Essay by Eric Worrall
Big tech pushing back on accusations of being AI obsessed climate wreckers.
Microsoft wants us to believe AI will crack practical fusion power, driving future AI
This BS ends at some point, right?
Thomas Claburn
Fri 9 May 2025 // 11:23 UTCMicrosoft believes AI can hasten development of nuclear fusion as a practical energy source, which could in turn accelerate answers to the question of how to power AI.
…
“The pursuit of nuclear fusion as a limitless, clean energy source has long been one of humanity’s most ambitious scientific goals,” three Microsoft Research boffins – Kenji Takeda, Shruti Rajurkar, and Ade Famoti – wrote in a post published Wednesday.
“While scalable fusion energy is still years away, researchers are now exploring how AI can help accelerate fusion research and bring this energy to the grid sooner.”
…
Such speculation is needed because AI is not currently sustainable. As the UN Environment Program observed last year, “The proliferating datacenters that house AI servers produce electronic waste. They are large consumers of water, which is becoming scarce in many places. They rely on critical minerals and rare elements, which are often mined unsustainably. And they use massive amounts of electricity, spurring the emission of planet-warming greenhouse gases.”
…
Enter Clippy: “I see you’re building a fusion reactor. Would you like some help with that?”
Read more: https://www.theregister.com/2025/05/09/microsoft_ai_fusion/
I have to say, I like this new green big tech activist plan to defer serious emissions reduction efforts until affordable nuclear fusion is available. If this is the latest green position, I can get on board with that plan, so long as they develop nuclear fusion on their own dime.
I can’t wait to see what AI comes up with re. a workable way to contain the plasma of a grid-scale fusion reactor.
While at it, AI should also solve the nuclear waste problem.
What problem? Vitrify the stuff and you can leave it out. Don’t need to put it underground.
Reprocess it. Then there is little to nothing left to vitrify.
Can’t be left out because it is still intensely radioactive. But the French vitrify and, according to a French colleague, store it “in a room.” They have been doing this for many decades. Carter killed our vitrification work many years ago.
And maybe figure out how many appendages humans have.
It would be nice if AI could tell us what a woman is.
Grok gives all the “definitions”. It’s pretty sad.
That’s pretty much what the M$ folks are saying in that A.I. “can” help with the plasma conainment problem. Nothing has been said about dealing with the 14MeV neutrons from the DT reactions or about where the tritium needed to start a grid scale reactor will come from.
That would be the giant fission reactor behind the curtain… 🙂
It’s not even (primarily) about containing the plasma. It’s what to do about the fast neutron flux which is why ITER will fail.
Isn’t it about what to do if containment fails and our Solar System then has two Suns?
Maybe it can fabricate a few trillion $ to pay for it as well. Heck, let’s have AI create $37trillion and pay off the national debt too. While it’s at it have it solve the unified field theory. After all there’s nothing AI can’t do.
In my bookmarks “Slashdot” is just before “WUWT”.
By chance today, literally just before reading your post, /. led me to a University of Texas press release titled “University of Texas-led Team Solves a Big Problem for Fusion Energy”, which includes :
OK, so it’s only for “stellarator” designs rather than the usual “tokamak” ones, but after “only” 70 years of active research efforts they have come up with a better algorithm.
All they have to do now is apply it (with or without the help of AI), iron out the “wrinkles”, and then produce a commercially viable 1-3 GW design …
simple.
/s
Magnets are being used to contain the plasma. But what is the containment material that reliably converts the energy and neutron flux (on the order of 10^16 neutrons per square centimeter per second) at very high energy from a streak of plasma at 100,000,000 degrees or so into steam to run the electric generator? Or is there some other magic technology that will convert such intense heat into electricity? So far, the answer seems to be Unobtanium.
Most power plants boil water to turn turbines.
Defanging the anti nuclear energy lobby would be a good first step. They are assured to be opposed to fusion as much as fission, as they are basically Luddite safetyists.
Tom.
You seem to have misspelled Satanists!
It will forever be a decade or more until it is ready.
There ought to be a unit of time measurement named for the number of years until commercial fusion power. I believe the de facto value is ~40 years.
(Some may argue 42 as I suspect that is the more precise answer!)
The magic of this number is that any 23-year old college graduate may safely predict it even fresh out of school, and still be retired before being proven wrong.
We never never going to fly, nor go into space, until we did.
You’ll never ever see commercial fusion power. Bank on it.
Can you run an exorbitant science project to sustain a fusion reaction for a long time? I have no doubt. Just as you can make hydrocarbons out of CO2 and water if you put in more energy than you can get out of the fuel you produce. The only problem is money in minus money out > 0. It costs you more than you get back.
A commercially viable power plant needs to produce power at a cost less than what consumers are prepared to pay. And that is not remotely feasible. Of course governments might be able to distort markets to such an absurd degree through subsidies and bans on all other sources of power, that fusion could be made to appear commercially feasible. Bu only if and it’s a huge if, true breakeven could be exceeded. Otherwise no matter how much you subsidize, you can’t produce power at any price if you need to use more energy than you get out.
Limitless cheap energy is a cruel joke foisted on generations of gullible taxpayers.
I remember as a 23 year saying that fusion may be 40 years away – this was over 45 years ago. The person I said that to replied that the 40 year estimate was essentially saying it will never happen.
Until it suddenly is closer. Interesting that you say a “decade or more”. It’s usually “it’s always 30 years away.”
Max More: “Until it suddenly is closer. Interesting that you say a “decade or more”. It’s usually “it’s always 30 years away.”
My monthly copy of the American Nuclear Society journal Nuclear News always has something in it about the latest announcements & pronouncements in nuclear fusion.
If the opinions I see being expressed among those in my circle are indicative, the older readership of the ANS Nuclear News believes that talk of nuclear fusion is all hype, and nothing but hype.
Right! It’s all “green” to keep what we have and add to it reliably and wait until fusion arrives. Yea. We all agree.
Of course, we’ll all retire beneath the green before fusion, but what the hey, energy on demand will stay!
Do data centers consume large amounts of water? Does the water burn off, evaporate or burn away?
The data centers generate a ton of heat. You have to dissipate the heat or the servers will fail. There’s lots of options to cool them, it need not be water, but you do have to cool them.
Supercomputer processors are refrigerated with liquid nitrogen or other inert gasses. Too much heat to dissipate by regular means.
Some yes, there’s other methods like immersion cooling, but one you get the heat away from the processor you still have to get it from there to the outside world.
There are lots of data centers near where I live. Most seem to have forced draft cooling towers as do many thermal (coal, gas) electricity generators. Whether these cooling towers are to manage computer heat or to manage the thermal cycle of their fossil backup power generators I do not know. Perhaps both.
My point is the water isn’t ‘consumed’ when used for cooling. Most of the water remains after use.
According to this paper ‘The US needs a Bigger (Energy) Boat by D Romito, Pickering Energy Partners (Rosito 20240620.pdf) a typical data centre consumes around 275m gallons of water a day – equivalent to the annual water use requirements for all new oil and gas drilling.
That seems awfully high. I tossed this question to perplexity.ai and it claims water evaporated is 12,300 gallons / day for each megawatt of power used by a data center. This scales to 1.23 million gallons / day for 100 megawatts. If the figures you cite above are accurate, an AI data center consuming 275M gallons/day would consume over 22 gigawatts.
The Perplexity answer noted the evaporation rate is 1% of the circulation volume, so the 275M gallons might be what is circulated in a day with 1%, or 2.75M gallons evaporated. That would suggest a 223 MW data center consumption, which is still quite large.
For context, the 1.15 GWe Plant Vogtle reactors (units 1 & 2) circulate 500,000 gallons per minute and evaporate 15,000 of that (about 3%).
Yes this addresses More Soylent Green’s point about using the word “consume”. They probably recirculate and cool with an evaporator.
It could happen. And we may be running out of nitrogen too! Panic!
What could possibly go wrong?
Westworld: Where nothing can go worng…
Fusion has been vaporware since 1948. It does not exist and AI can do nothing not in a data base.
My former chairman, Lyman Spitzer, in the astrophysical science department at Princeton headed the program and developed the Stellarator which is now coming back into prominence. It is well known how fusion works. We cannot duplicate those conditions until we can control gravity.
In the meantime, nuclear FISSION is proven to be the safest power source – and it does exist.
There is sufficient uranium (and thorium) to power the entire planet for a millennium, safely.
Let’s build the needed 5000 USC co-producing breeder fission plants and have energy for the planet, giving us the TIME to crack the fusion nut. The nuclear power plant has an operational life of 60-75 years, even the early reactors are mostly operating still unless one of our brilliant elite has blown them up.
Dealing in futures is what our elite politicians do, like Macron, Starmer, and Merz were just caught doing in a train car. We need a stable, reliable, and affordable power source,.
Exactly.
If existing technology fission generators were installed widely, the development of fusion generators could fade into a backwater academic exercise. Minimally funded by taxpayer r&d grants? – maybe not.
“power the entire planet for a millennium”
All kinds of assumptions in that number.
Probably it was made before electric car mandates and AI were more than sci-fi.
But I thought that “giving cheap, abundant energy to humanity would be like giving an idiot child a machine gun”.
Or –
“giving election majorities to globalists is like giving a jihadist a pick-up truck”
“Microsoft believes AI can hasten development of nuclear fusion as a practical energy source, which could in turn accelerate answers to the question of how to power AI.”
Fission is already practical and emissions free, so why not just use that?
“The proliferating datacenters that house AI servers produce electronic waste.
Just like the electric cars, windmills and solar panels you want us to switch to?
They are large consumers of water, which is becoming scarce in many places.
That’s why we don’t build them in places where water is scarce!
They rely on critical minerals and rare elements, which are often mined unsustainably.
Just like the electric cars, windmills and solar panels you want us to switch to?
Is there an echo in here?
And they use massive amounts of electricity, spurring the emission of planet-warming greenhouse gases.”
Fission is already practical and emissions free, so why not just use that?
(Again, is there an echo in here?)
There is definitely an echo; I believe it comes from your words rattling around in the nearly empty crania of the Climate Apocalypse Hoax true believers! At least it is all relatively well synchronized by their “thought” installers; otherwise it would soon devolve to white noise!
I was in the nuclear power business from 1966 to 1996 and have watched many “brilliant” ideas come and go — mainly go — during that time. Fusion may become commercial in the distant future — never say never — but it will long be a dream and may join the ash heap of history with many other dreams. In the meantime it provides jobs for some really smart people and suppliers who are happy to accept the challenge and the money that comes with it.
An awful lot of scientific research is simply welfare for highly educated people.
aka “welfare queens in white coasts”
aka white collar welfare
Perhaps the matter – antimatter reactors will come online sooner?
/h
The three big questions for fusion reactors have remained the same for decades:
1) How do we get the heat out in a usable form?
2) How do we get the waste product (He) out without shutting down the fusion process?
3) How do we get new fuel (hydrogen isotopes) in without shutting down the fusion process?
I’d like to see anyone, or anything (AI), answer those questions.
Tokamak style fusion reactors are a dead-end becasue they don’t provide a path that can answer those three questions.
“I just started fusion in a tokamak reactor, Hurray! Now what do I do?”
Run away…. run away….
I don’t trust AI and I have no faith in it. Let them build what they want so long as it is on their own dime, so long as they build nuclear power plants for their energy to insure the rest of us have plenty of affordable, reliable power.
“While scalable fusion energy is still years away”…any words make sense in the second half of the sentence.
I am an outlier when it comes to AI. I have been in the computer world since 1979. And my thoughts about AI are straightforward. Garbage In Garbage Out, amplified 100 fold.
I started in 1988, and I’m of the same mind.
79 for me, and I agree too. From who I know, those of us who have been around a while seem to have a similar opinion.
I’m tired of the AI hype. It’s still subject to GIGO and biases.
My monthly copy of the American Nuclear Society journal Nuclear News always has articles which include some reference to ‘solving the climate crisis’, and to ‘deep decarbonization’, and to ‘a tripling of nuclear capacity by 2050.’
And lately, in the last year or so, to the emergence of AI-driven demand for more electricity and to the role of AI for use in various nuclear industry applications.
The emergence of AI is a magic elixer which, according to those who promote it within the nuclear industry, can be applied in a variety of ways to make nuclear power more cost competitive with other means of generating electricity.
Thirty years ago, the emergence of Expert Systems was being promoted as a means of doing exactly the same thing. Over the next two decades, that concept quietly disappeared off the radar screen.
Back to the future.
When the cost and schedule feasibility estimating for Nuclear Renaissance 1.0 was being done in the mid-2000’s, it was assumed that all the extensive lessons learned in the 1980’s about how to keep nuclear projects on cost and on schedule would be diligently applied to the newest reactor construction projects.
Those lessons learned from the 1980’s are extensively documented in the historical record. And so an AI application could, in theory, sink its teeth into that extensive documentation and spit forth the kind of sound advice that humans were offering to nuclear project managers in the late 2000’s when VC Summer and Vogtle 3 & 4 started construction.
What actually happened instead was that the original project management teams for VC Summer and for Vogtle 3 & 4 ignored every piece of sound advice that was offered them and then made every serious mistake it is possible to make in managing a nuclear construction project. VC Summer was canceled and the original Vogtle management team was shown the door.
So here is a question:
Would the original project managers for VC Summer and for Vogtle 3 & 4 have heeded all the sound advice that was given them had that advice been generated by an AI application — as opposed to being produced by humans who had accumulated years of practical experience in how to deliver nuclear projects on cost and on schedule?
Henry Petrowski made a comment about world class bridge disasters occurring at approximately 30 year intervals. This was the amount of time for the large majority of engineers who were around when the last disaster occurred to have retired and replaced by younger engineers.
Institutional knowledge matters.
Same thing happens in software maintenance. New kid on the block comes in and see something that isn’t implemented the way he was taught in school he changes it. If a complex product has been in the field very long many problems repeat themselves every couple of releases.
The real question is how good is AI at civil engineering? It was having to construct a big building that was the source of all the problems at VC Summer, Vogtle, and recent experiences in Europe at constructing new nuclear plants. (Note that the Chinese don’t have this problem, because they treat their buildings as “Made in China.”)
Realistically, the best that AI has to offer the nuclear industry is an efficient way to go through all of the paperwork … and there is a lot of paperwork in an industry as heavily regulated as the nuclear industry. So, this is essentially a machine solution to a human-made problem. If you look at it that way, there is some hope.
AI will not “discover” the solutions to the inherent problems with nuclear fusion. The “smarter” our phones become and the more “artificial” our intelligence becomes, the dumber the people become, and they are even more likely to believe in magical solutions.
+10
Brian, it was the gross mismanagement of every facet of those two projects from the start of construction that was the root cause of the bulk of the huge cost and schedule overruns at VC Summer and at Vogtle 3 & 4.
What was seen at VC Summer and at Vogtle 3 & 4 was an exact repetition of the management issues seen at nuclear projects in the late 1970’s and early 1980’s. For those like myself who had been a direct witness to those issues in the late 1970’s and early 1980’s, it was deja vu all over again.
The original management teams at VC Summer and at Vogtle 3 & 4 knew what was being expected of them them for every facet of their two projects, but they failed to deliver on their promises and obligations.
VC Summer was canceled outright and the original management team at Vogtle 3 & 4 was replaced. Vogtle 3 & 4 still experienced cost growth issues after the original management team was sent out the door, but it was a consequence of growing worldwide competition for nuclear-grade components, equipment, and services.
Vogtle 3 & 4 went forward after the management shake-up because a public policy decision had been made by Georgia’s politicians that the project would continue regardless of what it would eventually cost, and regardless of how long it would eventually take.
Going with nuclear is a public policy decision. Buying nuclear is done for purposes of gaining energy reliability and security, not because it is the cheapest way to generate electricity.
And so if we want the benefits of nuclear’s reliable energy and security capabilities, we are obliged to pay a premium for nuclear over and above what we would pay for equivalent gas-fired generation.
What we as energy consumers should expect from the nuclear construction industry, and what public policy decision makers should expect in promoting and approving nuclear projects, is that we pay the lowest premium for nuclear that is reasonably possible within the regulatory and industrial base environment we now face.
“The real question is how good is AI at …”
Don’t forget, AI is usually trained on peer review papers and/or data on the internet. Since 80% to 85% of that information is incorrect (i.e. garbage), until someone has figured out how to program a strong B.S. Meter AI can’t be more than 10 to 15% accurate.
I don’t know about those boys, but I see some pretty absurd answers from “AI”.
It’s like asking a stoner a philosophical question. These AI machines often go off and “hallucinate.”
See my comment above :<)
What is “electronic waste”?
Burnt out integrated circuits.
Nuclear fusion is an engineering matter, as indeed are all practical applications of science.
A better understanding of fusion won’t build the machines needed.
A plasma hotter than the sun half a meter from magnets cooled to nearly absolute zero, is an engineering challenge for a long ways down the road….it took humanity a long time from controlling fire to simple batteries…
+10
so long as they develop nuclear fusion on their own dime.
No matter who foots the up front cost, everyone pays one way or another. Maybe it is a fee to use the AI. Maybe it is something else.
D-T fusion is not green. It produces high energy neutrons.
Boron fusion is potentially green, but is much more difficult than D-T.
Magnetic confinement for D-T fusion is not going to work because 80% of the energy is carried away as neutrons which cannot be contained magnetically. Every time the tokomat fusion starts it will snuf itself out unless you add energy. While irradiating everything around it
Using what, LLM’s? Don’t make me laugh. They’ll have us looking for eezo (Mass Effect) to fuel the damn things.
And always will be. Give it up. The Sun gets to generate net positive energy because it gets gravity for free. Us mere mortals have to pay for a gravity equivalent with energy, and even if we could somehow nudge that over into a small net energy surplus, it’s never going to be fiscally profitable, and the ERoEI will probably never even break even in a thirty-year lifespan, never mind give a positive return.