Oracle wants to power 1GW datacenter with trio of tiny nuclear reactors

Oracle is going nuclear over growing demand for AI datacenters, and that's not a metaphor for Larry Ellison's mood. On Monday's quarterly earnings call, Oracle's founder, chair and CTO revealed the database giant and cloud provider had secured building permits for a trio of small modular reactors (SMRs) to power a datacenter …

[IvyKing]

USN's safety record with the Naval Nuclear Propulsion program has been exceptional, due in part to "the kindly old gent" being in charge at the beginning. One description of him was "basically honest but totally unscrupulous". The problem with SMR's is more regulatory/financial than technical.

[Gene Cash]

He was smart. He realized very early on that "if we have ONE nuclear accident, no matter how small, we're out of business"

Would be nice if, you know, we had leaders that planned ahead and understood technology and consequences these days.

[Yet Another Anonymous coward]

Although perhaps at the cost of combat effectiveness. He prioritized the safe operation of 'his' boats over the actually being used for anything.

Famously he didn't allow any commanders who had been decorated in WWII, on the grounds that winning a medal meant they had behaved recklessly.

Bomber command not losing a single airframe in WWII would have been an achievement but not necessarily an optimal goal

[Claptrap314]

I had not heard this story.

It is a major flaw of the lower ranks to discount the critical need for the flag ranks to manage their politics before their warfighting. Took me long enough to figure that out. (It is also a major flaw of the flag ranks to manage their politics to the exclusion of their warfighting).

[JRStern]

... and so you never heard about any accidents.

[Anonymous Coward]

>The problem with SMR's is more regulatory/financial than technical.

The problem with SMRs is that they are powered by wishful thinking.

There is no real technical basis for the expectation that they will be better - quite the opposite, small reactors inherently tend to result in more waste/GWh. They linearly scale up the probability of accidents. Many of them are actually single use throw away designs when you realise that the factory recycling proposed is exactly as realistic as plastic recycling has proven to be - if the plastic was radioactive as well.

Unfortunately they divert and muddy the waters of developing actually better nuclear designs, or as is the case here, they distract from the massive energy waste of vast data centres whose primary function is the concentration of wealth with little creation of new value.

[Displacement Activity]

data centres whose primary function is the concentration of wealth with little creation of new value.

Wow. And I thought it was to watch kitten videos.

[Like a badger]

"The problem with SMRs is that they are powered by wishful thinking."

I struggle to see how they'll be worse (in cost, waste, or safety) then the alternative huge, over-complex approaches like the Areva EPR ("At least three times over budget or your money back!). Hinkley C is currently heading towards an outturn cost of £46 billion...for one 3.2MW power station.

As an aside, that's roughly how much China invests annually in new coal power stations to give around 180GW of new capacity.

[Anonymous Coward]

It is pretty easy to see the potential for 20 x 50MW SMR's to have 20x the fault count of 1GW reactors. There is not any real reason (beyond wishful thinking) to believe the control systems will be inherently more reliable. The reactors are not going to actually be simpler.

Then consider the 1GW plant is not a cost-down design. It is not unlikely that the SMR will have a higher fault rate than bigger, more thoroughly engineered reactors as they will be a cost down.

That's if all 20 are on a single site.

Spread them around 20 separate data centres, and now you have 20x the attack surface.

Whereas on a nuclear site they will be run by Areva or someone else vaguely credible, in data centres they will be run by MS (recently bluescreened half the worlds air traffic) Google (just turns off anything after 5 years when they lose interest like a toddler) Amazon (left as an exercise for the reader)

The taxpayer will be left to pickup the cleanup and decommissioning costs for these, as all these companies simply abandon everything as soon as it doesn't make them a river of gold.

[Wellyboot]

I'd be surprised if the model for DC power wasn't along the lines of a simple zero liability model -

Contract to splash a billion or two to a 3rd party company that will provide x Gwh per day for y years. 3rd party company then build SMRs with excess capacity that can be flogged onto the open market.

It's down to legislation to provide a plan to prevent said 3rd party company just shutting up shop as the SMRs reach EoL instead of decommissioning.

[Snake]

RE: the problem

The problem with SMRs will be practical, not theoretical.

Oracle may have gotten permission to build an SMR, but getting permission to run a nuclear reactor is a whole different ball game. You need approval from the NRC, and to get *that* you need to not only provide workable evacuation and emergency procedures but also need to gain the acceptance and approval of these procedures from the local governments involved with the implementation of the aforementioned actions.

In other words, you need to get the acceptance from the local populace & their politicians of the proposed evacuation plan before the NRC will even consider the possibility of granting you an operating license.

I would advise my gracious European friends to read about Shoreham Nuclear Power Plant to see how well that can work out.

[collinsl]

Re: RE: the problem

> Shoreham Nuclear Power Plant

Didn't realise we had any in Sussex, I thought the closest place to that was Dungeness, and those are decommed.

[Alan Brown]

The problem is that Submarine sized reactors are entirely self contained and have an infinite supply of heatsinking just outside the hull

Scaling from 60-100MW up to 3500MW (you need that much to get 1000MW electrical output) in a steam based/water cored system is "difficult" as the engineering stresses go up with the CUBE of power

Rickover entrusted the design to Weinberg and Weinberg did a great job in producing a laboratory demonstrator based on the ww2 X10 plutonium breeder using available materials but it wasn't designed to scale to large sizes and was intended to be a stepping stone to better designs (It's wet steam because in the 1950s navy engineers deeply understood boilers and steam turbines. It can't be dry steam because the water inside the reactor is the thermaly limiting factor and waste happens because that wasn't a design concern for a demonstrator)

Weinberg went on to build a better mousetrap (The MSRE) and the USA response was to kill the project because it would have entirely divorced civil nuclear power from the weapons cycle thanks to not needing enriched uranium (which is an unwanted waste product of making bomb grade plutonium) and as such potentially endangerd the nuclear weapons industry by making enrichment plants and potentially even uranium mining itself subject to limitation treaties

"Small modular reactors" are fundamentally a fallacy because on the generation side you need as large a heatsource as possible - current thermal plants are about 800-1000MW because that's essentially the current materials limit of steam turbines. Smaller ones are uneconomic and nuclear heat with its wet steam is a MAJOR maintenance headache on the non-nuclear (revenue) side of the wall because it drives up maintenance costs due to erosion of the turbine blades

Weinberg's 1960s MSRE design addressed this by not using water in the core (hotter, not pressurised, 98% lower waste production) and the USA has tossed away a 60 year lead on clean energy systems due to 1970 cold warrior politics

[Pascal Monett]

"pilot projects have not gone well"

Why ? What have the problems been ? How many pilots have there been ?

I'd like a bit more info on that. Until we get actual fusion, I am convinced that nuclear (preferably Thorium) is the way to go. So what are these SMRs based on and why is it apparently so hard to get them working on land when they are working fine in nuclear submarines ?

[Julz]

Well

For starters, they are not surrounded by an unlimited supply of cooling fluid.

[Anonymous Coward]

Re: "pilot projects have not gone well"

I guess design is just so much easier when

(1) money is no object

(2) it's someone else's job to clear up the mistakes.

[IGotOut]

Re: "pilot projects have not gone well"

3. If it goes wrong it kills a few thousand soldiers and some inconvenient* marine life. On land it has the potential to do far, far more (visible) damage.

4. If there is an issue, it gets hidden due to "National Security"

[collinsl]

Re: "pilot projects have not gone well"

Also worth noting that nuclear engineers on naval vessels with reactors are very well trained, will staff it 24/7, and are very aware of their responsibilities and are invested in keeping the reactor from going critical due to their proximity to it etc.

Putting a small reactor on land with likely more lax government regulation than the current large power generation sites would probably lead to them being overseen remotely by some outsourced management company who then hire poorly trained workers to go and fix the thing who have no interest in it working and no investment in it not going critical beyond them not being in the blast zone when it happens.

[Matthew Elvey]

Real? Misleading?

Where's the evidence that Oracle "secured building permits for a trio of small modular reactors (SMRs) to power a datacenter"? This article links to another that doesn't mention 'nuclear' at all. I've looked for a record of an Oracle earnings call on Sep 9, and found it with some difficulty. Search for "Oracle Corporation - Q1 FY25 Earnings". It requires registration and won't be up long; I grabbed some sound bites. Larry starts talking 31 minutes into the call. He initially talks about huge data centers, but I heard no mention of nuclear, 'till the Q&A. 45 minutes into the call, he says: We're in the middle of designing a data center. A > 1 GW data center; they already have location and power. They've already got the building permits for 3 nuclear reactors. These are the SMR that power the data center. This is how crazy it's getting. So he says "they've", not "we've". I think there's some telephone being played or people are playing fast and loose with terminology. I also see that other's are running with the story anyway. Saw one report, "Oracle’s founder did not elaborate on the location for the data center or any construction timelines. Power Engineering has reached out to the company for more details and will report back when we learn more."

I agree that Thorium-base molten salt small modular reactor is the most promising technical path to carry nuclear energy into the new era of meeting the market demand, but I can't believe ones already got a building permit. On the other hand, a building permit is not an operation permit. That's probably the hardest one to get?

[JRStern]

Acres of GPU clusters OMG

>and it will contain acres of Nvidia GPU clusters

I'll pass on further nuke comments but acres of GPU clusters is OMG SMH.

[disillusioned fanboi]

While we're here, a word on Fusion

I'm currently studying Fusion for my son's school project. I can say with certainty that a fusion power remains science fiction.

Most fusion projects aim for Deuterium and Tritium fusion. It happens at an achievably low temperature, and has been demonstrated in various test environments. However Tritium is not naturally available, the best source is from the Canadian nuclear power stations which will soon be decommissioned. The D+T fusion produces energy in the form of a fast neutron, which we don't really know what to do with. There are various theories, some produce Tritium, but globally it sucks.

The next most easy is Deuterium with Helium3 (isotope of helium with only one neutron). Its nice in that it produces energitic charged particles which we could probably directly use to produce electricity. See this projects pitch for more info: https://www.helionenergy.com/articles/explaining-helions-fusion-fuel-choice-d-he-3/

However Helium3 is not naturally available on Earth. It might be available on the Moon. Globally its science fiction.

ITER is a D-T experiment to prove that excess energy can be generated. But there is no plan to capture the excess energy. Its not a power plant. And its costing billions. So fusion will not help solve any current problems. Its fun to research, but it has no foreseeable practical use.

Sorry guys, its not what I wanted to hear either....