News Oracle will use three small nuclear reactors to power new 1-gigawatt AI data center

[Admin]

Oracle revealed during its quarterly earnings call that it has secured permits to build a trio of small modular nuclear reactors that could deliver power for a 1-gigawatt AI data center.

Oracle will use three small nuclear reactors to power new 1-gigawatt AI data center : Read more

 

[King_V]

This seems kind of an odd choice, given the time scale involved.

 

[sunnykabs]

Oracle revealed during its quarterly earnings call that it has secured permits to build a trio of small modular nuclear reactors that could deliver power for a 1-gigawatt AI data center.

Oracle will use three small nuclear reactors to power new 1-gigawatt AI data center : Read more

If you need 800MW, why build 3 SMRs, Might as well build an AP1000, 1.1GW. Proven to work, already complete design ready, certified. 4 -5 years construction and online in 5-6 years.

 

[Sluggotg]

If you need 800MW, why build 3 SMRs, Might as well build an AP1000, 1.1GW. Proven to work, already complete design ready, certified. 4 -5 years construction and online in 5-6 years.

I did a couple of stints in the US Navy Nuclear Power Program and 30 years at a Commercial Nuclear Power plant as an Operator. Large ones, (like the one I worked at), are generally refueled once every 2 years, (it used to be each year). When the plant is down for refuel and maintenance you are production nothing. With SMRs, (like the ones from Nuscale), you can shut down a single reactor for maintenance/refueling. This saves time and money.

The new SMRs are really nice. Operating and Maintaining them is much easier. The article compares them to Naval Reactors. They are nothing like Naval Reactors. Navy Reactors are fueled to last many years and are extremely Heavy Duty. They are not commercially viable nor are they a good choice for civilian plants.

It is a long subject butI hope Oracle pulls it off!

 

[mras2]

If you need 800MW, why build 3 SMRs, Might as well build an AP1000, 1.1GW. Proven to work, already complete design ready, certified. 4 -5 years construction and online in 5-6 years.

Cause, a nuclear plant only has 80-90% uptime.
You need 3x for a 2+1 redudancy.
No SMR does 500-666MW,
Currently best PROJECTED SMR's, does about 150MW electricity tops a piece.
This will end up being extremely expensive, and a poor businesscase...
Article says a GW, thats 1000MW, not 800MW.

 

[gg83]

I love the concept of SMR!!!! It's great to see big companies push for the tech.

 

[ThomasKinsley]

I'm sure tech companies will handle nuclear energy with the exact levels of wisdom, thoughtfulness and care that they handle their codebase.

 

[tennis2]

Thought process:

• HA, I want a mini nuclear power plant in my PC
• You know what looks like a nuclear cooling tower, LN2 pots
• etc etc
• What if you could capture the waste heat from a CPU (or multiple) and compress it into something hot enough to generate electricity (thermoelectric generator or something). I wonder what the recovery efficiency would be.... could you regain..50%(?) of the power you used/wasted?
  • Ed: Looks like 5-15% efficiency on TEGs...bummer

 

[USAFRet]

I'm sure tech companies will handle nuclear energy with the exact levels of wisdom, thoughtfulness and care that they handle their codebase.

Kodak, ostensibly a "tech company", had their own small reactor for decades.
https://www.nbcnews.com/id/wbna47417980

 

[emike09]

I'm 100% for nuclear. People with stigmas against it need to assess the facts and realize its potential. 583 people died in a single aviation incident in 1977, but we're all still flying more than ever because we realize the benefit and convenience of it.

 

[Kondamin]

I'm sure tech companies will handle nuclear energy with the exact levels of wisdom, thoughtfulness and care that they handle their codebase

Disney has a license, let that sink in

 

[ThomasKinsley]

Kodak, ostensibly a "tech company", had their own small reactor for decades.
https://www.nbcnews.com/id/wbna47417980

Apparently they had scientists working on the Manhattan Project too. Wild stuff.

 

[The Historical Fidelity]

Cause, a nuclear plant only has 80-90% uptime.
You need 3x for a 2+1 redudancy.
No SMR does 500-666MW,
Currently best PROJECTED SMR's, does about 150MW electricity tops a piece.
This will end up being extremely expensive, and a poor businesscase...
Article says a GW, thats 1000MW, not 800MW.

Just giving the article the benefit of the doubt…maybe they mistook the anticipated thermal power of the 3 SMRs for the electrical output. The rule of thumb is around 1/3rd of thermal power = maximum electrical output. For example, the nuclear power station by me has a thermal capacity of 3853 MW per unit and a maximum electrical capacity of 1280 MW per unit.

 

[Geef]

Cause, a nuclear plant only has 80-90% uptime.
You need 3x for a 2+1 redudancy.

So instead of a RAID array with 2+1, you have a RAIPP... Hmm.. I guess 'Power Plants' replacing Disks on RAID won't work. Nobody wants to work at a 'RAIPP plant.'

That night at a bar. Lady "So what do you do?" Guy "I work at the local RAIPP plant!"

 

[das_stig]

Just build them in a desert or other unpopulated area, feed it the grid and then let the grid supply it to your DC free and agree a price on the surplus the grid can use for other uses, win for all.

 

[The Historical Fidelity]

Just build them in a desert or other unpopulated area, feed it the grid and then let the grid supply it to your DC free and agree a price on the surplus the grid can use for other uses, win for all.

Two problems with that, transmission loss vs having the reactors on site would be a dealbreaker in principle, and the whole point of on-site power production is to isolate from power grid faults.

 

[TheJoker2020]

I'm sure tech companies will handle nuclear energy with the exact levels of wisdom, thoughtfulness and care that they handle their codebase.

Thankfully they won't be getting their hands on any nuclear facilities, and I really hope not their codebase, best leave it to the Pro's

 

[TheJoker2020]

Two problems with that, transmission loss vs having the reactors on site would be a dealbreaker in principle, and the whole point of on-site power production is to isolate from power grid faults.

This is currently one of the many problems with the insane politicucks that have been running the UK (my country) for the last 27-years, they have gone balls to the wall "wind power" and we now do not just have (what is essentially) the most expensive electricity in the World.

The UK has (by design) deindustrialised to the point that we have no Aluminium smelters at all meaning 100% import and export of Aluminium to/from the UK which is the opposite of "green" , no coal power stations (that are burning coal, we have one burning wood imported from mainly Canada because of the "greenness".!).

No new Nuclear power stations have been built in the UK in 20-years but existing ones have come offline, and now the politicucks in charge are further ruining the countryside laying hundreds and hundreds of miles of thick copper power cables because they need this vast network of power cables that link together thousands of windmills strewn well beyond the length and breadth of the country, which is where my point links directly to yours, the transmission losses can be absolutely huge, and this is one of the many (many, Many) drawbacks of wind power (especially, solar also, but not as much). The massive use of copper cabling and having to dig up the countryside (for the 20% that is underground by distance), and build pylons everywhere because the windmills are just not good enough on their own for the other 80% of this massive expensive power cable network.!

If you want to know what insanity look like, this is it.! SMR and other new Nuclear types (there are several, Thorium Sodium is talked about the most) are obviously the way forward, but they are still years away, so what is Oracle's timescale here.?

 

[Kondamin]

Just build them in a desert or other unpopulated area, feed it the grid and then let the grid supply it to your DC free and agree a price on the surplus the grid can use for other uses, win for all.

You want those in a location with abundant cooling water, and considering the history of safety and nuclear power plants, i'd much rather have one of those in my backyard than any other form of large scale power generation

 

[TheJoker2020]

Kodak, ostensibly a "tech company", had their own small reactor for decades.
https://www.nbcnews.com/id/wbna47417980

There has been a small one (that is public knowledge) in the centre of London for many years now, there are more than people think, and most of them are for governmental / military use as per the one in London, there will no doubt be many more in use like this dotted around the World.

 

[Kondamin]

There has been a small one (that is public knowledge) in the centre of London for many years now, there are more than people think, and most of them are for governmental / military use as per the one in London, there will no doubt be many more in use like this dotted around the World.

plenty of universities have nuclear reactors, it's a very useful technology with a lot of practical applications.
ending nuclear power will do nothing reducing the waste it will just lower the amount of money that will flow in to managing it.

 

[TheJoker2020]

Apparently they had scientists working on the Manhattan Project too. Wild stuff.

They went to the US government shortly after the first nuke was set off and asked them what was going on because they had very clear evidence that something had happened, and if they could tell because it showed up on their film (that they were selling to the C_A for their spy planes) and other film manufacturers would also be able to tell as the fallout spread.!

The film was then later used to identify the spread of fallout, how much, where and when. Literal patterns (wind) shown for detonation after detonation (for decades not just 1945) via Kodak film.!!! Two impressive stories in one...

As for whether any of their scientists worked ON the Manhattan Project before this event I do not know, I am not an expert in this field, but know of the above Kodak incidents (it was on QI (Quite Interesting) many years ago) and it's not something I would forget unless I had dementia.

 

[TheJoker2020]

plenty of universities have nuclear reactors, it's a very useful technology with a lot of practical applications.
ending nuclear power will do nothing reducing the waste it will just lower the amount of money that will flow in to managing it.

Very true, I had almost forgotten about them, possibly for the following reason. They rarely make more power than they produce so are never included amongst the number of "Nuclear power stations / reactors" listed in the "news" because they produce nothing and the articles rarely talk about the technology, only the power output number and of course the heavy emphasis about nuclear danger.!

 

[TheJoker2020]

3853 MW

This one.? If so, you are easy to find just by that number, that was the first hit of the search for 3853 MW

https://pris.iaea.org/PRIS/CountryStatistics/ReactorDetails.aspx?current=764

 

[edzieba]

Research reactors (like Kodak's, and those at universities) are not power reactors. Their electricity output is zilch. The vast majority are also pulsed reactors that generate a very rapid ramp up and ramp down in reactivity and can be cooled simply by being immersed in a sufficient volume of water - total output power over time is extremely low. A power reactor on the other hand produces heat in a steady state, and it is dealing with that constant reactivity and heat output that is the challenge. Power reactors are far more complex and with more support machinery than research reactors.

As for economic viability: large datacentres draw so much power that they effectively require that power capacity to be added to the grid SOMEWHERE in order to support them. Building the power plant alongside the power consumer cuts out a lot of grid capacity work, and makes pricing the cost of building that plant a direct cost rather than something that needs to be spread over a few years/decades of projected power usage charges (which need to price in risk, e.g. that the datacentre operator folds and ceases operating a year into the contract and that power plant is now effectively unfunded).