Is AI going nuclear?
This week we’ll take a look at some of the recent buzz around nuclear energy for data centers, starting with a deep dive into a recent interview with the head of the DOE’s Loan Program Office, Jigar Shah.
-JM
Nuclear? Perhaps!
David Roberts from VoltsWTF logs a great interview with Jigar Shah, head of DOE's Loan Programs Office.
“The US needs to double or triple its nuclear capacity by 2050”
“Today, we're talking about persistent load growth that's six times higher than what we've experienced the last 20 years.”.
“[The hyperscalers are] writing white papers saying we don't know how to meet our 24 x 7 standard in 2030 without nuclear”
published: Feb 21, 2024 Source: VoltsWTF
Jake’s take: This is a friendly and thorough conversation around our growing need for “clean firm” energy sources, specifically nuclear. Jigar should need no introduction to this audience…
Full disclosure: I grew up in a household that was decidedly anti-nuclear. I fully acknowledge my bias here! That said, I believe that well-respected voices (like Jigar’s) are helping once-dogmatic folks like me to embrace the notion of nuclear power as something like a “least-worst option”.
I’m also very curious as to whether Mark Jacobson’s “No miracles needed” approach to our energy needs can accommodate the expected growth in demand.
Here’s what stood out to me from the podcast:
Nuclear is really hard, really expensive, and we have no existing economies of scale to support it.
“When it came time to build Vogtle, the National Building Trades trained 13,000 people in how to build a nuclear power plant”
“… I'm hoping that someone builds a new reactor soon so that they all don't get other jobs somewhere else”
“This country lost the ability to do big things — and on purpose.”
In the US, we seem to be much more effective when we think big, but start small and act locally. I believe the ongoing success of DERs (Distributed Energy Resources) like decentralized solar and storage is a good indicator here.
We missed the early opportunity to be the global leader in standardized, modular nuclear reactors.
“We have 92 operating reactors today in this country and no four of them are the same. So we decided that every single nuclear plant should be a snowflake”
I appreciated the history lesson here. The obvious question then is: Can we / Should we try to catch up? Does the US have a global obligation to lead the way in nuclear energy? Or, per Jacobson, would we better off taking all of time, money, and resources required to do so and apply it to what is already working and can scale (namely, renewables+storage)?
The data center load concerns are for real!
“And we have AI saying — I mean, Microsoft alone, I think, is saying we need 10,000 MW for ChatGPT. You've got Google, Amazon, Apple, Meta.”
“But let us just run a thought experiment for a second: The deciders of this are the hyperscale data centers, because they themselves need 200,000 MWh, 200 TWh of new power by 2030”
“close to 20 GW of new data centers that we need by 2030”
“So you know what they're [the hyperscalers] choosing to do instead? They're building a bunch of Bloom boxes powered by natural gas to run a lot of those off grid — and they're not off grid, it's like undergrid. So they have locations where the utility has half the power that they need. They get the other half from Bloom, and then they're waiting for the utility to catch up in two years when the interconnection gets full. And then, of course, the Bloom box continues to run. So I just want to make sure we're crystal clear about the fact that this is not a theoretical conversation.”
I’m not sure how precise Jigar’s numbers are, but you can find the report he references here, which does show 5yr growth expectations doubling from 2022 to 2023.
The hyperscalers (Amazon, Google, etc.) are driving major energy projects, and the federal loans to fund them.
“Microsoft and Google have publicly said that they need to be 24x7 matched with clean energy by 2030, they have a set of options”
“[Google’s] paper suggests that they believe that nuclear power is the likeliest to be able to meet this 24x7 matching requirement that they have.
“So then what they do with that data is they go to a utility and they say, ‘we would like a hamburger today, and I'll give you two hamburgers tomorrow.’"
“… so they're saying our CFO needs a data center next week because it's dependent upon us making $20 billion a quarter in profits. So we'll do that. And in exchange, we'll agree to pay an above-market tariff for 24x7 matched clean power that you will start construction of to supply us with this 24x7 matching that we need, whether it's in 2030 or whether it's in 2035.”
“So, if the utility says we just came to an understanding with Microsoft and Google and Apple and Amazon and Meta for geothermal at $150 a MWh or offshore wind at this price or whatever it is, we would do that loan.”
This section of the conversation took me a minute to digest. I think it goes something like this:
Climate-conscious Hyperscaler makes public commitment to clean energy.
Hyperscaler can’t buy 100% clean energy yet, because it isn’t available, so they contract in “24x7 clean energy matching”. They continue to consume less-than-clean local energy but pay (a premium) for clean energy to be generated somewhere else (consumed by someone else). This lets the Hyperscaler say they purchase 100% clean energy, even if that’s not exactly what they’re consuming.
Hyperscaler can’t even get enough local dirty energy, so they install natural gas-powered fuel cells to generate what they need. As Jigar mentioned, they’ll continue to run even after the initial clean energy goals are met, which I suspect is due to a long expected payback period on the fuel cells.
Demand for 100% clean energy is so high (driven by the need to build the next data center, etc.) that the Hyperscaler can’t find enough existing energy-matching contracts. So, they negotiate a deal with the utility to buy 100% clean energy in the future, and at a premium price.
Utility co takes that contract for future clean power to Jigar and says “give me a loan to build the new clean power plant I need to fill this contract”.
Jigar seems to be saying that he’ll fund nuclear projects all day long if the utilities come to the DOE with contracts to provide nuclear power to the hyperscalers.
For a look at volumetric vs. emissions vs. temporal (24x7) energy matching, check out: Most net-zero energy strategies have no real emissions benefits, researchers warn.
For a critical view of 24x7 energy matching: Every load an island: Requiring hourly matching of clean electricity purchases would raise emissions
SMR’s (small modular reactors) are not the panacea that many have suggested…
“They're neither small nor modular, but they are reactors”
“it is structurally more expensive to do an SMR than to do an AP1000 (an average-sized reactor)”
We’re also still many years away from commercial viability, with Jigar noting that the first commercial microreactor should come online “the day before 2030”.
Also, SMRs may “actually increase the volume of nuclear waste in need of management and disposal, by factors of 2 to 30” according to a recent study.
For some basics on SMRs, see this IAEA page.
And, don’t worry about the nuclear waste…
“coal plants in our country have produced ten times more nuclear radioactive waste than our nuclear fleet”
“I could make my bed out of it and sleep on it, and I would get more radiation from the atmosphere than I'm getting from the waste”
Though I appreciate the sentiment that coal is the real enemy, the idea that nuclear waste is nothing to worry about runs counter to everything I’ve ever learned. I’m certainly not an expert, but a quick look at the U.S. NRC seems to confirm that this stuff is nasty: “10 years after removal from a reactor, the surface dose rate for a typical spent fuel assembly exceeds 10,000 rem/hour – far greater than the fatal whole-body dose for humans of about 500 rem”
All in all, a really interesting conversation. I’d love to see David Roberts host a debate between Jigar Shah and Mark Jacobson on their dueling "All-of-the-Above vs. No-Miracles-Needed" approaches to our energy demands!
AWS Acquiring Hyperscale Data Center Directly Connected to Nuclear Plant
Cumulus data center campus owner and nuclear power plant owner Talen Energy agreed to sell the facility to AWS for approximately $650 million
Published: Mar 5, 2024 Source: MicroGridKnowledge.com
Jake’s take: Just in case you thought the conversation was too theoretical… Hyperscaler Amazon will be specifically and intentionally powered by nuclear energy.
…and I can see how it makes sense to put the data center “behind the fence”. The DC doesn’t then burden the downstream transmission lines, and its construction likely benefitted from easier permits due to its colocation with the power plant.
Also, $650M is big business! Cumulus Data reportedly developed the site speculatively and then found a buyer. I’ve gotta imagine we’ll see more deals like this in the coming years.
Future data centres may have built-in nuclear reactors
“A normal data centre needs 32 megawatts of power flowing into the building. For an AI data centre it's 80 megawatts”
“Our industry has to find another source of power”
“nuclear power is the only answer to the pressures the new AI data centre world faces”
Published: Feb 14, 2024 Source: BBC
Jake’s take: The focus of the article is SMRs in the UK, about which most of the interviewees are bullish. There is an acknowledgment of the timing difficulties; as one contributor puts it: “it will take a long time to deploy a nuclear-configured data centre in the UK, and AI is happening now”
That’s a wrap! EN+AI seems to be moving in a less frequent, but more detailed direction. Let me know what you think- do you prefer the lighter-weight, more frequent (roughly weekly) posts, or the more detailed (but slower, roughly monthly) view? Or, both?
Either way, thanks for reading!
In case you missed it- here are your favorite links from previous editions of EnergyNews.ai:
Rethinking Concerns About AI’s Energy Use
Daniel Castro provides a comprehensive look at the concerns around AI’s energy consumption.
published: Jan 29, 2024 Source: Center for Data Innovation, PDF here
Why AI and energy are the new power couple
AI is increasingly vital in managing complex, data-rich power systems, especially with the rise of renewable energy sources. This IEA.org article highlights AI's role in improving the predictability and efficiency of power supply and demand, particularly in renewable energy. It also delves into AI's contribution to predictive maintenance, ensuring more robust and reliable energy infrastructures.
source: iea.org