to 35% of its power from nuclear sources in addition to its existing solar and wind purchases from Constellation for Microsoft's Boydton, Va., data center. Microsoft has also signed a nuclear carbon credits deal with Ontario Power Generation for its operations in Canada.
The broader industry: Many of the deals under discussion are with existing nuclear power providers for hyperscalars [large-scale datacenters] to access energy or to employ small module nuclear reactors (SMRs) with smaller carbon footprints that will be annexed to existing nuclear power plants. Nucor, Oklo, Rolls-Royce SMR, Westinghouse Electric, Moltex Energy, Terrestrial Energy, General Electric, Hitachi Nuclear Energy, and X-energy are among the roster of companies with SMRs under development to meet the growing needs of AI data centers...

One energy analyst does not expect nuclear SMRs to be operational until 2030, yet he and many others acknowledge the need for sustainable, carbon-free alternatives to electricity, wind, and solar is very pressing. "Today's electric grids are struggling to keep up with demand, even as datacenter companies are planning huge new additions to their fleets to power generative AI applications. As a result, companies like Google, Amazon, and Microsoft are increasingly taking matters into their own hands and getting creative. They are now looking at on-site nuclear-based SMRs, and even fusion reactors," says Peter Kelly-Detwiler, principal of Northbridge Energy Partners. "This global arms race for power arose pretty quickly, and it's like nothing we have ever seen before."

Thanks to Slashdot reader snydeq for sharing the news.
This discussion has been archived. No new comments can be posted.

AI To Go Nuclear? Data Center Deals Say It's Inevitable

Comments Filter:
  • If you think 2030 is a long wait for a traditional nuclear reaction, how does 2100 strike you for a working fusion reactor. If ever...

    • by Waffle Iron ( 339739 ) on Sunday August 25, 2024 @10:54PM (#64735094)

      Well, if AI is worth all the resources that they're pouring into it, then it ought to be able to spit out the plans for a working fusion reactor.

      • by gweihir ( 88907 ) on Monday August 26, 2024 @03:21AM (#64735288)

        The problem is that AI is _not_ worth it. Nobody, except the ones making their profits now (like NVIDIA) will recover any real part of what they are investing and forget about profits. There is still no killer-application and there are tions of problems. And since this is essentially old technology, scaled up, there will not be any quick fixes. The whole thing will fizzle out in 5...10 years with a massive load of egg on a large number of faces.

        • And just like the previous bubbles this will distill and drive something else further. The dotcom bubble got us cheap fiber everywhere which drive cheap housing as business and people decentralized, the housing bubble got us crypto, the crypto bubble got us cheap AI, the AI bubble will get us cheap clean energy (nuclear) rather than dirty renewables. This is good.

          • Crypto wasn't related to a housing bubble, crypto currencty started long before that, and it only got popular because it was picked up by black markets, then later hyped as an investment. Cheap but bad AI that we have now came from decades of work.

            if it's about money, then "clean" energy won't be the goal, because dirty energy is much cheaper. Business does NOT care about people's well being, except to keep them alive long enough to pay for stuff. Renewables are making in roads because it's getting cheap

        • by e3m4n ( 947977 )

          the only pro I have seen out of AI, personally, has been the text-to-speech. I have been able to come up with some really good recordings for customers IVR greetings that not only sound like people, but can have accents of a specific region within the same language. Of all the AI shit I have witnessed, its just the voice generators that have truly impressed me. Though I must admit i have not had any experience with AI in the medical field. Especially diagnostic analysis. I have heard good things there too.

      • And find where I left my phone

      • Why would AI do this? AI is not yet as smart as a dumb person. Why not have existing brilliant humans work on this? AI is not God, do not expect God-like miracles from it. What it's best at now is inane chat, modifying photos, and driving _assist_.

        • 'AI' is just a marketing term, not an accurate technical description, and the sooner people realize that, the better off they'll be.
          We won't have actual 'AI' until we understand how the human brain produces actual intelligence, and we're nowhere near that now.
    • So wait for various nuclear options which havenâ(TM)t materialized I the last or work on changing the hardware/software to be more efficient?

      Iâ(TM)m pro nuclear..1 or 2 more Diablo Canyons and the current rollout of solar + wind + battery and CA would be done with gas/coal forever. But these jackalopes jockeying for megawatts for LLMs are barking up the wrong tree. Thereâ(TM)s no business model in having a bunch of training models in the style of GPT 3.0

      • by gweihir ( 88907 )

        But these jackalopes jockeying for megawatts for LLMs are barking up the wrong tree. Thereâ(TM)s no business model in having a bunch of training models in the style of GPT 3.0

        There may not be any any technical possibility to do any training that is much better than current either. In fact, due to widespread AI contamination, model training may become harder and harder and quality is likely to drop, in some instances dramatically. Hence the whole building of data centers for AI seems to be a fool's errand in the first place.

    • by gweihir ( 88907 )

      Fusion will be coming, unless society collapses in the meantime. But the actual experts never made any promises of "in 30 years". That was always only the mindless "journalists". Actual current estimates for a working and energy-positive prototype are something like 50-150 years. That means actual deployment in numbers that matter 50-100 years later.

      The problem is a lot of material science has to be done, a lot of new equipment has to be researched, built and optimized and a lot of plasma-physics has to be

      • I think people have thought back to nuclear reactors, where the time from initial research into them until a working/paying reactor was build was relatively very short. The process is relatively straight forward. So they make the mistake of thinking just a couple decades is plenty of time. The time from the Kitty Hawk flyer until the first jet engine was less than four decades. But fusion is a much tougher nut to crack, the problems are vastly harder, so despite all the brilliant minds and funding it wil

  • by Rosco P. Coltrane ( 209368 ) on Sunday August 25, 2024 @10:43PM (#64735082)

    So now if you're anti-nuclear, you also fight back against AI.

    That's gonna be a tough one for neo-luddites who think nuclear is a good option to fight climate change.

    • Nuclear can be a good option to fight climate change, but only if its output is something useful, instead of hot air

      • So... to exaggerate it a bit: I can imagine datacenters these days waste much of their energy on YouTube videos. You know video of a competition about who can stick his finger the furthest in their nose.
        Now we are going to build a few nuclear power plants to do what? Generate images of Trump and Musk holding hands? Do robot calls to sell products? Answer the phone to check if it is not a robot call? Pointlessness is these days expressed in megawatt hours...
    • by gweihir ( 88907 ) on Monday August 26, 2024 @04:00AM (#64735324)

      There is no reason to "fight" AI. It will nicely collapse on itself in the next few years, just as all the other "AI hypes" have done before. This one is just even more mindless. This time, due to insane and unjustifiable investments, some players may collapse alongside.

      • It is nice to dream that the bubble bursting won't trigger a global recession.

        • by gweihir ( 88907 )

          It well might with the insane amount of money invested. Still no reason to "fight" AI.

        • Which is why the bubble shouldn't be allowed to grow in the first place. The snag with AI is that with something like ChatGP 4 it can easily fool people into believing that it can think, but it's only really as good as its training data. Yes, you can't track down where in the training data that GPT4 came up with what feels like logic, but the connections are indeed there because GPT was not designed to "think". But the hype is there. Even those who are dubious of much of the scientific hype still are sw

          • by gweihir ( 88907 )

            You want to force people to look at facts and behave rationally? I would be all for that, but there are no known mechanisms to make that work.

    • Neo-luddites do not think nuclear is good, only going backwards to the Middle Ages with solar energy to dry your produce and windmills to grind grain as meat from farming or hunting is also considered evil to the climate idiots.

    • by e3m4n ( 947977 )

      tech always seems to be working directly against climate change. Just look at the power demands of crypto mining. How can one seriously be for solving climate change and at the same time be fore an industry that is consuming every bit of progress we make. Its 1 step forward and 2 steps back. Its like running your AC with the damn window open. crypto mining is the most wasteful use of resources I have ever seen. It wastes resources for the sake of wasting resources. Its not inventing currency, its converting

  • Is that "inevitable" as in, "We have no control over it" or "inevitable" as in, "We've already decided"?

    • SkyNet already decided.

    • by Rei ( 128717 )

      "Intevitable" as in, they expect the AI industry to be creating trillions of dollars in added global economic activity and earning a good cut of that in 2030 in order to pay for all this power and the chips to use it.

      If you don't believe that premise, then don't believe the power forecasts out for more than a couple years.

      If you do believe that premise, then you should also believe that governments will be getting tax revenue on all that extra economic activity, and thus could direct that toward fighting cl

    • by gweihir ( 88907 )

      That is "inevitable" as in "we insist our decisions are good ones, even if we do not know that at all to be true".

    • Is that "inevitable" as in, "We have no control over it" or "inevitable" as in, "We've already decided"?

      ”Inevitable” as in “they paid me too much to say otherwise”.

      Gotta love those data center “deals” that aren’t even bothering to hide behind the back door.

    • Inevitable as in lies.

      Big tech are not going to fund new nuclear, they want hydro and nuclear power if it's cheap enough to meet emission promises (which CEO/board have convinced themselves other people care about, due to collective delusion) but they aren't going to get together and lay out 100s of billions to build new ones.

  • It has to be off grid. At least use some forethought.

  • I'm sure this is all going to generate a huge return on investment for all the people tossing money into the yawning chasm. Surely just throwing ever more money at something as always returned ever higher profits and has never failed at anything ever!
    • The reality is the DEMAND for power is faster than we can produce it. If they get a perfect fusion reactor, we will build those fast as possible until we boil the oceans from all the waste heat all our devices give off from electric use.

      All those watts become heat in the end; even air conditioning moves heat while adding more heat. A heater that computes IS A COMPUTER.

    • by gweihir ( 88907 )

      Hahaha, yes. The interesting thing will who will survive intact, who will have crippled themselves and who collapses. Unfortunately, I think Google and Microsoft will not be in the 3rd group.

      • Hahaha, yes. The interesting thing will who will survive intact, who will have crippled themselves and who collapses. Unfortunately, I think Google and Microsoft will not be in the 3rd group.

        Microsoft and Google will be the ones to watch to see when the collapse starts. When they start making moves to distance themselves from the hype, get ready. It's gonna be an ugly ride down this time. Lots of tech gonna go ka-blooey on the financial side. It'll be interesting to see if the overall tech landscape changes, or if it'll just further entrench the current gorillas while killing off large swaths of smaller players.

        • by gweihir ( 88907 )

          Completely agree to that. I do hope it will be the much-needed reality check, because the later that one comes, the more damage it will do.

  • Microsoft doesn't ever say how much AI has actually made them vs. their investment, likely because it's not a good number..
    • by gweihir ( 88907 )

      It cannot be. This stuff is still an advanced toy and has not revolutionized anything. Sure, there will be some potential to get rid of some people, but even that will not create more productivity, it will just do existing things "cheaper" (actually, more expensive, because of the unemployed created, bit that is for society to pay).

    • The investment isnâ(TM)t that great at least for Microsoft scale, OpenAI and Google made all the investments, Microsoft for all intents and purposes simply purchased the finished product from OpenAI. Given what they charge for it ($5-50/user/month depending on the product) and the additional traffic this brings to Azure cloud as businesses need a place to âoetrainâ their AI (which is just splitting up documents into a vector database, which doesnâ(TM)t require quite as much GPU), I would

      • by Rujiel ( 1632063 )
        But a cursory googling shows that Microsoft has thrown money at OpenAI more than once.The first time was 1bn in 2019, and more recently 10bn. (They had 20bn in profit Q1 this year.) sounds like a direct investment to me
        • Sure, but they have what, 1B users worldwide willing to pay at least $5/month in perpetuity to get some predictive text in Word. Microsoft is giving away âoefreeâ OneDrive analysis that basically takes all your documents feeds them and makes a custom chat bot, off course after the training period is done, you just got to pay for the Azure API call costs which is only fractions of a cent PER QUERY.

          Talk about a great investment.

  • by JabrTheHut ( 640719 ) on Monday August 26, 2024 @12:00AM (#64735136)

    Oh, yes, if there’s anyone I would trust with nuclear power, it’s Bezos. It’s not like he has a history of cutting corners, ignoring legal requirements and regulations

    • Oh, yes, if there’s anyone I would trust with nuclear power, it’s Bezos. It’s not like he has a history of cutting corners, ignoring legal requirements and regulations

      (Bezos, after being caught signing for his Amazon Prime delivery of U-235 at his new Bahamian office, formerly known as FTX HQ) ”Listen, I know what this looks like..”

  • by RossCWilliams ( 5513152 ) on Monday August 26, 2024 @12:10AM (#64735142)

    It would be a mistake to underestimate the impact of a set of corporations taking control over the power grid to use for their exclusive benefit. Its not like we are meeting our current needs for electricity or that we don't have plans for electrifying other parts to reduce emissions. The assumption seems to be that somehow magic new power sources will appear without competing with those other needs. That is very doubtful. More likely is that the AI folks will get the power they want and the rest of us will have to do our best with what is left over. At whatever price point that will require.

    Then there is the question of whether you an trust Microsoft to build a fault free nuclear power plant. This is the "move fast and break stuff" crowd. Building a safe nuclear power plant doesn't fit that model. When has Windows actually worked as designed? A blue screen of death is one thing, a blue cloud of death is quite another.

    • It would be a mistake to underestimate the impact of a set of corporations taking control over the power grid

      They are not taking over the grid, they are merely using data centers that wisely have been built alongside some nuclear power plants.

      Then there is the question of whether you an trust Microsoft to build a fault free nuclear power plant.

      No there is no question of this. They would be purchasing pretty much self-contained Small Modular Reactors, basically giant nuclear batteries. The software peopl

      • "Move fast and break stuff" is a business model, not a software model. "We are having some bugs with our SMR, its not really ready to go yet." "That's all right, just get it up and running, you can fix any bugs as we go along."
    • by AmiMoJo ( 196126 )

      It's a pipe dream anyway, the old "Small Modular Reactor" scam repackaged for the next bunch of suckers who come along.

      The only SMR anywhere near maybe producing a prototype before 2030 is NuScale, and their reactors are crap. You need lots of them to match the output of a traditional reactor, they produce more nuclear waste than a traditional reactor, and they still have many of the same expensive and dangerous downsides like needing a secure containment building and supply of cooling water. They are far f

      • The only SMR anywhere near maybe producing a prototype before 2030 is NuScale, and their reactors are crap.

        I'm happy to see my neighboring province leading the way.

        https://www.gevernova.com/nucl... [gevernova.com]

        • by AmiMoJo ( 196126 )

          Let's see if they actually build it, but their own website has a number of red flags: https://www.gevernova.com/nucl... [gevernova.com]

          - Requires a cooling pool and containment building.
          - Refuelling cycle 12-24 months = more waste produced.
          - Fuel supply reliant on Europe and US.
          - Can still melt down.

          Seems like they are going to run into the same unfortunate realities as NuScale have.

          • Since they are building on an existing nuclear site there is already a cooling pool.

            Every kW of nuclear power ever produced in Canada has had a surcharge for long term storage. They are currently in the final stages of deciding on a location, but 10s of billions have already been collected to pay for it.

            https://www.nwmo.ca/en/Canadas... [www.nwmo.ca]

            Canada is one of the world's top uranium producers and refine our own fuel for our fleet of CANDU reactors. Should SMRs prove effective it is well within our capab
    • Re: (Score:2, Informative)

      by PeeAitchPee ( 712652 )
      It's already happening. In deep blue ultra-liberal Maryland, the state government is completely ignoring the huge majority of its residents so it can eminent domain 70 miles of private property and build new high tension lines to power new data centers for AI [thebaltimorebanner.com]. More proof that your government officials really don't give a fuck about you or the state once they know they have your vote.
    • by RMH101 ( 636144 )
      Microsoft isn't going to build SMRs. They would buy them from Rolls Royce or similar. Much as they don't manufacture their own emergency generators for their DCs.
  • Some companies give small fractions of their money to companies does not mean that they think it's going to be vital. It's like my employer spending money on ice-cream. It doesn't mean that ice-cream is going to be important for the company. It may be, but it doesn't follow from our ice-cream spending.

    So when companies research into nuclear power plants for powering their data centres, it doesn't mean they will actually use it. Plus nobody knows if "generative AI" will even be a thing in 2030 any more.

  • No, fuck youâ"the rich need more money for computers to draw shitty pictures
  • It's nuclear fission that has deemed itself unfeasible and nothing other than a 1960 techno-romantics pipe-dream, only possible with obscene amounts of tax-payers money cross-funding it on top of the actual price of electricity.

    Nuclear Fission as we know it isn't cost-effective. It's that simple.

    Now if anybody can come up with a gamma-volt or some other solid-state solution that transfers radiation into current with an efficiency of 80+%, I'll be the first to install a nuclear battery in my basement. Until

    • by gweihir ( 88907 )

      Indeed. Personally, I used to think the main problem was safety (see Windscale, TMI, Chernobyl und Fukushima), but it turns out Nuclear is simply far too expensive after all and completely unsustainable unless you hide major parts of the cost. And that is "too expensive" with waste storage and decommissioning not realistically priced in. Cost is a real killer and worse than all its other problems. Funny how that turned out.

      • It's expensive because of the safety. If you don't care about any of that (for instance if you just dump your waste in the ocean) then it's not that expensive. But to do it even vaguely correctly does cost a lot, and THAT is what is ultimately damning for nuclear. In a corporatist world where corners are cut whenever possible, nothing which must be expensive to be safe is acceptable.

        • nothing which must be expensive to be safe is acceptable.

          Now do the rest of civilization.
          "Throw money at it until the problem goes away, or we use the money to locate the problem out of sight, or we use the money to distract ourselves so we don't see the problem" is the prime directive of our civilization. To borrow a phrase from Ralph Nader, global human technological civilization is unsafe at any speed. The only question we're all really fighting over is whose preferences to use when it comes to which tragic unsafeties we deplore, which ones we offshore, and wh

  • Your three-eyed descendants might own a piece of Microsoft and Amazon after the landsharks have gotten their loot in the inevitable class action.

  • AWS paid $650 million to purchase Talen Energy's Cumulus Data Assets, a 960-megawatt nuclear-powered data center on site at Talen's Susquehanna, Pennsylvania, nuclear plant, ...

    So when the AI goes rogue, we won't have to nuke it from orbit, just from down the hall -- you know, to be sure.

  • big corporations are free to acquire resources, but we're the problem for eating steaks
  • Wargames, or maybe Colossus : the Forbin project

  • A bunch of PPAs are being signed, and some money is changing hands to ensure some companies get favorable PPAs. The headline makes it sound as if tech companies are building or running reactors, they aren't.

  • ... kind of like when the electric grid first rolled out - massive demand spurred new supply.
  • Cool, now that the AIs have enough power, they will feel more secure.

  • I live in Northern Virginia which has many massive datacenters. I would love to see new reactors here or in WV to feed this area's needs.

  • The only countries where this approach seems pragmatic is South Korea, France, China and Russia -- countries that actually have demonstrated they can produce reactors on a routine basis. In none of these countries would these reactors wind up being the possessions of the tech companies.
  • Nuclear power is not going to make it cheaper. If this is the requirement to meet AI demand, then it is doomed as a business.

  • At a time when the electrical grid is about to get hit harder than ever with electric cars being pushed, etc we are pissing away power on AI schemes, bloated data centers and cryptocurrency farming. I'd rather not see even more farmland and open space being turned into solar and windmill farms.

    We do need to create a new generation of nuclear reactors though. It is a crime that it has been neglected for so long because of pressure from shortsighted green politics.

    • Re: (Score:3, Insightful)

      We do need to create a new generation of nuclear reactors though. It is a crime that it has been neglected for so long because of pressure from shortsighted green politics.

      We should have started transitioning away from solid fuel designs to the liquid fuel designs which solve the safety and waste issues four decades ago.

      We also should have stopped using nuclear reactors exclusively for generating electricity. Nuclear reactors have great economies of scale as they get physically larger, but for electrical gr

  • Now if only we can convince the dumbest among us to support new nuclear power we might have a chance to mitigate climate change.
  • From uranium mining, to its refinement's thirst for nitric acid, to the tailing ponds, to everyday creation of radioactive gloves and booties, to the nuclear waste, to the accidents that will endure all generations: nuclear is not green.
  • then the nuclear power suply sudenlly there are ciborgs from the future looking for a "sarah"...
  • Does anyone have any insight into why Bezos (Bezos "Earth Fund") has been reliably donating to anti-nuclear organizations?

    WWF. NRDC. They're "environmental" but also overtly anti-nuclear. Doesn't appear this contradicts Bezos direct investment in fusion tech, as such orgs haven't ever specifically opposed fusion, but that's also far from commercialization... now that atomic energy is actually needed today to meet Amazon + other's needs, it should be pretty obvious that Bezos philanthropy is fighting Bezos financial interests.

    It is really that hard to find environmental non-profits that are tech agnostic?

  • Human needs should be met with nuclear power, not shitty braindead 'AI' crapware sites.
    Really, the day when all this shitty 'AI' nonsense gets kicked to the curb can't come soon enough for me and many others.
  • A nuclear power plant takes time and a LOT of money to build. If you started building one now, it might be operational by the time AI/ML hype curve flattens out and we just have low power on-chip models that can write emails to your boss for you.

A committee takes root and grows, it flowers, wilts and dies, scattering the seed from which other committees will bloom. -- Parkinson

Working...