Amazon buys stake in nuclear energy developer in push to power data centres

I've been saying for years that we'll know the world is serious about energy when we see big pushes into nuclear power. We're here now. This is the point at which all the dumb "but what about a gorillion turbines and solar panels" start to die off (except for where they're legitimately a great solution) and talk about the only viable power source moving forward will start to grow. This is super exciting to me.

I just wanted to take the opportunity to say something positive about these companies, since it's so rare that I find anything positive to say.

I applaud these efforts. The amount of energy that's going to be burned by these things is crazy. Whether or not that energy use is wasted is an open question.

I think it's a positive thing if the companies making these bets don't offload the costs of those bets onto public utilities, and also that they're looking at energy sources that have less of an adverse environmental impact than things like coal.

Kudos, guys.

It feels like we're in a golden age of hard tech:

* A major advance in spaceflight [0]

* a ton of private investment in nuclear power [1]

* AI models performing at PhD-level on some tasks [2]

I know it can feel low-status to admire these accomplishments -- it feels like we're aligning with/submitting to the people behind them -- but I perceive technological growth to be accelerating across a bunch of fields that matter to me.

[0] https://news.ycombinator.com/item?id=41827362

[1] https://news.ycombinator.com/item?id=41858961

[2] https://www.nature.com/articles/d41586-024-03169-9

Related:

Google commits to buying power generated by nuclear-energy startup Kairos Power

https://news.ycombinator.com/item?id=41840769

Three Mile Island nuclear plant restart in Microsoft AI power deal

https://news.ycombinator.com/item?id=41601443

There's a nice synergy between data center operations and nuclear power. Similar requirements for high reliability, complex systems, redundancy, even things like electrical grids and cooling, and industrial HVAC are crossover points

In other words, a lot of engineers could work quite happily in both fields.

AI training demand should in theory be extremely elastic. Scale up when and where power is cheap, scale down when and where it is expensive. Training larger models takes a long time, and storing or moving the current training checkpoint is not that expensive.

Inference is somewhat elastic - people want relatively low latency - they might be able to tolerate an extra round-trip around the world, but probably not waiting until a time when there is more total capacity.

However, the big impediment to using cheap and green power is the capital cost of the training hardware; that can't be moved around in a hurry, so its capacity goes unused when the sun isn't shining and the wind isn't blowing. Much of the cost of the high-end data centre oriented GPU hardware is likely not incremental cost for nvidia - it is recovering fixed costs, or profit. In addition, people buying the hardware fear it will depreciate, so they have a limited time frame to use it.

So it is fair to say that it is nvidia's pricing strategy that is a significant driver of Google, Microsoft and Amazon investing in nuclear.

I would have thought that X-Energys HTR would be better suited for high-temperature process heat applications than electricity production for DCs

This is great news. The Xe-100 reactor design is not that different from the Kairos Hermes reactor that was yesterday in the news [1]. They both use TRISO pebbles, they are both slow (or thermal) reactors, and they are both SMR's that produce very nearly the same output (80 MW for Xe-100, 75 MW for Hermes). The only difference is the coolant: it is helium for this reactor, and FLiBe molten salt (a mix of lithium and beryllium fluorides). Helium is a gas, so it's much less dense than a molten salt, which is a liquid. Surprisingly, helium has a very high specific heat (5.2 J/gK), higher even than the specific heat of water (4.2 J/gK). FLiBe has about 2.4 J/gK specific heat. Helium can travel through the core at a much higher speed, so overall, the disadvantage of being a gas is not that great. The Hermes reactor is projected to have about twice the power density of the Xe-100 reactor.

There are some great things about Helium though. One is that Helium-4 (which represents 99.9998%) is the only isotope in the universe that has exactly zero neutron absorption cross-section. All other coolants absorb some of the neutrons in a reactor, but helium doesn't. It is also an inert gas, so it does not pose the various corrosion problems posed by other coolants and moderators (like water for NuScale, FLiBe for Kairos, sodium for Natrium, lead for Westinghouse).

China did already put in production two reactors of the same type as Xe-100, the HTR-PMs. They appear to be working just fine.

Good luck to X-Energy.

[1] https://news.ycombinator.com/item?id=41840769

We now have the three tech giants moving in on nuclear. I think we will have a race to see which SMR provider comes online first even though Nuscale canceled its pilot plant, sending shockwaves through the industry last year, there's a clear demand now. I wouldn't be surprised if Microsoft signs with an SMR provider in the coming weeks because the other players have. I would have said it would be Nuscale, but Sam Altman might move them to Oklo.

Wait... But everyone says wind+solar+batteries are so much cheaper than nuclear. Why aren't big corpos pouring money in these instead to get even more output?

Is Uranium relatively easy to find to power these reactors, or is the next wave of startups going to be space/deep ocean mining operations to find it?

I feel like this is only happening because financial projections were discovered to be impossible to meet with the current power generation and the trends for building new power were insufficient.

That was a great deal to dismentel AREVA in France to kick them from the edge of the Nuclear business and bring power back to America \o/

I'd rather wait for fusion. Fission is dirty and creates more radioactive waste than we know how to deal with. A tsunami hitting a fusion plant would not do much except knock it out of commission. Fission on the other hand ....well, we've seen that already. We've seen many times how messy fission is.

Make them pre-pay into a clear up fund. And carry massive insurance funds.

No privatizing the profits but then socializing the cost of the waste disposal (and leaks).

If the public has to pay for it as a EPA superfund site for cleanup, well then Amazon should then be sued out of existence for this.

And... there it is, 11 paragraphs in: "The first Xe-100 SMR is being developed at a Dow manufacturing site on the Texas Gulf Coast, with financial support from the US government." (Emphasis mine.)

From Admiral Rickover's 'Paper Reactor' memo (1953)[1]: "An academic reactor or reactor plant almost always has the following basic characteristics:

1. It is simple. 2. It is small. 3. It is cheap. 4. It is light. 5. It can be built very quickly. 6. It is very flexible in purpose (“omnibus reactor”) 7. Very little development is required. It will use mostly “off-the-shelf” components. 8. The reactor is in the study phase. It is not being built now." (Emphasis mine, again.)

I've seen so many of these come (and go), I think they actually use Rickover's paper as a blueprint for their marketing pitch. Xe's site on the Xe-100[2], "A simple design & modularized components" where, "Our elegant and simple design maximizes the use of off-the-shelf components manufactured and shipped to site using existing road & rail." <shakes head>

I think safe and affordable nuclear power could do a great deal of good. But I've seen "small modular reactor real soon" (including pebble-beds) for thirty years, so I'm reflexively skeptical. I checked to see if I could find any evidence this one is beyond the paper reactor stage, but my search-fu was insufficient to find it, presuming it exists.

[1] https://whatisnuclear.com/rickover.html (The whole paper is well worth reading, IMHO) [2] https://x-energy.com/reactors/xe-100

At today's live event (to accompany the news) Amazon Web Services (AWS) announced plans for 5 gigawatts of small modular nuclear reactors: https://www.youtube.com/watch?v=2hFBKIt_YfI

Let's begin with a quote from Yann LeCun (Vice-President, Chief AI Scientist at Meta):

  AI datacenters will be built next to energy production sites that can produce 
  gigawatt-scale, low-cost, low-emission electricity continuously.

  Basically, next to nuclear power plants.

  The advantage is that there is no need for expensive and wasteful
  long-distance distribution infrastructure.

  Note: Yes, solar and wind are nice and all, but they require lots of land
  and massive-scale energy storage systems for when there is too little sun
  and/or wind. Neither simple nor cheap.

https://x.com/ylecun/status/1837875035270263014

No battery farm can protect a solar/wind grid from an arbitrarily extended period of bad weather. If you have battery backup sufficient for time T and the weather doesn't cooperate for time T+1, you're in trouble.

Even a day or two of battery backup eliminates the cost advantage of solar/wind. Battery backup postpones the "range anxiety deadline" but cannot remove it. Fundamentally, solar and wind are not baseload power solutions. They are intermittent and unreliable.

Nuclear fission is the only clean baseload power source that can be widely adopted (cf. hydro). After 70 years of working with fission reactors, we know how to build and operate them at 95%+ efficiency (https://www.energy.gov/ne/articles/what-generation-capacity). Vogtle 3 and 4 have been operating at 100%.

Today there are 440 nuclear reactors operating in 32 countries.

Nuclear fission power plants are expensive to build but once built the plant can last 50 years (probably 80 years, maybe more). The unenriched uranium fuel is very cheap (https://www.cameco.com/invest/markets/uranium-price), perhaps 5% of the cost of running the plant.

This is in stark contrast to natural gas, where the plant is less expensive to build, but then fuel costs rapidly accumulate. The fossil fuel is the dominant cost of running the plant. And natural gas is a poor choice if greenhouse emissions matter.

Google is funding construction of 7 nuclear reactors. Microsoft is paying $100/MWh for 20 years to restart an 819 MW reactor at Three Mile Island. Sam Altman owns a stake in Oklo, a small modular reactor company. Bill Gates owns a stake in his TerraPower nuclear reactor company. Amazon recently purchased a "nuclear adjacent" data center from Talen Energy. Oracle announced that it is designing data centers with small modular nuclear reactors.

In China, 5 reactors are being built every year. 11 more were recently announced. The United Arab Emirates (land of oil and sun) now gets 25% of its grid power from the Barakah nuclear power plant (four 1.4 GW reactors, a total of 5.6 GW).

Nuclear fission will play an important role in the future of grid energy, along with solar and wind. Many people (e.g., Germany) still fear it. Often these people are afraid of nuclear waste, despite it being extremely tiny and safely contained (https://en.wikipedia.org/wiki/Dry_cask_storage). Education will fix this.

Nuclear fission is safe, clean, secure, and reliable.

The market has spoken. We need a shitload of reliable electricity and can't rely on the sun or wind.

Instead of this, it might be interesting to see Amazon (or similar provider) offer some kind of compute tier whose performance (not memory or storage, just cpu/gpu) is dependent the amount of power from sun/wind being generated. For many internal applications, I could see that being an acceptable compromise for some amount of cost savings.