Remember when they told us in CS class that it's better to design more efficient algorithms than to buy a faster CPU? Well here we are building nuclear reactors to run our brute force "scaled" LLMs. Really, really dumb.
You have a point. But it doesn't make sense to seek for the next unrealized breakthrough (low energy, brain-comparable power consumption) AI leap yet, when existing products are already so transformative.
The one company I really want to see involved in dangerous things with clean up and serious environmental risks is the company that has serious production QA problems, an attention span of about 2 minutes, regular bouts of corporate schizophrenia and a policy of forcing half the planet to abandon working hardware.
Nothing good can come of this.
Microsoft needs to start asking if it should do something before it does it.
Have we seen Microsoft actually put any skin in the game yet? All the pre-purchase announcements are virtually risk free for Microsoft. They've agreed to buy a certain amount of power at a certain price, if the counter-party can deliver it. But they're not pre-paying, they only pay when the electricity gets delivered. If they never deliver, Microsoft isn't out any money.
Ugh, I rdread this topic because nuclear is as close to as a religion as you get on HN. SMRs just arne't better in any way that matters [1].
And while I personally hope we have economical commercial power generation in the future, I'm not convinced that'll ever happen due to one massive problem: energy loss from high-energy neutrons, which have the added problem that they destroy your very expensive containment vessel. Stars deal with this by being massive, having fusion happen in the core (depending on the size of the star) and gravity, none of which is applicable to a fusion reactor.
I'm reminded of the push recycling of plastic. Evidence has surface that this was nothing more than oil industry propaganda to sell more plastic [2]. A lot of "recycling" is simply dumping the problem into developing countries and then just looking the other way. We used to do this to China until they stopped taking plastic to "recycle".
I can't help but think that Microsoft issuing some press releases about nuclear is nothing more than marketing to contributing to the data center explosion that will inevitably drive up your electricity bills because you'll pay for the infrastructure that needs to be built and will be paying the generous (and usually secret) subsidies these data centers engotiate.
dating myself here but I remember in the 90s reading a really funny spoof article about Microsoft announcing they had developed nuclear weapons. Didn't even seem that implausible at the time.
I would have linked it here but none of the search engines are turning up anything at all, and in fact I don't think it's even possible to find stuff like that with search engines anyore.
The Windows 98 license actually did forbid using Windows in nuclear power plants (along with other high risk areas). That was due to some interaction with the Java license and I always considered it a very fortunate fluke.
Awesome. I'm convinced nuclear is the only realistic path toward an energy-laden sustainable future, I've yet to understand the fear mongering beyond political faction bearing and token counting in terms of district employment numbers or some such third-order nonsense... there's nothing safer in terms of human lethality.
Molten salt reactors, micro-reactors, modularity. It's the miltech we had in the 60s, on the path to commercialization and commoditization.
It's all proven technology and the obvious exemplar is the nuclear-powered navies, micro-cities that can roam, submerged within the depths of, or riding atop the world's oceans, for decades at a time. We've been doing this for over 70 years.
It's only a matter of time. AWS has a campus in PA already next to the power plant at Susquehanna, plugged in. They're invested in small modular reactors.
Google has contracts and investments toward the same end. This fits the pattern we're seeing across big tech, and it's driven by the non-negotiable power demands of AI.
I don't balk at the climate-changists, I'm more curious about the anti-Nuke sentiments on HN; what am I missing?
This is a big deal, not because Microsoft wants to build reactors but because it highlights the real bottleneck: nuclear fuel. There’s already a growing uranium deficit, conversion and enrichment capacity are thin and geopolitically fragile, and next-gen reactors need HALEU, which barely exists today. Building new reactors is the easy part — scaling the fuel supply chain takes years.
The article mentions Helion. Those guys were supposed to have their Polaris demo machine running by now. But they've become very quiet about that. Press releases about it in 2024, but not much in 2025.
Polaris is supposed to pass theoretical breakeven, and maybe even technical breakeven - more electrical power out than they put in. That would be a huge event.
These are very long term bets. MS isn't betting much here, just staying involved just in case. Which is prudent but not much of a commitment. A big commitment for a trillion dollar company would have a big dollar value. Like billions of dollars. That's not what's happening here.
I like the idea of small reactors from a technical point of view. But I'm also a realist. To match current renewables growth (or even put a minor dent in it), many tens of thousands of these things are needed. They don't put out a lot of energy. In wind number of turbines it's something like 2-5 turbines per reactor. There already are tens of thousands of wind turbines. Plonking down a few hundred wind turbines is routine business. Getting the first small reactor online is still in progress.
In other words, small reactors are not happening anytime soon. Certainly not in the next decade. If there are a few hundred active small reactors in 15 years that would be really amazing. And if that happens at a reasonable cost (big if) relative to wind, solar, and batteries, that would be even better. But we'll be well into the second half of this century before these things are putting a dent into other sources of energy. And that's only if it all works out in terms of cost and technology. 25 years is not that long in nuclear. Long planning cycles are common. These things have a lot to prove.
I'm skeptical on especially the cost aspect. Nuclear proponents tend to gloss over the fact that nuclear has always been expensive. Things like waste handling and security add extra cost and small reactors just complicate that further. Small reactors have a lot to prove and the rosy projections tend to dodge the harder issues here. There's a lot of magical thinking around this topic.
In any case, a few hundred of these things would be a meaninglessly small drop in the ocean in terms of energy output. It's not coming even close to the yearly growth with solar, wind, and batteries. And MS needs data centers sooner than even those would be coming online. And the energy to power them. Wherever that's going to come from, it's (mostly) not going to be nuclear any time soon. Unless they drastically scale down their AI expansion plans. And long term this is a cost game. MS is going to need lots of cheap energy. Expensive energy just raises its cost. Unless small reactors fix the cost issue, MS won't be using a lot of small reactors.
I see a lot of skepticism in the comments, but if you’re going to gamble, SMRs seem like a pretty good bet. Nuclear is still in its mainframe era, where everything is bespoke and costly. Modularization enables repeatability, which is the heart of optimization. Doing something smaller, but more often, is how you get good at most things!
There’s a hundred and one “yes, but” objections to make, but our energy transition needs to throw everything at the wall and see what sticks. I don’t think it’s a choice between nuclear and other renewables. We need them all.
We recently had an election in AU where "nuclear" was on the agenda as the (losing) party/coalition were promoting nuclear as a "solution" to our aging coal-generator fleet.
The trouble is that:
a) "baseload" is a misnomer, what is required is storage to cover periods when "the sun doesn't shine and the wind doesn't blow"
b) CSIRO (our government research organization) releases a regular report called "Gencost" [1]. It has shown regular decreases in solar and wind, with costs for other solutions (coal/gas/nuclear) growing during the same period
c) The problem for nuclear power in AU is doubled because there is no local infrastructure or engineering or industry for the nuclear fuel cycle
d) AU home solar is world leading, with now a government subsidy available for home battery storage to soak up the midday peak, one state (SA) regularly runs on 100% renewables
e) SMRs do NOT exist in a commercially deployable way. There are any number of research and demo-scale possible SMRs, but none that are immediately able to be deployed
f) SMRs are too SMALL to replace existing coal gen, especially compared to the capacity of solar and wind farms, with offshore wind only just being started in AU
> c) The problem for nuclear power in AU is doubled because there is no local infrastructure or engineering or industry for the nuclear fuel cycle
One might say this is an advantage, with no home industry asking for local supply chains to be built up (at significant cost and risk). Solar panels, batteries and wind turbines are not generally made in Australia, right? For the fuel cycle, Urenco for enrichment, and Westinghouse or Orano for the uranium processing and fuel fabrication would be possible deals with allies.
> e) SMRs do NOT exist in a commercially deployable way. There are any number of research and demo-scale possible SMRs, but none that are immediately able to be deployed
SMRs are not the entirety of nuclear, and were not the entirety of the Coalition nuclear plan. Large reactors do exist and are being built around the world. Rosatom are able to do so (Egypt, Turkey, Bangladesh), KEPCO has done so in the UAE, and China is exporting to Pakistan. As an aside, a GE-Hitachi BWRX-300 first-of-a-kind (not demo-scale or research) is being built in Darlington Canada, so SMRs are being deployed.
> f) SMRs are too SMALL to replace existing coal gen, especially compared to the capacity of solar and wind farms, with offshore wind only just being started in AU
I'm baffled that anyone could propose this as a good thing. I've worked in industries that are supported - just smaller than the US and the price gouging is substantial. Parts will be designed for the US market, we need to adapt. Bonus points when buying incompatible European / US designs.
Solar panels, batteries, and wind turbine have all the necessary ancillary parts in warehouses close to where they are needed. They also have all the expertise, the cranes, the transport regulations nailed down.
Oh, who would have thought. Fusion and nuclear is a money pit.
Because complexity is expensive and those two are by far the most complex ways of generating energy, one of which is even so complex it hasn't even achieved net plus anyway.
Even a giant like Microsoft doesn't have unlimited funds to burn.
I am user of Microsoft Nuclear Cloud Plant Manager Xp SP3 and after update my control panel restarted causing power outage.
Now every time i click power status it is crashing, while power output is rising, making pipes hot.
I already tried msreactor /scannow but don’t want to reinstall reactor as last time I did it I lost my city (and support only told to use boron or move to another area).
37 comments
[ 4.0 ms ] story [ 57.2 ms ] threadIt will come, give it time.
Nothing good can come of this.
Microsoft needs to start asking if it should do something before it does it.
And while I personally hope we have economical commercial power generation in the future, I'm not convinced that'll ever happen due to one massive problem: energy loss from high-energy neutrons, which have the added problem that they destroy your very expensive containment vessel. Stars deal with this by being massive, having fusion happen in the core (depending on the size of the star) and gravity, none of which is applicable to a fusion reactor.
I'm reminded of the push recycling of plastic. Evidence has surface that this was nothing more than oil industry propaganda to sell more plastic [2]. A lot of "recycling" is simply dumping the problem into developing countries and then just looking the other way. We used to do this to China until they stopped taking plastic to "recycle".
I can't help but think that Microsoft issuing some press releases about nuclear is nothing more than marketing to contributing to the data center explosion that will inevitably drive up your electricity bills because you'll pay for the infrastructure that needs to be built and will be paying the generous (and usually secret) subsidies these data centers engotiate.
[1]: https://blog.ucs.org/edwin-lyman/five-things-the-nuclear-bro...
[2]: https://www.npr.org/2020/09/11/897692090/how-big-oil-misled-...
I would have linked it here but none of the search engines are turning up anything at all, and in fact I don't think it's even possible to find stuff like that with search engines anyore.
Molten salt reactors, micro-reactors, modularity. It's the miltech we had in the 60s, on the path to commercialization and commoditization.
It's all proven technology and the obvious exemplar is the nuclear-powered navies, micro-cities that can roam, submerged within the depths of, or riding atop the world's oceans, for decades at a time. We've been doing this for over 70 years.
It's only a matter of time. AWS has a campus in PA already next to the power plant at Susquehanna, plugged in. They're invested in small modular reactors.
Google has contracts and investments toward the same end. This fits the pattern we're seeing across big tech, and it's driven by the non-negotiable power demands of AI.
I don't balk at the climate-changists, I'm more curious about the anti-Nuke sentiments on HN; what am I missing?
'Going Nuclear - How the Atom will save the world' by Tom Gregory https://open.spotify.com/show/7l8wXWfPb3ZAUmd1pfUdv3?si=52fe...
Earlier this summer it was quantum computing, more recently optical computing, seems like the next one is going to be fusion!!
Polaris is supposed to pass theoretical breakeven, and maybe even technical breakeven - more electrical power out than they put in. That would be a huge event.
I like the idea of small reactors from a technical point of view. But I'm also a realist. To match current renewables growth (or even put a minor dent in it), many tens of thousands of these things are needed. They don't put out a lot of energy. In wind number of turbines it's something like 2-5 turbines per reactor. There already are tens of thousands of wind turbines. Plonking down a few hundred wind turbines is routine business. Getting the first small reactor online is still in progress.
In other words, small reactors are not happening anytime soon. Certainly not in the next decade. If there are a few hundred active small reactors in 15 years that would be really amazing. And if that happens at a reasonable cost (big if) relative to wind, solar, and batteries, that would be even better. But we'll be well into the second half of this century before these things are putting a dent into other sources of energy. And that's only if it all works out in terms of cost and technology. 25 years is not that long in nuclear. Long planning cycles are common. These things have a lot to prove.
I'm skeptical on especially the cost aspect. Nuclear proponents tend to gloss over the fact that nuclear has always been expensive. Things like waste handling and security add extra cost and small reactors just complicate that further. Small reactors have a lot to prove and the rosy projections tend to dodge the harder issues here. There's a lot of magical thinking around this topic.
In any case, a few hundred of these things would be a meaninglessly small drop in the ocean in terms of energy output. It's not coming even close to the yearly growth with solar, wind, and batteries. And MS needs data centers sooner than even those would be coming online. And the energy to power them. Wherever that's going to come from, it's (mostly) not going to be nuclear any time soon. Unless they drastically scale down their AI expansion plans. And long term this is a cost game. MS is going to need lots of cheap energy. Expensive energy just raises its cost. Unless small reactors fix the cost issue, MS won't be using a lot of small reactors.
There’s a hundred and one “yes, but” objections to make, but our energy transition needs to throw everything at the wall and see what sticks. I don’t think it’s a choice between nuclear and other renewables. We need them all.
The trouble is that:
a) "baseload" is a misnomer, what is required is storage to cover periods when "the sun doesn't shine and the wind doesn't blow"
b) CSIRO (our government research organization) releases a regular report called "Gencost" [1]. It has shown regular decreases in solar and wind, with costs for other solutions (coal/gas/nuclear) growing during the same period
c) The problem for nuclear power in AU is doubled because there is no local infrastructure or engineering or industry for the nuclear fuel cycle
d) AU home solar is world leading, with now a government subsidy available for home battery storage to soak up the midday peak, one state (SA) regularly runs on 100% renewables
e) SMRs do NOT exist in a commercially deployable way. There are any number of research and demo-scale possible SMRs, but none that are immediately able to be deployed
f) SMRs are too SMALL to replace existing coal gen, especially compared to the capacity of solar and wind farms, with offshore wind only just being started in AU
[1] https://www.csiro.au/en/research/technology-space/energy/ele...
One might say this is an advantage, with no home industry asking for local supply chains to be built up (at significant cost and risk). Solar panels, batteries and wind turbines are not generally made in Australia, right? For the fuel cycle, Urenco for enrichment, and Westinghouse or Orano for the uranium processing and fuel fabrication would be possible deals with allies.
> e) SMRs do NOT exist in a commercially deployable way. There are any number of research and demo-scale possible SMRs, but none that are immediately able to be deployed
SMRs are not the entirety of nuclear, and were not the entirety of the Coalition nuclear plan. Large reactors do exist and are being built around the world. Rosatom are able to do so (Egypt, Turkey, Bangladesh), KEPCO has done so in the UAE, and China is exporting to Pakistan. As an aside, a GE-Hitachi BWRX-300 first-of-a-kind (not demo-scale or research) is being built in Darlington Canada, so SMRs are being deployed.
> f) SMRs are too SMALL to replace existing coal gen, especially compared to the capacity of solar and wind farms, with offshore wind only just being started in AU
This is why the Coalition plan proposed large reactors in addition to SMRs. https://www.theguardian.com/australia-news/2025/feb/04/nucle...
I'm baffled that anyone could propose this as a good thing. I've worked in industries that are supported - just smaller than the US and the price gouging is substantial. Parts will be designed for the US market, we need to adapt. Bonus points when buying incompatible European / US designs.
Solar panels, batteries, and wind turbine have all the necessary ancillary parts in warehouses close to where they are needed. They also have all the expertise, the cranes, the transport regulations nailed down.
Because complexity is expensive and those two are by far the most complex ways of generating energy, one of which is even so complex it hasn't even achieved net plus anyway.
Even a giant like Microsoft doesn't have unlimited funds to burn.
I already tried msreactor /scannow but don’t want to reinstall reactor as last time I did it I lost my city (and support only told to use boron or move to another area).
Please help!