AI Data Center Need Nuclear Power- Lots of It

31 thoughts on “AI Data Center Need Nuclear Power- Lots of It”

1. Forget nuclear. The chips will become more efficient and require less power. In the mean time locate in areas with hydro plants that way you will have cheap power. Solar with batteries can be built faster and cheaper than nuclear and lasts just as long with lower operating costs.

2. Most of the advanced reactors are the children of earlier approaches that failed commercially for various reasons, including being too small to be economically viable. That problem remains but the DOE continues to cheer lead the efforts by pumping in billions and billions of taxpayer dollars. The primary beneficiaries are not the consumer, but the DOE labs themselves.

   If a large amount of energy is required, build large conventional reactors that are well proven and reliable. The advanced reactors are neither.

   Vast unnecessary overregulation is a huge financial problem for nuclear energy – roughly causing a 30% premium on construction costs. Much of that problem would dissipate if the Nuclear Regulation Commission would comply with recent Congressional law. However, the NRC staff continues to churn out ever more convoluted and complex regulations. I would normally expect the NRC commissioner to step in since NRC executives are not doing their job, however there are not many commissioners left.

   In passing, conventional water reactors need less uranium ore than most of the advanced reactors that require much more enriched uranium (HALEU) to function. The former do not need the billions of taxpayers dollars being poured into HALEU facilities, existing facilities are adequate, although some expansion would be wise. Did I mention we import most of the uranium needed for nuclear power?

   In closing, the nuclear industry has taken a page out of the renewable energy playbook. Billions and billions of taxpayer dollars are subsidizing advanced reactors.

   • Amen

   • Fossil fuel also gets billions of taxpayer dollars in subsidies, that does not get mentioned much for some reason.

     • Might want to re-think that claim. The Federal budget over the last several years shows green energy got most of the direct subsidies. Attempting to ferret out the various other types of help (direct and indirect) is nearly impossible. Would be wiser to stay out of that subjective minefield.

       • Not really because it is a fact. You can argue over where the subsidies come from (state, federal whatever) and who got what when. For instance I would agree that under Biden’s term he boosted subsidies to green energy massively. But then Trumps terms cut back subsidies to green energy massively. But overall and over decades fossil fuel has had many billions of tax dollars of subsidies, it is just a fact. Why do you think the fossil fuel companies give so much money to US politicians? Its not charity, easy to Google the figures.

   • Terrestrial Energy seems to have a “promise” of a good solution. They use 5% enriched Uranium, burn up most of the actinides – producing less waste and have a long run time before swapping out units.

     Of course molten salt is an engineering challenge so who knows if they can pull it off. And who knows how much taxpayer money they have burned through.

     • We’ll they’re in the running with 11 other developers under the Executive Order 14301, which allows test reactors to be built under DOE review without NRC license.

       “….advanced reactor that satisfies thresholds established by the Department sufficient to demonstrate that, from the perspective of technical development and financial backing, the reactor may feasibly be operational within 2 years from the date a substantially complete application is submitted….The Secretary shall approve at least three reactors pursuant to this pilot program with the goal of achieving criticality in each of the three reactors by July 4, 2026.”

       Words on paper. Strong words.

       • The whole DOE effort is utterly decoupled from economic reality. The entire effort never considered the competitive position of advanced reactors relative to other energy production methods (gas, renewables, etc), let alone among the various proposals. That is a fact, as proven by reading the various grant solicitations.

         The entire effort is a vast make-work operation for the DOE, with the taxpayer being dammed. Thanks President Trump. Not surprised Musk left in utter disgust by the hypocrisy of the administration.

         The NRC has major issues as far as overregulation is concerned, by the law is crystal clear that the are responsible for regulation commercial nuclear power, not the DOE. Hope somebody drags the Trump administration into court for defying the law.

     • Terrestrial Energy’s reactor coolant is also home for the nuclear fuel. Just what the operating crews need, nuclear fuel wandering around in the piping systems.
       I am truly weary of the DOE resurrecting zombie technologies from the past with zero regard for the taxpayers funding the inevitable debacles.

3. No. Nuclear is intellectually the perfect choice, but unless there is strong public support for it, it is not a good choice. For Example, Germany had many safe, efficient, well managed nuclear plants but the public turned against nuclear power and they then closed the plants.
   The best solution is a Solar facility integrated into an actual farm. Modern environmentalist will complain about the environmental impact of building a solar facility even if you build in the middle of the desert, but even the most insane environmentalist realizes that we need farms.
   Solar farm with battery packs is the fastest power facility to build and has the least public disapproval.

   • Solar makes some sense in a broad band either side of the equator where seasonal variations in sunlight are modest. Though you want something else for running things at night.
     In high latitudes like Canada or Europe north of the Alps solar is useless because your energy needs are greatest when you have the least sunlight.

     • Clearly all industrial/commercial buildings and parking lots should be covered in solar panels, and in times near peak generation, the excess power should be switched into aluminum or H2 production the moment it would require baseload nuclear to derate.

       None of this is complex.

       A thousand 5-15MWe microreactors is obviously not preferable to 4-12 AP1000. It’s just dumb, and the TRISO fuel is very expensive.

       • Creating hydrogen is thermodynamically and economically out-of touch with reality. If burning a gas is necessary to produce work, use natural gas. Vastly less expensive.

         • Is it less economical than grounding out excess power?

           Renewables can drive the cost of electricity negative.

           If you don’t agree, propose another solution.

           Bill Gates attempting to load follow a sodium fast reactor is cockamamie.

           • Not sure what we are disagreeing on.
             In the past, if I wanted to dump unneeded power into the grid, I would have to pay money to the grid operators. We would did not run the plant. Same market principle should be applied to unneeded green energy.

           • As far as Gates is concerned, I agree.
             Why on earth would a utility want a reactor whose coolant catches fire if exposed to air, turns to rock if the temperature drops below about 400 F, and can cause violent steam explosions if exposed to water.
             I have been in the power business for over half a century and have started up and run a large number of power plants. I would not go any where near Gate’s reactor for fear of me and my crews being at mortal risk of death and injury.

       • TRISO fuel can be pricey, particularly if a nuclear plant using TRISO does not employ a containment structure. The needed quality assurance and regulatory measures for using TRISO fuel help drive up costs to eye watering levels.
         There are, however alternative approaches that can significantly reduce TRISO costs.

         • At the end of the day TRISO is manufactured using physical vapor deposition of graphite and silicon carbide using specialized, high temperature, vacuum kilns – similar processes used in semiconductor manufacturing. TRISO will never NOT be an order of magnitude more expensive than pressing oxide in a pill press and firing it in a kiln with H2. Light water reactor fuel costs ~$8/MWh. HALEU TRISO is estimated $240/MWh by the same disingenuous type of people that said they could build a NuScale for $1B. GE passes 100 tons of oxide through a pill press in Wilmington North Carolina every 3 months. Light water reactors burn dirt.

           • Nuclear fuel costs are less than ten percent of production costs for big conventional reactors. Therein lies a way to accommodate higher TRISO fuel costs. Make the plant outputs large while being thermodynamically very efficient. That being said, a number of advanced reactors using TRISO fuel are inherently very small for various reasons, although they are more efficient than conventional water reactors. However, that is offset by much higher enrichment – roughly 15 percent versus around 4-1/2 percent for water reactors.

             • If [my company] is projecting to sell power for $45/MWh this year, then our fuel costs are closer to 20% of the expected revenue.

               As you know, the feed/product relations for enrichment are pretty simple. For 1 ton of 5% we throw away 10 tons of depleted U. For 20% it is 41 tons of depleted U. Most of the “advanced” TRISO guys are targeting 10% enrichment and will throw away 20 tons, and the cost of U3O8 has risen. May be time to re-evaluate the frame of reference of all the variables you’re juggling what offsets what.

               Take the X-energy or PBMR example TRISO fueled reactors – they need the elevated enrichment because they are tall, skinny piles, because that allows for radial heat conduction to the vessel walls in the accident [isolation] case. The enrichment is 10% because the cores are tall and skinny and depart from the textbook “right circular cylinder” optimum – and this is for graphite where the neutrons need to travel half a meter to slow down (< 5cm in light water). NuScale also from the same "small core syndrome" and leaks 10% of it's neutrons (as opposed to 2-3% for a 4GW LWR). So, your "efficient" TRISO reactors are actually very inefficient WRT fuel economy, fuel cost. It's been a while, but I've calculated the volumetric throughput of high-level waste (spent fuel) for the TRISO reactors is up to 25-50x that of LWRs, simply because the spatial density of uranium in the PBMR is 0.05g/cc compared to the 2.3g/cc of LWR. All the fuel that has ever arrived at my company is still on site. TRISO makes sense for niche use where cost is no object (military). If FtStVrain (BISO) had operated well enough to require refueling, all these things would be better appreciated by the academics.

               • The actual fissile/fertile material (heavy metal) needs of the two reactor types are very similar, with the more efficient TRISO fueled reactors creating less heavy metal waste on an electrical generation basis. The TRISO fueled reactors have a significantly lower power densities than conventional reactors, with graphite moderation used instead of water. The volume of the TRISO fueled core (including graphite blocks) is much larger than the conventional reactors. However, the core of the TRISO reactors cannot melt, unlike a conventional core.
                 Not so sure the volume of the core is all that pertinent from a high level waste standpoint. The amounts of radioactive waste are similar

4. If we need the power then the same solution applies that is always applied.

   The solution is to break ground on five gigawatts of new capacity annually. In equilibrium this should put 5 gigawatts on the grid every year. Continue that for 30 years with GE/Westinghouse reactors. And then never stop. Maintain your supply chain. Maintain your talent. It’s simple communist state central planning stuff – it only works when financed by the state.

   • Agree. Just build Westinghouse AP1000 reactors, with natural gas plants picking up the slack until the big reactors show up. In passing, the NRC staff should be required to comply with the law or get drop-kicked over the side of the ship.

5. “The resurgence of nuclear energy is driven by the massive power demands of AI data centers.”

   The ‘resurgence’ is not present tense. It may happen, but what you are referring to is actually a stock bubble. I never thought we’d see one in this field, but the power thirst of [present] AI and futurist hype from Altman and a thousand internet inflencers is inflating a bubble that will pop. hard. And then we will build ap1000s.

   Do I have to remind the crowd that the Great White Hope NuScale design was federally blessed almost 6 years ago and nobody wants a sixpack for $6B, when they can build a gas plant of similar size for < $1B.

   All these initiatives to build 10 micro reactors this year, by 'companies' like Aalo that didn't exist 2 years ago… One does not simply just build these things… and for what? Only the golden hammer procuring military could afford to pay $300+/MWh at the crank. Nothing is new. It could all have been done and hasn't been done just like the nuclear rocket that was canceled a couple months ago with no press conference.

   • If you mean the DRACO project that was sponsored by DARPA and canceled, then this is just another of so many examples of short-sighted budget squeezing to give tax breaks to the rich for nothing, that the Trump administration has been doing, ironically though maybe not coincidentally in this case, spurred on by Elon Musk’s DOGE. More here: https://www.neimagazine.com/news/draco-project-cancelled/. Perhaps Musk sought to undermine a competitor to his failing Starship. Starship will never reach Mars. Perhaps no one but me believes that now, but as the failures mount up, or even minor successes stretch years into the future, with no realistic timeline in either the present administration (or the next one), and aging Musk is increasingly side-lined – easier to do now that Gallup shows he’s the most hated man in America, far more unpopular than even Trump – and China keeps moving towards a crewed lunar landing, reality will become unavoidable.
     America has ceded leadership in so many ways, so fast and so much that most Americans can’t/won’t even see it…until they do. And it was all unnecessary.

     • Yes. Spot on. I might also add that blocking for decades-long projects tens of billions of dollars in investments is quite risky:
       we know that human level intelligence on an organic architecture can run on 300 Watts (not gigawatts, not megawatts, just Watts).

       • No. The DRACO fail is much less sinister than you imagine. It has nothing to do with Musk turning into Rockefeller and undercutting competition or America ceding leadership in technical fields – nuclear energy isn’t TSMC proprietary 3nm lithography, it is heat and mass transfer. In the remote chance some novel approach to heat and mass transfer is demonstrated elsewhere, the US (and RF, Jap, Korea, France, etc.) could either license or mirror (steal) it. I posted a link for a DRACO update here 3 months before the funding was withdrawn; the article was full of sober realizations by the DARPA project manager explaining why the timeline would slip years due to technical challenges. He was foreshadowing the outcome that I predicted: the paper gains in ISP using Nuclear Thermal Propulsion do not compensate for how marginal these systems are in power-to-weight, lifetime, reliability, etc., and just plain PITA. You and many others do not have the perspective to understand how truly impractical dumping cryogenic H2 (a moderator) through white hot tungsten cermet, but the bright minds at DARPA figured it out. Putin has tested and recently threatened to test his “infinite range” NTP powered cruise missile, and a turbojet with a temp as low as < 1500 degC (solid fuel) may indeed function, but it is not a threat, more of a dirty bomb. Imagine a 500lb explosive charge and splattering 100kg of barely burned HEU – more of a terror weapon than a paradigm shift. Not even a good dirty bomb (few fission products).

   • The NUSCALE licensing costs are well over 1/2 billion dollars. Guess who paid most of that cost? Taxpayer.
     The nuclear industry needs to demand that the NRC comply with recent Congressional law directing simplification of nuclear regulations. That being said, who actually picks up most of those licensing costs for advanced reactors? Taxpayer. The nuclear industry has little incentive to grow a backbone, particularly with a hostile NRC staff that can (and does) wrap schedules around the axle if challenged.

     • I believe NuScale received DOE “cost matching” grants, so Fluor and VC had to come up with half of the development funding to get the DOE part, but yes – taxpayer money. In 2011 one of NuScale’s larger investors (Kenwood Group) pleaded guilty to operating a Ponzi scheme; later that year NuScale was bought by Fluor. I agree that the DOE dragged NuScale over the finish line, then spiked the ball, and still lost the game, because the design is a polished turd that reduces fuel economy while increasing the number of licensed staff relative to 60-year-old Gen2 reactors. It doesn’t improve much – perhaps only CDF (core damage frequency per PRA).

       So, tell me: now that NuScale has the original and now uprated designs approved, how the NRC is wrapping things around the axels? They’re not – the design is a LWR so the NRC is fine with it. THERE ARE NO CUSTOMERS. There are negotiations and MOUs and studies for projects in Romania (where they run CANDU well) but THERE ARE NO CUSTOMERS. Because it sucks. Because it is expensive. Because gas plants are highly efficient (45%+) and emit two parts water for one part CO2. If at any time anybody wanted a 600MW station, or a 75MW station, Westinghouse/GE would have built it for said customer. What we have is noise on the internet – a hype bubble – a lot of chatter and bad predictions – as always. There are always new people on the internet that became interested (were born) yesterday.

       • Early grants involved twenty percent cost matching, but do not know where it ultimately ended up. The taxpayer share involves a net of billions of dollar for all the advanced reactors. The net NUSCALE licensing costs are publicly available and well over 1/2 billion dollars.
         By wrapping the effort around the axle, I am referring to costs, including building the facility. There is zero doubt that nuclear plants are vastly more expensive (about a factor of seven) than conventional power plants.
         To be fair, the conventional plants do not require massive amounts of steel and concrete. That being said, the nuclear regulations do, in fact, significantly slow activities and that significantly increases completion times.
         I am not sure you fully appreciate the severe impacts of never-ending “bring-me-another-rock” inquires by the NRC staff and the excruciating paper work associated with the vast overregulation. That directly causes efforts to become snail-like in terms of progress.
         I’ve been in the power industry for more than than half a century and am more than a little familiar with both nuclear and conventional plant efforts. I would never opt for a nuclear plant because the financial risks are immense and essentially unknown. I”ll pick a gas fired plant every time because the financial risk is quite manageable and the combustion turbines are easy to operate.

Comments are closed.