AI's Power Hunger Drives Tech Giants Towards Nuclear Energy

AI is consuming more power than the grid can handle -- nuclear might be the answer

New partnerships are forming between tech companies and power operators -- ones that could reshape decades of misconceptions about nuclear energy. Last year, Meta (Facebook's parent company) put out a call for nuclear proposals, Google agreed to buy new nuclear reactors from Kairos Power, Amazon partnered with Energy Northwest and Dominion Energy to develop nuclear energy and Microsoft committed to a 20-year deal to restart Unit 1 of the Three Mile Island nuclear plant. At the centre of these partnerships is artificial intelligence's voracious appetite for electricity. One Google search uses about as much electricity as turning on a household light for 17 seconds. Asking a Generative AI model like ChatGPT a single question is equivalent to leaving that light on for 20 minutes. Read more: AI is bad for the environment, and the problem is bigger than energy consumption Having GenAI generate an image can draw about 6,250 times more electricity, roughly the energy of fully charging a smartphone, or enough to keep the same light bulb on for 87 consecutive days. The hundreds of millions of people now using AI have effectively added the equivalent of millions of new homes to the power grid. And demand is only growing. The challenge for tech companies is that few sources of electricity are well-suited to AI. The grid wasn't ready for AI AI requires vast amounts of computational power running around the clock, often housed in energy-intensive data centres. Renewable energy sources such as solar and wind provide intermittent energy, meaning they don't guarantee the constant power supply these data centres require. These centres must be online 24/7, even when the sun isn't shining and the wind isn't blowing. Fossil fuels can run continuously, but they carry their own risks. They have significant environmental impacts. Fuel prices can be unpredictable, as exemplified by the gas price spikes due to the war in Ukraine, and the long-term availability of fossil fuels is uncertain. Major tech companies like Google, Amazon and Microsoft say they are committed to eliminating CO2 emissions, making fossil fuels a poor long-term fit for them. This has pushed nuclear energy back into the conversation. Nuclear energy is a good fit because it provides electricity around the clock, maximizing the use of expensive data centres. It's also clean, allowing tech companies to meet their low CO2 commitments. Lastly, nuclear energy has very low fuel costs, which allows tech companies to plan their costs far into the future. However, nuclear energy has its own set of problems that have historically been hard to solve -- problems that tech companies may now be uniquely positioned to overcome. Is nuclear energy making a comeback? Nuclear power has long been considered too costly and too slow to build. The estimated cost of a 1.1 gigawatt nuclear power facility is about US$7.77 billion, but can run higher. The recently completed Vogtle Units 3 and 4 in the state of Georgia, for example, cost US$36.8 billion combined. Historically, nuclear energy projects have been hard to justify because of their high upfront costs. Like solar and wind power, nuclear energy has relatively low operating costs once a plant is up and running. The key difference is scale: unlike solar panels, which can be installed on individual rooftops, the kind of nuclear reactors tech companies require can't be built small. Yet this cost is now more palatable when compared to the expense of AI data centres, which are both more costly and entirely useless without electricity. The first phase of OpenAI and SoftBank's Stargate AI project will cost US$100 billion and could be entirely powered by a single nuclear plant. Nuclear power plants also take a long time to build. A 1.1 gigawatt reactor takes, on average, 7.5 years in the U.S. and 6.3 years globally. Projects with such long timelines require confidence in long-term electricity demand, something traditional utilities struggle to predict. To solve the problem of long-range forecasting, tech companies are incentivizing power providers by guaranteeing they'll purchase electricity far into the future. These companies are also literally and financially moving closer to nuclear power, either by acquiring nuclear energy companies or locating their data centres next to nuclear power plants. Destigmatizing nuclear energy One of the biggest challenges facing nuclear energy is the perception that it's dangerous and dirty. Per gigawatt-hour of electricity, nuclear produces only six tonnes of CO2. In comparison, coal produces 970, natural gas 720 and hydropower 24. Nuclear even has lower emissions than wind and solar, which produce 11 and 53 tonnes of CO2, respectively. Nuclear energy is also among the safest energy sources. Per gigawatt-hour, it causes 820 times fewer deaths than coal, 43 times fewer than hydropower and roughly the same as wind and solar. Still, nuclear energy remains stigmatized, largely because of persistent misconceptions and outdated beliefs about nuclear waste and disasters. For instance, while many public concerns remain about nuclear waste, existing storage solutions have been used safely for decades and are supported by a strong track record and scientific consensus. Similarly, while the Fukushima disaster in Japan displaced thousands of people and was extremely costly (total costs of the disaster are expected at about US$188 billion), not a single person died of radiation exposure after the accident, a United Nations Scientific Committee of 80 international experts found. Read more: With nuclear power on the rise, reducing conspiracies and increasing public education is key For decades, there was little effort to correct public perceptions about nuclear fears because it wasn't seen as necessary or profitable. Coal, gas and renewables were sufficient to meet the demand required of them. But that's now changing. With AI's energy needs soaring, Big Tech has classified nuclear energy as green and the World Bank has agreed to lift its longstanding ban on financing nuclear projects. Big Tech's billion-dollar bet on nuclear The world has long lived with two nuclear dilemmas. The first is that, despite being one the safest and cleanest form of energy, nuclear was perceived as one the most dangerous and dirtiest. The second is that upgrading the power grid requires large-scale investments, yet money had been funnelled into small, distributed sources like solar and wind, or dirty ones like coal and natural gas. Now tech companies are making hundred-billion-dollar strategic bets that they can solve both nuclear dilemmas. They are betting that nuclear can offer the kind of steady, clean power their AI ambitions require. This could be an unexpected positive consequence of AI: the revitalization of one of the safest and cleanest energy sources available to humankind. Michael Tadrous, an undergraduate student and research assistant at the DeGroote School of Business at McMaster University, co-authored this article.

AI is consuming more power than the grid can handle. Nuclear might be the answer

New partnerships are forming between tech companies and power operators -- ones that could reshape decades of misconceptions about nuclear energy. Last year, Meta (Facebook's parent company) put out a call for nuclear proposals, Google agreed to buy new nuclear reactors from Kairos Power, Amazon partnered with Energy Northwest and Dominion Energy to develop nuclear energy and Microsoft committed to a 20-year deal to restart Unit 1 of the Three Mile Island nuclear plant. At the center of these partnerships is artificial intelligence's voracious appetite for electricity. One Google search uses about as much electricity as turning on a household light for 17 seconds. Asking a Generative AI model like ChatGPT a single question is equivalent to leaving that light on for 20 minutes. Having GenAI generate an image can draw about 6,250 times more electricity, roughly the energy of fully charging a smartphone, or enough to keep the same light bulb on for 87 consecutive days. The hundreds of millions of people now using AI have effectively added the equivalent of millions of new homes to the power grid. And demand is only growing. The challenge for tech companies is that few sources of electricity are well-suited to AI. The grid wasn't ready for AI AI requires vast amounts of computational power running around the clock, often housed in energy-intensive data centers. Renewable energy sources such as solar and wind provide intermittent energy, meaning they don't guarantee the constant power supply these data centers require. These centers must be online 24/7, even when the sun isn't shining and the wind isn't blowing. Fossil fuels can run continuously, but they carry their own risks. They have significant environmental impacts. Fuel prices can be unpredictable, as exemplified by the gas price spikes due to the war in Ukraine, and the long-term availability of fossil fuels is uncertain. Major tech companies like Google, Amazon and Microsoft say they are committed to eliminating CO emissions, making fossil fuels a poor long-term fit for them. This has pushed nuclear energy back into the conversation. Nuclear energy is a good fit because it provides electricity around the clock, maximizing the use of expensive data centers. It's also clean, allowing tech companies to meet their low CO commitments. Lastly, nuclear energy has very low fuel costs, which allows tech companies to plan their costs far into the future. However, nuclear energy has its own set of problems that have historically been hard to solve -- problems that tech companies may now be uniquely positioned to overcome. Is nuclear energy making a comeback? Nuclear power has long been considered too costly and too slow to build. The estimated cost of a 1.1 gigawatt nuclear power facility is about US$7.77 billion, but can run higher. The recently completed Vogtle Units 3 and 4 in the state of Georgia, for example, cost US$36.8 billion combined. Historically, nuclear energy projects have been hard to justify because of their high upfront costs. Like solar and wind power, nuclear energy has relatively low operating costs once a plant is up and running. The key difference is scale: unlike solar panels, which can be installed on individual rooftops, the kind of nuclear reactors tech companies require can't be built small. Yet this cost is now more palatable when compared to the expense of AI data centers, which are both more costly and entirely useless without electricity. The first phase of OpenAI and SoftBank's Stargate AI project will cost US$100 billion and could be entirely powered by a single nuclear plant. Nuclear power plants also take a long time to build. A 1.1 gigawatt reactor takes, on average, 7.5 years in the U.S. and 6.3 years globally. Projects with such long timelines require confidence in long-term electricity demand, something traditional utilities struggle to predict. To solve the problem of long-range forecasting, tech companies are incentivizing power providers by guaranteeing they'll purchase electricity far into the future. These companies are also literally and financially moving closer to nuclear power, either by acquiring nuclear energy companies or locating their data centers next to nuclear power plants. Destigmatizing nuclear energy One of the biggest challenges facing nuclear energy is the perception that it's dangerous and dirty. Per gigawatt-hour of electricity, nuclear produces only six metric tons of CO. In comparison, coal produces 970, natural gas 720 and hydropower 24. Nuclear even has lower emissions than wind and solar, which produce 11 and 53 metric tons of CO, respectively. Nuclear energy is also among the safest energy sources. Per gigawatt-hour, it causes 820 times fewer deaths than coal, 43 times fewer than hydropower and roughly the same as wind and solar. Still, nuclear energy remains stigmatized, largely because of persistent misconceptions and outdated beliefs about nuclear waste and disasters. For instance, while many public concerns remain about nuclear waste, existing storage solutions have been used safely for decades and are supported by a strong track record and scientific consensus. Similarly, while the Fukushima disaster in Japan displaced thousands of people and was extremely costly (total costs of the disaster are expected at about US$188 billion), not a single person died of radiation exposure after the accident, a United Nations Scientific Committee of 80 international experts found. For decades, there was little effort to correct public perceptions about nuclear fears because it wasn't seen as necessary or profitable. Coal, gas and renewables were sufficient to meet the demand required of them. But that's now changing. With AI's energy needs soaring, Big Tech has classified nuclear energy as green and the World Bank has agreed to lift its longstanding ban on financing nuclear projects. Big Tech's billion-dollar bet on nuclear The world has long lived with two nuclear dilemmas. The first is that, despite being one of the safest and cleanest forms of energy, nuclear was perceived as one of the most dangerous and dirtiest. The second is that upgrading the power grid requires large-scale investments, yet money had been funneled into small, distributed sources like solar and wind, or dirty ones like coal and natural gas. Now tech companies are making hundred-billion-dollar strategic bets that they can solve both nuclear dilemmas. They are betting that nuclear can offer the kind of steady, clean power their AI ambitions require. This could be an unexpected positive consequence of AI: the revitalization of one of the safest and cleanest energy sources available to humankind.