Image Source: Unsplash

With the supposed need for vast new electricity generation to fuel the artificial intelligence (AI) boom, AI companies are pushing nuclear power as one solution to provide that power for the many data centers they plan to build.

Count me skeptical of the boom and therefore of the need for vast new electricity generation capacity. See here, here, here, here and here. AI boosters usually talk about expanding existing nuclear power technologies, that is, fission reactors that run on uranium and (more dangerously) on plutonium.

But it is well to keep in mind that there are two kinds of nuclear power: fission and fusion. For now, there are no commercial fusion reactors since with current technology it takes far more than the equivalent of a kilowatt of energy to produce a kilowatt of electricity. This is because it takes a lot of energy just to get a fusion reaction going. The current state of affairs in fusion reminds me of the old joke about the manufacturer who admits he loses a nickel on every sale, but claims he makes it up in volume.

Fortunately, fusion researchers are smarter than this and await the day when fusion technology can produce more energy than it consumes. That waiting has spawned another well-worn joke about the coming of clean, limitless fusion energy, namely, that it's only 25 years away and always will be (whether fusion energy will be clean, that is, non-radioactive, is debatable).

It's no surprise then that with the AI industry saying it needs a lot more energy now, the predicted advent of net-energy-positive fusion is being moved up. In this case Commonwealth Fusion Systems, a startup spun off by the Massachusetts Institute of Technology, claims that by 2027 it will achieve the feat of producing more energy from a fusion device than is consumed. The Chinese government is a bit more vague, saying its research program may within a few years produce more energy than is consumed by a fusion reaction.

When this achievement is announced, it will be important to read the fine print. 11 years ago, scientists working on fusion at the Lawrence Livermore National Laboratory in California were able to produce more energy output in a fusion experiment than was used to produce the fuel. That feat, however, didn't take into account the amount of energy needed by the entire system, which was 118 times more than the energy output.

Some media outlets (who apparently did not read or understand the background materials) erroneously reported that the experiment had, in fact, achieved the feat of producing more energy than it consumed.

In 2022, the same laboratory declared it had achieved a net energy gain (read the second subheading) from a fusion reaction. Again, reading the fine print is important. As this article points out, "while a single shot may produce more energy than the fuel absorbs, the entire facility, from lasers to cryogenics to control systems, still consumes far more power than it delivers."

Said simply, you have to look at the whole system to understand the energy balance. This analysis suggests that the entire system actually consumed about 100 times the energy output of the experiment. The experiment did mark progress. But we remain nowhere near producing net energy from fusion reactions, not least because there is currently no system that can provide more than a momentary burst of energy instead of the sustained reaction seen in conventional fission reactors.

The Chinese government said it expects to have a pilot fusion plant operating by the 2030s or 2040s. First, that's pretty far away (and vague), and the realization of commercial fusion power is much further away, even if this plan comes to fruition. A pilot plant is only the second stage of the development of commercial fusion power. First comes the prototype which helps validate the technology. Then comes the pilot plant which demonstrates that such technology will, in fact, integrate successfully with the existing electric grid.

Then comes a demonstration plant which is a full-size test of the economic and commercial viability of the technology. At this stage, utility managers want hard evidence that such plants are reliable and profitable. Demonstration plants could be as far off as the 2050s or 2060s, again, even if we assume the schedule for pilot plants proves to be doable. And then, utilities would have to decide to try to build their own fusion plants and that might only begin in the late 2050s. Widespread adoption might take another 20 to 30 years.

Even if fusion generating plants turn out to be feasible, the idea that they are going to provide any near-term fix for our energy needs or for addressing climate change is completely misguided.

Energy transitions take time. They occur over more than one generation. In times of great stress such as ours, people look for miraculous solutions. Fusion seems like one of those solutions. But it will almost certainly not turn out to be miraculous and, if feasible, will be painstakingly slow to emerge as a major energy source for human civilization.

More By This Author: