The artificial intelligence boom has already changed how we understand technology and the world. But developing and updating AI programs requires a lot of computing power. This relies heavily on servers in data centres, at a great cost in terms of carbon emissions and resource use.
One particularly energy intensive task is “training”, where generative AI systems are exposed to vast amounts of data so that they improve at what they do.
The development of AI-based systems has been blamed for a 48% increase in Google’s greenhouse gas emissions over five years. This will make it harder for the tech giant to achieve its goal of reaching net zero by 2030.
Some in the industry justify the extra energy expenditure from AI by pointing to benefits the technology could have for environmental sustainability and climate action. Improving the efficiency of solar and wind power through predicting weather patterns, “smart” agriculture and more efficient, electric autonomous vehicles are among the purported benefits of AI for the Earth.
It’s against this background that tech companies have been looking to renewables and nuclear fission to supply electricity to their data centres.
Nuclear fission is the type of nuclear power that’s been in use around the world for decades. It releases energy by splitting a heavy chemical element to form lighter ones. Fission is one thing, but some in Silicon Valley feel a different technology will be needed to plug the gap: nuclear fusion.
Unlike fission, nuclear fusion produces energy by combining two light elements to make a heavier one. But fusion energy is an unproven solution to the sustainability challenge of AI. And the enthusiasm of tech CEOs for this technology as an AI energy supply risks sidelining the potential benefits for the planet.
Google recently announced that it had signed a deal to buy energy from small nuclear reactors. This is a technology, based on nuclear fission, that allows useful amounts of power to be produced from much smaller devices than the huge reactors in big nuclear power plants. Google plans to use these small reactors to generate the power needed for the rise in use of AI.
This year, Microsoft announced an agreement with the company Constellation Energy, which could pave the way to restart a reactor at Pennsylvania’s Three Mile Island nuclear power station, the site of the worst nuclear accident in US history.
However, nuclear power produces long-lived radioactive waste, which needs to be stored securely. Nuclear fuels, such as the element uranium (which needs to be mined), are finite, so the technology is not considered renewable. Renewable sources of energy, such as solar and wind power suffer from “intermittency”, meaning they do not consistently produce energy at all hours of the day.
These limitations have driven some to look to look to nuclear fusion as a solution. Most notably, Sam Altman of OpenAI has shown particular interest in Helion Energy, a fusion startup working on a relatively novel technological design.
In theory, nuclear fusion offers a “holy grail” energy source by generating a large output of energy from small quantities of fuel, with no greenhouse gas emissions from the process and comparatively little radioactive waste. Some forms of fusion rely on a fuel called deuterium, a form of hydrogen, which can be extracted from an abundant source: seawater.
In the eyes of its advocates, like Altman, these qualities make nuclear fusion well suited to meet the challenges of growing energy demand in the face of the climate crisis –- and to meet the vast demands of AI development.
However, dig beneath the surface and the picture isn’t so rosy. Despite the hopes of its proponents, fusion technologies have yet to produce sustained net energy output (more energy than is put in to run the reactor), let alone produce energy at the scale required to meet the growing demands of AI. Fusion will require many more technological developments before it can fulfil its promise of delivering power to the grid.
Wealthy and powerful people, such as the CEOs of giant technology companies, can strongly influence how new technology is developed. For example, there are many different technological ways to perform nuclear fusion. But the particular route to fusion that is useful for meeting the energy demands of AI might not be the one that’s ideal for meeting people’s general energy needs.
Innovators often take for granted that their work will produce ideal social outcomes. If fusion can be made to work at scale, it could make a valuable contribution to decarbonising our energy supplies as the world seeks to tackle the climate crisis.
However, the humanitarian promises of both fusion and AI often seem to be sidelined in favour of scientific innovation and progress. Indeed, when looking at those invested in these technologies, it is worth asking who actually benefits from them.
Will investment in fusion for AI purposes enable its wider take-up as a clean technology to replace polluting fossil fuels? Or will a vision for the technology propagated by powerful tech companies restrict its use for other purposes?
It can sometimes feel as if innovation is itself the goal, with much less consideration of the wider impact. This vision has echoes of Meta CEO Mark Zuckerberg’s motto of “move fast and break things”, where short-term losses are accepted in pursuit of a future vision that will later justify the means.
Sophie Cogan receives funding from the EPSRC Fusion Centre for Doctoral Training.
