News US govt wants to talk to tech companies about AI electricity demands — eyes nuclear fusion and fission

[COLGeek]

I think that at least one "dead horse" has been beaten to a pulp. Points made, now lets move on. Shall we?

 

[enewmen]

There will not be a choice. Fortunately modern fission reactors are lightyears ahead of where they were when most of the ones we have running now were built.
We are going to need a lot of them, not just a handful, dozens and dozens.
The fusion guys say we are ten years away, but they have been saying that for at least five decades, it might not even happen for another five, and as far as practical usable fusion reactors on the grid, maybe twice that. I seem to recall we are three to five trillion dollars away, and no one has the political will to spend that much.
Wind and solar are a very small fraction of the grid, and in 20 years they will still be, they won't save us.
Fission it likely will be. Better hurry though, our needs are going to double if not triple in the next couple of decades.

It looks like you're talking about Thorium reactors. Lots of small local-use thorium reactors will solve many problems. These reactors are MUCH cheaper, safer, don't require water, and smaller than the big uranium reactors seen today.

 

[enewmen]

Here in Indiana they actually passed a bill allowing old power plants to be retrofitted with micro nuclear reactors. IE replacing Coal/Oil/Natural Gas heat sources with a nuclear one, more or less recycling the turbines and generators.

Rather progressive for a formerly no-nuke state. I don't think anyone has actually taken up the task to do one, but it is currently allowable.

Been a long time since anyone has actually built a molten salt reactor, and they didn't work super well when they were tested by the UK and US.

They have better tools now, though. Computers and the ability to simulate every practical aspect before actually trying to build one. I suspect China will probably beat everyone to it, they have built up an enormous stockpile of Thorium from their rare earth metal mining.

Molten salt thorium reactors are being built and used NOW in China for use in remote areas and large container ships. Search:

TMSR-LF1, KUN-24AP​

 

[watzupken]

As I was expecting, this AI craze is going to hit a brick wall soon. It is not just about the availability of the hardware, but the fact that you have all these power guzzling hardware that is growing exponentially, with limited power generation and ageing infrastructure to support. It is not sustainable. All these sustainable electric source are both unreliable and will not generate sufficient power to feed such power requirements.

What I feel is annoying is that they are shutting down nuclear power plants for whatever reasons, which could otherwise increase supply and make electricity more affordable for households. And now for the sake of AI, they are looking at nuclear power again. So the government prioritizes AI over real people? This is a serious problem.

 

[kjfatl]

Modern nuclear reactors are much safer than previous generations. Fukushima was a mix of generation 2 and 2+ boiling water reactors. Generation 2 reactors were the first generation of civilian power designs built in large number in the 60’s and 70’s. Generation 1 reactors were direct copies of the first pressurized water naval reactor designs scaled up in size and output.

Unfortunately, reactors stopped being built in the U.S. starting around the 3 mile Island incident which was blown completely out of proportion at the time and continues to be dramatized by people who know nothing about the incident (aka Netflix). Chernobyl was a communist f’up at its finest. They saw nothing wrong with using a military plutonium breeding reactor design for civilian energy production. RBMK reactors are the most unsafe reactor designs ever invented, whether it be the positive void coefficient due to the decision to use graphite to moderate the neutrons (IE: the more coolant that turns into steam within the reactor vessel, the more potent the nuclear reaction becomes. This quickly runs away in a positive feedback loop), or the graphite tipped control rods (graphite moderates neutron energy allowing more effective absorption of neutrons into uranium which causes the atom to split, so the control rods actually increased the fission reaction further when the rod’s tips first lower into the reactor.).

American reactor designs were engineered to exhibit negative void coefficient due to using water to both cool and moderate the neutrons (IE the more coolant turns to steam, the less potent the nuclear reaction becomes. This can automatically shut down the reaction in a negative feedback loop if too much steam in the reactor vessel develops.)

Unfortunately, national protests after 3 mile and Chernobyl means that the newest reactors the USA has is generation 2. However, we continued to R&D newer designs and if we started building nuclear reactors here again, we would be building generation 4+ designs which are loaded with passive safety mechanisms and failsafes. (For example, the LiFThR [Lithium Fluoride Thorium Reactor] is a completely liquid design where molten lithium fluoride salt is both the coolant and the carrier fluid of diffuse uranium fluoride salt. [This is much safer than having hundreds of kilograms of uranium densely packed in the core ready to meltdown if conditions aren’t perfect. Also molten salt replacing water as coolant means there is no longer a way for hydrogen gas to be created due to neutrons destroying the bond between water’s hydrogen and oxygen, which means no Fukushima containment building explosions.] Only when the molten uranium fluoride salt is traveling through the graphite core are neutrons moderated to allow fission. The LiFThR design also has a failsafe where if the power to operate active safety equipment is lost, the freezer unit at the bottom of the molten salt loop can no longer keep the salt plug frozen, is quickly melted by the residual heat of fission, and opens allowing the entire salt coolant and uranium salt to drain out of the reactor into a tank purposefully designed to be able to passively cool the reaction mixture from decay heat indefinitely.

Vogtle units 3 and 4 are Generation III+ reactors. Unfortunately the project to build them was badly mishandled. They were supposed to be the first of 100 or more similar reactors.

 

[williamcll]

Did you mean Japan (Singular)? No other country is close enough to feel the effects. And that is accounting for the heavy water they have to purge once in a while.

No, Korea is having the worse of it and both Chinas are looking at inland/foreign fish to source. The heavy water needs more further decontamination before dumping it which they are doing right now.

I do agree that they should be fine with newer reactors though. But I think it's cheaper and less riskier to just buy out the thousands of spa resorts across the country and convert them into geothermal generators.

 

[The_Future_Is_Here]

Here's a novel concept,

How about instead of building more power stations we all focus on REDUCING our energy usage, with a HUGE focus on reducing the amount of energy that Data Centers & Computers use.

Data Centers are quickly becoming the most energy hungry industry, and I don't think that we should allow that to happen.
We should be investing in reducing our energy usage, not offsetting it, and not increasing our energy production through the power grid.

A genuine and United effort needs to be made to massively cut back on the amount of energy we are using in every part of our daily lives, with of course a heavy focus on the largest energy users.

Develop more efficient ways of manufacturing stuff, focus silicon Electronics Designs on the reduction of power & the reduction of wasted energy, make things more efficient in how they utilize the energy that we already have.

This drive to just build yet more power stations right across the world is stupid and very short sighted, we need to be focusing on condensing our energy usage, and in the long term reducing our overall global energy usage.
Demands will continue to increase only if we allow industries to expand in an unregulated wild west type manner.

Put in place some guidance, regulations, and barriers to prevent any rogue players from taking advantage of blatantly obvious loopholes, and force everyone in every industry & every place of work, in every job position, to think hard about what waste they are producing and what energy they are needlessly wasting.

In my lifetime the average home computer CPU has gone from consuming roughly 25W at peak to consuming north of 400W, and the efficiency curve for that near 400W CPU is horrendous, it shouldn't even be allowed to ship as a product.

We need tighter regulations of all things electronic & the power that they are allowed to draw in daily operation and in peak operation, hard limits of say 50W as a initial target and then slowly lower that peak power draw figure over time.
That will still drive innovation, it's just that the focus will be getting the most performance per-watt of energy from your Processor but not just restricted to the hardware, the software & firmware will also need to go through some big changes to truly make a substantial and meaningful difference to the energy that these components are using.

I'm not denying that we do right now need more energy because of our potential future power needs, but that doesn't mean we have to focus on yet more energy production. Other options exist and we will benefit more in the long run if we were to focus on those alternative options globally.


I await replies from others to see what their opinions are on this take.

 

[The Historical Fidelity]

Vogtle units 3 and 4 are Generation III+ reactors. Unfortunately the project to build them was badly mishandled. They were supposed to be the first of 100 or more similar reactors.

Yeah, not badly mishandled per se, more like we haven’t built a commercial reactor in ~40 years so we will be relearning a lot of construction details lost to the prior generation of reactor construction crews. You know what they say, “if you don’t use it, you lose it” and it applies to nuclear construction as well.

 

[MatheusNRei]

Here's a novel concept,

How about instead of building more power stations we all focus on REDUCING our energy usage, with a HUGE focus on reducing the amount of energy that Data Centers & Computers use.

Data Centers are quickly becoming the most energy hungry industry, and I don't think that we should allow that to happen.
We should be investing in reducing our energy usage, not offsetting it, and not increasing our energy production through the power grid.

A genuine and United effort needs to be made to massively cut back on the amount of energy we are using in every part of our daily lives, with of course a heavy focus on the largest energy users.

Develop more efficient ways of manufacturing stuff, focus silicon Electronics Designs on the reduction of power & the reduction of wasted energy, make things more efficient in how they utilize the energy that we already have.

This drive to just build yet more power stations right across the world is stupid and very short sighted, we need to be focusing on condensing our energy usage, and in the long term reducing our overall global energy usage.
Demands will continue to increase only if we allow industries to expand in an unregulated wild west type manner.

Put in place some guidance, regulations, and barriers to prevent any rogue players from taking advantage of blatantly obvious loopholes, and force everyone in every industry & every place of work, in every job position, to think hard about what waste they are producing and what energy they are needlessly wasting.

In my lifetime the average home computer CPU has gone from consuming roughly 25W at peak to consuming north of 400W, and the efficiency curve for that near 400W CPU is horrendous, it shouldn't even be allowed to ship as a product.

We need tighter regulations of all things electronic & the power that they are allowed to draw in daily operation and in peak operation, hard limits of say 50W as a initial target and then slowly lower that peak power draw figure over time.
That will still drive innovation, it's just that the focus will be getting the most performance per-watt of energy from your Processor but not just restricted to the hardware, the software & firmware will also need to go through some big changes to truly make a substantial and meaningful difference to the energy that these components are using.

I'm not denying that we do right now need more energy because of our potential future power needs, but that doesn't mean we have to focus on yet more energy production. Other options exist and we will benefit more in the long run if we were to focus on those alternative options globally.


I await replies from others to see what their opinions are on this take.

Without going into the merits of such a proposal, I see a couple of problems with it's implementation:

First, the economic impact of the proposal on the short term would probably be quite severe;

Second, you would need to ensure global adoption if you intend to do more than just shift the burden elsewhere and getting the whole world to agree on anything is next to impossible;

Third, putting a cap on power draw would also mean putting a ceiling on performance because there's only so much power efficiency you can get before physics starts to get in the way.

 

[Diogene7]

It is really crazy that billions of dollars are allocated to current AI technology that rely on hugely energy inefficient silicon transistor technology.

I would think the problem should be tackled differently, in order to accelerate the transition to much more energy efficient beyond CMOS technology : spintronics related technologies (MRAM, Intel MESO concept,…).

As a 1st step, I would use current AI datacenter to focus on finding options for what could be some of the best materials and architecture to create Non-Volatile-Memory (NVM) MRAM with good key performance indicators (ex: as possible, less than 1ns latency, retention > 20 years at 125C, R/W endurance > 10E18,…).

Then, there should be much, much more funding allocated to scale-up spintronics Non-Volatile-Memory (NVM) MRAM to High-Volume-Manufacturing (HVM).

This would help kickstart the transition to beyond CMOS spintronics technologies like MRAM (SOT-MRAM), Intel MESO concept,… that are key enabler to signicantly reduce (>100x to 1000 000x) digital logic and AI power consumption, depending on the architecture (analog or digital).

All this would make much more sense that stubbornly continue scaling up systems made up from highly energy inefficient silicon transistors.