Nuclear Power Is Actually Not Bad. Microsoft is Bad.
As a native of Central Pennsylvania, I’ve followed Three Mile Island since, well, ever. The 1979 accident happened several years before I was even born, let alone able to understand the principles of nuclear fission (I probably didn’t get there until my thirties). I have mixed feelings about the news I heard this morning that Constellation and Microsoft have inked a deal to restart the closed Unit 1 reactor. It’s not because I have particularly mixed feelings about nuclear power– more that I have mixed feelings about where the power is going from this plant. Let’s look at both elements here.
A Brief History of Three Mile Island
The power plant is, of course, notorious for a 1979 partial meltdown that occurred shortly after it came online. The accident occurred as a product of a combination of multiple failures of system legibility as well as mechanics. I am always telling my students in my surveying class about the importance of legibility alongside precision and accuracy– it is never enough to say you’ve gotten “precise” information if you don’t have accurate information, or if you are receiving (or delivering) that information to someone else in a way that they can’t readily understand. The examples I give are from using high-tech surveying equipment as opposed to low-tech equipment, given that you can get a lot more information from the high-tech equipment, but it’s a much steeper learning curve to know what to do with it.
Similarly, a nuclear power plant is a very high-tech ecosystem of machinery, relying not only on physics of atomic interactions to create heat but also on humans to be able to interpret thousands of signals from the machinery that supports this invisible and incomprehensibly weird thermal process that generates some of the safety and most reliable energy. In the case of the TMI accident, operators both received different information (poor legibility) and misinterpreted the information they received (poor training and operator error) about a single valve in the coolant system that led to a cascading series of failures that ultimately resulted in a nearly catastrophic event. The catalyzing series of events here took place over a few minutes starting in the wee small hours of the morning, while the following failures leading to the meltdown took all of a few hours.
It took days to figure out what had actually happened, and, indeed, years to assess the full extent of the damage.
Implications of TMI for the Industry
The accident raised existential debates over the future of the nuclear power industry after a decades-long construction spree, and was arguably the beginning of the end of the age of nuclear hype. Few reactors were built after TMI, with the notable exception of the expansion of the Vogtle plant in Georgia (built in 1976, but with new units that opened in 2013, 2023, and 2024). US nuclear production peaked in the late eighties, in spite of vast improvements to safety in reactor design and, indeed, completely new designs for reactors that are smaller and safer. The Chernobyl accident in 1986– far worse than TMI in every way, especially the human impact of how many people died and how many more were permanently uprooted from their homes- was less disastrous for the American industry as it highlighted the Soviet penchant for corner-cutting in designing “too-big-to-fail” systems that, well, did, and subsequently lying about what happened– but it still haunts the popular imagination.
Critics argued that the TMI failure showed that this technology was, unequivocally, too dangerous to be used. Charles Perrow wrote about the idea of “normal accidents” that are inevitable to occur in systems with preposterous levels of complexity, in which a preposterous number of things can go wrong, in a highly technologically advanced society. Perhaps the most famous note on the historical context of the meltdown is that it occurred less than two weeks after the release of The China Syndrome, a thriller film with Jane Fonda, Jack Lemmon, and Michael Douglas, about a turbine trip and resulting depletion of feedwater leading to a core meltdown, in a situation in which gauges indicate conflicting readings. This is almost to the exact letter what happened at TMI.
As an unapologetic junkie of all media, I find this historical aside pretty interesting, but one might deduce that it’s not completely outside of the realm of something that could be imagined. 1979 was, after all, Peak Crisis– ranging from the 1973 oil crisis to the Iranian Revolution, which would effect a second oil crisis, and, well, we could go on. Also, peak Cold War– not only thinking about the paranoia of nuclear annihilation by way of domestic, civil accidents, but also by way of warfare, for example, in the 1977 James Bond film, The Spy Who Loved Me.
Still, exhaustive epidemiological studies found no increased incidence of cancer, and even in the process of cleaning up the site or dealing with the accident itself, very few workers were even exposed to radiation. Highly sophisticated radiation sensors used in dozens of locations near the plant and by researchers in the region found more radiation indicated from atmospheric nuclear weapons testing than from the Three Mile Island disaster. To this day, the only death attributable to TMI in its entire history is a contractor who was killed in a jobsite safety failure while unloading a truck in 2021.
The announcement that Constellation (formerly associated with my former employer Exelon, love you guys) will restart the reactor is welcomed by many folks in the pro-nuclear crowd and many folks in the pro-decarbonization crowd (unless you’re more on the side of “we must stop consuming everything,” in which case you detest the idea of any power plants being built anywhere of any variety). I am not opposed to nuclear power, by any means, but I am thoroughly skeptical of the idea of using it to power artificial intelligence and data centers run by giant, unaccountable tech companies that have proven to be extremely inefficient with their resources. Google and Microsoft have seen massive spikes in energy demand in the past year owing to things like the fact that every Google search now is unavoidably tied to their silly AI feature. Even a few months ago, it was possible to turn it off.
The technology works. It’s not perfect. But in a society that cannot be bothered to reduce how much we consume, we must figure out less catastrophic ways to consume. I say this not boldly, but tongue-in-cheek. Clearly, the answer is not to build a thousand nuclear power plants. The answer is to use a combination of market mechanisms and better technology to rein in demand in a way that is more equitable for consumers and more environmentally beneficial.
The big problem with restarting a nuclear power plant to sell cheap power to giant tech companies is that they’re getting a great deal on that power and it doesn’t translate to broader grid decarbonization during an era when this must be a vital goal for every sector of the economy. In other words, it doesn’t matter too much that this is producing low-carbon electricity given that demand for electricity has spiked in the past couple of years owing to the proliferation of data centers and artificial intelligence. Accordingly, data centers, whose operators prioritize uptime at any and all costs, are effectively designed in a way that results in twice the infrastructure and twice the power consumption.
Public Policy to Fix It
There are ways to shift the onus of decarbonization from residential ratepayers (n.b., putting on my best Gretchen Whitmer accent here: “hard-working tyaxpayers”) onto the folks who generate the most pollution– by equalizing the rates that we pay.
Better Ratemaking. To the uninitiated, industrial and commercial consumers pay a far lower rate for electricity because they use so much more of it. Some of this makes sense– wholesale purchasing is always cheaper. They’re not only using much more power, they’re also typically paying for in larger “chunks.” If a utility knows that a steel mill is going to be operating an electric arc furnace on continuous double-shifts for five years straight at an estimated number of however many thousands of kilowatt-hours at a time, that’s reasonably straightforward as far as corporate accounting and capacity planning are concerned. Utilities do not find the same solace in thinking about homeowners who might upgrade from incandescent bulbs to compact fluorescents to LEDs to moon-bulbs, or what have you, nor in the prospects of residential customers installing rooftop solar and depriving them of that revenue stream.
This is the world we live in, but the advent of distributed energy resources is likely to slowly but steadily destabilize the monopolistic configuration of most public utilities. I imagine that less attention will be paid to the utility models of old, in which monumental, centralized power plants run the show and huge, high-voltage transmission carries the electrons to and fro, but rather more attention to local, more resilient grids, with more local generation resources.
Things that can be implemented at the state level of utility regulation:
- Incentives for small-scale distributed generation, paired with higher rates for AI and data centers. These companies have enough money and profit and making them pay their fair share will perhaps impinge upon their ability to stockpile hundreds of billions in cash– it will not drive up costs for the lay consumer.
- Exploration of new generation technology including nuclear power, ranging from large-scale to small-scale operations.
- Better regulation of data centers and AI technology in general.
But changing rates in a way that would force the biggest consumers to pay more is unlikely to get much traction. Companies that consume a lot of power have deep pockets for lobbying. There is one bright ray of sunshine coming through the clouds, though, and that’s that Americans have increasingly soured on the amount of control and power Big Tech has. The partisan split, in which Republicans seem to have a substantially higher distaste for these companies, is largely attributable to the perception that social media censors conservatives, even when we have oodles of evidence that social media in fact glorifies conservative voices because Russian disinformation actually sells quite well. Also, Cambridge Analytica? The Handbuilt City used to enjoy a great deal of engagement and referral traffic from Twitter, now X, which has become a cesspool of neo-Nazis and Elon Musk making r*pe jokes about Taylor Swift.
Anyway, we should embrace the downright miracle of any bipartisan agreement in the year of our lord 2024. Suffice it to say that the possibility of reining in Big Tech’s excesses by making power purchasing more equitable and sustainable is probably the last thing I would have had on my bingo card had I made up such a card years ago, but then again, my predictions aren’t always accurate. One thing I think I can predict, though, is that things are gonna get worse on the AI energy front before they get better. Is this a step in the right direction? It’s certainly a step in a direction.
