Source: bemissu/Buildings/Landmarks, Backgrounds/Textures/shutterstock.
The current AI arms race has reached a bottleneck that can be relieved only by new sources of electrical generation. This scramble for power has garnered interest from hyperscale companies in previously resisted forms of energy, such as nuclear. While fission has been the exemplar of nuclear energy to date, the fusion sector has seen a recent surge in interest, with global funding exceeding $7.1 billion across 50 startups.
The Take
Our market forecast reports show that global datacenter power demand will likely maintain a compound annual growth rate of 14% to 2029, with the US growing by 18% in the same time period. Alongside the current demand for power, governments around the world continue to express the importance of energy transition, working to remove greenhouse-gas-intensive generation sources in favor of clean and renewable ones. In the wake of these and other considerations, conventional nuclear (fission) has accumulated considerable hype as of late for energy efficiency and the ability to provide 24/7 carbon-free power. As such, hyperscale companies have funneled unprecedented funds into these nuclear solutions as one of many options to ensure future energy availability for their operations.
Context
Meta Platforms Inc. signed a 20-year agreement with Constellation Energy Corp. in June to procure 1,121 MW of “emissions-free nuclear energy” from the Clinton Power Station nuclear plant. In 2024, Amazon Web Services invested $650 million for a 1,200-acre site next to the Susquehanna nuclear facility for datacenter development. The partnership outlines an initial 300 MW of behind-the-meter power, with periodic increases thereafter. Microsoft Corp. signed a contract to restart Constellation Energy’s facility at Three Mile Island, an estimated $1.6 billion project. Google’s partnership with Kairos Power aims for initial deployment in 2030 with a total of 500 MW by 2035. Amazon.com Inc. has also made a $500 million investment in the X-energy reactor company, with plans to bring 5 GW of power online by 2039.
The aforementioned projects all involve nuclear fission. There is, however, a second approach to nuclear energy known as fusion, and while the outlook is distant, fusion reactions are touted as a more efficient and sustainable form of energy generation. While fission has been the exemplar of nuclear energy, the fusion sector has seen a recent surge in interest, with global funding exceeding $7.1 billion across 50 startups. Google and Microsoft have shown a particular interest in fusion through significant investments and partnerships with fusion companies Commonwealth Fusion Systems, TAE Technologies and Helion.
Fission reactors, however, have their own drawbacks, such as highly toxic waste and a currently energy-dependent nature for most nations. The hope of fusion, and the “why” behind heavy investments into the approach by hyperscalers, is that it looks to greatly reduce — or even completely eliminate — these drawbacks. Fusion companies offer a variety of approaches, all aimed at generating the most clean and efficient source of energy to date. That said, the joke made about fusion is that it is “always a decade away,” so the prospect of now having go-live dates on the calendar makes the space all the more interesting to watch.
What is an AI Datacenter?
Before outlining fusion technologies, we must first place it in context with the more common fission reactions. Simply put, fission is a reaction where a neutron is fired at the nucleus of an atom, splitting it into at least two, smaller nuclei. While there are various fission methodologies, Uranium-235 is the most common fuel source for fission reactors; i.e., the atom that is fissioned. Nuclear is often marketed today as a clean source of energy but it comes with a unique set of issues. The waste created by fission is less abundant than that of fossil fuels, but it is difficult to draw parallels because it is of a different class. Fission reactors create what is classed as medium- to high-level radioactive waste, which is thermally hot, radioactive and must be shielded for storage. While this varies by element, high-level wastes can take thousands of years to decay to safe levels of radiation. A second issue, perhaps more pertinent for those concerned with relatively immediate gratification, is that 95.4% of US uranium must be imported from a small number of countries. While the US government is looking to ramp up uranium production, as it stands, an increased reliance on nuclear energy will likely march in lockstep with increased dependence on uranium-producing nations.
Fusion, on the other hand, is when two nuclei are fused into a single heavier nucleus. These reactions require immense heat and pressure, akin to stars, where fusion occurs naturally. The most common isotopes studied for fusion reactors are variants of hydrogen gas known as deuterium and tritium, which have a comparatively low ignition temperature of 160 million degrees Celsius. Because the heat produced in fusion reactors is well beyond the melting point of any manufacturing material, the superhot plasma must be confined using a magnetic cage, barring it from touching the walls of the container. The most popular design is known as a tokamak or a doughnut-shaped magnetic container. Tokamak designs yield a high degree of control but suffer from a sizable and complex design, making it more challenging for potential commercialization. A second method gaining attention from a few companies is known as a field-reverse configuration. This produces a self-stable torus of plasma, often said to resemble a hollowed-out sausage. One benefit of the field-reverse-style reactor is a smaller physical footprint and less intricate design, potentially facilitating commercialization.
Although a variety of isotopes can be fused to create energy, the following discussion will focus on those relevant to the players in the fusion space. As mentioned, the most well-studied method is deuterium-tritium fusion. While promising in terms of understanding and lower temperatures, it comes with a few issues. One is deuterium-tritium, which emits relatively high amounts of radioactive neutrons (80% of energy) that will melt and degrade the mechanical equipment around it, meaning the expensive and complex machinery will need replacement. The second is the scarcity of tritium. It is a rare isotope that must be “bred” by firing neutrons into a lithium blanket, which is often built into the reactor itself so the neutrons being emitted then create more tritium. While this method is backed by the most research, one criticism levied against it is that it is a better method for neutron generation than power generation.
There are also aneutronic or mostly aneutronic methods that emit low levels of neutrons. The first is deuterium and helium-3, which is thought to be a cleaner alternative that emits significantly fewer neutrons. Obversely, it has a higher temperature requirement of 200 million degrees Celsius and potentially undesired deuterium-deuterium and deuterium-tritium fusion do co-occur, which is not aneutronic and will reduce energy output. Helium-3 is also rare and will be expensive to produce; one method of production is the neutronic deuterium-deuterium fusion. The final method for our purposes is one that uses the abundant fuel sources hydrogen and boron. Beyond fuel abundance, this method also in theory boasts the least neutron release (<0.5 %). The downside of hydrogen-boron fusion is extremely high temperatures, upward of 1 billion degrees Celsius, significantly higher than what has been achieved by existing reactors. In terms of waste, fusion produces low-level radioactive waste, which has significantly shorter-lived radioactivity. The more aneutronic the fusion method, the less waste produced.
For transparency’s sake, it must be made clear that the longest stable fusion reaction is just over 22 minutes and produced negative energy. Only one lab has claimed a case of positive energy generation or Q>1, an efficiency metric with >1 indicating power generation greater than what it takes to run the reactor, also known as breakeven. This claim was made by the National Ignition Facility at the Lawrence Livermore National Laboratory, one of the US Department of Energy’s research labs under the National Nuclear Security Administration; it is worth noting that this claim is contested in the industry. Regardless, there are currently no examples of fusion power generation via any method at appreciable scale.
Want insights on datacenter trends delivered to your inbox? Join the 451 Alliance.
