Hacker News Reader

Comment by Filligree | Hacker News Reader

Nuclear is a reasonable source of green power, but this is the fourth-or-so time I've seen a MAANG suggest they're going to use it. I'd ignore this until someone breaks ground.

View on HN

MANGA?

(Or would that be the alternative history version where the companies were Japanese)

MAANG is pretty established, because it just follows from FAANG and Facebook->Meta.

I've never seen it before. And MANGA is much funnier and more memorable. Netflix seems like it should be ditched in favor of Microsoft though.

Also haven't seen it before. MAGMA would be a pretty sweet abbreviation!

It’s a pretty common acronym

Microsoft Apple Google Meta Amazon

You have to pronounce "MAANG" like Tony Montana would.

If we're gonna just arbitrarily name enshitified companies I propose the following instead:

Facebook

Ubisoft

Chrysler

Kroger

Enron

Verizon

Exxon

YouTube

TikTok

Twitter

Intel

Netflix

Google

Oh I see what you did there. Very accurate acronym from that list.

Somewhat confusing when GP uses it to refer to Microsoft, though.

*breaks atom

It seems like a really dumb idea to use the most expensive form of electricity generation by some margin for a very predictable work load.

Most expensive if you discount reliability, longevity and persistence in energy generation.

A predictable energy generation source for a predictable work load makes sense.

Nuclear fans grossly overstate the cost of dealing with renewable intermittency, especially if the load has dispatchability. Servers don't necessarily have to run 24/7.

> Servers don't necessarily have to run 24/7.

I'm curious, do you actually run servers for a living? I'm wondering what a typical use case is (aside from crypto mining) where powering off and on physical servers due to demand, cost, or availability of power is a thing.

In fairness, training machine learning models is definitely one of areas where only running when it’s cheap to run makes some amount of sense. It’s something you want done eventually, not something that is responding to an immediate need.

In practice…yeah. The machine’s your running your models on are expensive, the people working on the model waiting extra time to check the new model are expensive, and not keeping up with your competitors because you only work half the time is not generally a wise choice for someone like Microsoft.

I have worked for a smaller company that trained a model on AWS when spot instances were cheap…until they got large enough to purchase some hardware best suited to their use case that could run almost constantly, then doing everything except training their model in the cloud. So yeah I also doubt this makes sense in practice.

The cost of supplying the servers with renewables 90% of the time would be much cheaper than supplying them 100% of the time. The former can easily be done with some batteries; the latter requires more extensive consideration of rare dark-calm periods.

Another consideration is that servers can be located in the best places for renewables, not in places with high seasonal variability and relatively low winds.

> Servers don't necessarily have to run 24/7.

The benefit of your services having 25/7 availability aside, if the capital costs are high compared to operational costs, reducing operational time to save on operation costs is a bad trade.

"The wind is blowing! Quick, everyone run your Hadoop jobs!"

-- A++ customer experience

If you invest a lot of money into very expensive Nvidia training hardware you certainly want to run them as close to 24/7 as possible.

Dispatchable load usually means oversizing the dispatchable consumer to get the same overall output. This is already uneconomical for even particularly energy intensive industries (e.g. aluminum smelting). I would assume server hardware is a lot more capital intensive than that.

While I can’t speak to aluminium specifically, smelters in general have the problem that they’re pumping molten metal around. And the thing that keeps it molten inside the pipes is, usually, just inertia.

Pausing the plant doesn’t just mean pushing a button. Safely shutting it down is a long process of draining every single part of molten metal first, and might not even be doable; almost any other factory would have an easier time.

Molten cryolite rather; I think the metal pooled in the bottom of a pot could be drained off or even allowed to solidify (it would still be conductive.)

A related problem is that the walls of the pot cannot be allowed to get too hot. If this were not the case, they could simply insulate the pots to reduce heat losses to some arbitrarily low level.

What you think that will accomplish? We already schedule industrial loads for times of low demand from residential and commercial. Rescheduling those loads to coincide with high intermittent generation usually requires more transmission and distribution capacity.

Debating total cost depends almost entirely on the particular grid in question. For all we know, those "nuclear fans'" estimates could be spot-on but completely irrelevant to you.

No, but surely Microsoft could work out a deal with local utility companies to sell any excess electricity they generate?

Fossil fuel plants are far and away the most expensive form of electricity generation, we just allow them export a big chunk of the costs onto other people so we can pretend it's cheap.

Citation needed

Okay. Per [1] the cost of 1 ton of CO2 emitted is about $3,000. Per [2] fossil fuel plants emit about 1-2.3 pounds of CO2 per kWh, or about 0.0005 to 0.0012 tons per kWh. Multiplying that out (0.0005 * 3000) yields about $1.50 to $3.60 per kWh, accounting only for the cost of emissions.

Per [3] nuclear is about $0.07 per kWh. That's about 20-50 times cheaper than fossil fuel plants.

It would be fair to round that up a bit for construction and fuel extraction and refinement emissions, but don't forget to do that for fossil fuel plants, too.

[1] https://en.wikipedia.org/wiki/Social_cost_of_greenhouse_gas_...

[2] https://www.eia.gov/tools/faqs/faq.php?id=74&t=11

[3] https://css.umich.edu/publications/factsheets/energy/nuclear...

Source [1] shows $300 not $3000, which changes the math quite a bit.

China and India are building nuclear power plants pretty cheaply.

China's nuclear plans are consistently not being met. India's nuclear output is quite small, and is already exceeded by a factor of 2 by their PV output.

My point was that both are finishing several plants per year, with reasonable cost ($3k per kW compared to $8-12k per kW in the US) and within reasonable time frames of 5-10 years of construction. This is competitive with renewables if you take the cost of storage into account.

They are not competitive with renewables, which is why China and India are building so much more renewable capacity than they are nuclear capacity.

Ultimately, storage will be needed. But at that point, nuclear becomes even less desirable in those countries, because renewables are creating long periods where there's no demand to soak up the nuclear plants' output. You might object that this is a renewable problem, but I'd retort it's a nuclear problem -- they can't save much money by curtailing output, and this inflexibility is a blemish on nuclear.

With reasonable projections of the cost of storage, PV will end up much cheaper than nuclear in India, btw. India's a great place for PV; they could go full renewable with just PV and batteries, not needing wind or hydrogen.

Even with nuclear, ultimately storage will be needed. Throttling production of such an expensive piece of capital with low operational costs doesn't make much sense, it better to build to the average power needed and smooth out the peaks and valleys with a battery.

But if you have a battery, it makes more sense to fill it with low LCOE renewables rather than nuclear. So adding a battery cannot provide much revenue to the nuclear plant.

I agree, I just think it is funny when you have a large majority of your power coming from renewables or nuclear, you face similar looking problems. You are faced with the choice of artificially lowering a nuclear plants capacity factor by throttling down, or using batteries to keep the capacity factor high.

The price of grid batteries in the future will basically fully determine what power generation technologies we use.

China completes multiple reactors per year, but India does not. Here are the power reactor statistics for both countries from the International Atomic Energy Agency's Power Reactor Information System (sort by date of first grid connection):

India has connected 3 reactors to the grid since 2013-01-01:

https://pris.iaea.org/PRIS/CountryStatistics/CountryDetails....

China has connected 39 in the same time period:

https://pris.iaea.org/PRIS/CountryStatistics/CountryDetails....

China is barely building nuclear anymore. China added more wind and solar the past nine months than all of its nuclear reactors under construction will provide. Yes, that includes capacity factor.

https://twitter.com/yo_ean/status/1718633487454904718

Reply on news.ycombinator.com