Big tech companies are driving a nuclear resurgence, bringing shuttered facilities like 3 Mile Island back online and announcing plans to build small modular reactors. The new data centres being built today require so much electricity that national power grids can't keep up with them. Tech firms have worked out that they need to source their own power if they're going to build the kind of data centres that will make them competitive in AI.
The amount of power generated globally by nuclear power plants peaked in 2006, with new plant construction stagnating in the 1990s. The 2011 Fukushima disaster in Japan led to cancelled projects around the world. In Germany, Angela Merkel, a trained physicist, announced that Germany would phase out nuclear power, and Siemens announced that they would withdraw entirely from the nuclear industry to focus on renewables.
For a while, it seemed that Iran and North Korea were the only countries with a suspicious interest in building new nuclear power stations. In most of the rest of the world, projects were delayed or cancelled due to public safety concerns, regulatory hurdles, financing obstacles and fuel bottlenecks. It's not just the fear of disaster that slowed down nuclear energy, however. The global demand for electricity grew by just two point 2% in 2023, which was slightly less than the 2.4%
growth seen the prior year. While there had been strong growth in China, India, and the developing world, advanced economy showed substantial declines in their need for new power stations. For much of the 20th century, America's electricity demand had grown steadily and utilities built power stations to keep up with the growing demand.
Starting in the mid 2000's, the demand for additional capacity in the developed world flattened out while the US economy and population continued to grow and while we continued to add new electrical devices to our homes, factories, light bulbs, TV's, computers and home appliances all became more and more energy efficient, meaning that there was no real need to build new
generation capacity. The growth of data centers over the last few years, driven by working from home, cryptocurrency mining and now AI, boosted demand for electricity. As a side note, working from home which increased the need for data centres also reduced the amount of power being used overall as the server racks at data centres are more energy efficient than the office based servers which businesses had relied on in the past.
Other changes however, like the roll out of electric vehicles in recent years and plans to reassure the semiconductor industry, further boost the demand for electricity in advanced economies. Big tech companies like Microsoft, Amazon and Google are driving the nuclear revival as to get new power hungry data centres built in the United States and Europe, they have to solve the problem of power
generation. Their net 0 pledges mean that the sources of power that they've pledged to use have to be low carbon, and they've already invested heavily in wind and solar, but their data centres still need a steady base load. So for big Tech, investing in nuclear energy makes a lot of sense.
If we look at the pledges that big tech firms have made, almost all of them have promised to reach net 0 carbon emissions by 2040. These pledges often involve transitioning to renewable energy sources, investing in carbon capture technology, and optimizing their operations and supply chains for reduced energy consumption. Their pledges are somewhat believable, as they they're so profitable that they can actually afford to do this, unlike smaller, less profitable businesses.
At the most extreme end of the scale, Microsoft pledged almost five years ago to be carbon negative by 20-30, additionally promising that by 2050 they will have entirely stripped all of the carbon they emitted over the life of the firm from the
environment. As difficult as that goal will have sounded in 2020, it's only gotten more difficult over time thanks to AI. Microsoft now says that they expect to consume five to six times more electricity than they had planned on when they first made that promise, but they are still sticking to their goal. The reason for the huge step up in energy consumption is that AI consumes an awful lot of power. A ChatGPT query, for example, uses.
This is about 10 times as much power to process as a simple Google search does, and generating a single AI image consumes the same amount of energy as fully charging a smartphone. Microsoft revealed earlier this year that it's carbon emissions have increased by 30% since 2020, so they're well off target. The data centres being planned or built today consume more power than can be bought from utilities, and so if Big Tech wants a lot of 0 carbon power, their only realistic option is
nuclear power. Three Mile Island in Pennsylvania was the scene of the worst commercial nuclear accident in U.S. history, when a cooling malfunction in 1979 caused part of the core to melt in reactor 2, releasing radioactive gases and radioactive iodine into the environment. Schools near the reactor were quickly closed, pregnant women and small children were evacuated from the area, and residents were urged to stay indoors. Within a few days, 140,000
people left the area. In the aftermath, an investigation found that the average radiation dose to people living within 10 miles of the plant was about equal to that of a chest X-ray. Further studies found that the accident had no observable long term health effects on those living near the plant. Nevertheless, it was a big scare. The Three Mile Island accident was a turning point in the growth of of nuclear power.
Prior to the incident, the number of reactors under construction globally had grown steadily almost every year, but following the event, 52 plant nuclear reactors were cancelled in the United States alone. At the time of the incident, 129 nuclear power plants had been approved for construction worldwide, but of those only 53 were completed.
The Chernobyl disaster in northern Ukraine seven years later had an even bigger impact on public and political attitudes towards the safety of nuclear power and ended the growth in active reactors globally. Last month it was announced that Constellation Energy will reopen a nuclear plant, a Three Mile Island, to provide power to
Microsoft's data centres. The plant being reopened is Unit 1, which was just decommissioned five years ago, and Unit 2, where the accident occurred in 1979, will still remain closed. The recommission plant at Three Mile Island will provide more than 800 megawatts of power, all of which has been bought by Microsoft from Constellation Energy under a 20 year power purchase agreement. The plant is expected to be brought online in 2028 and to run until at least 2054.
The location of the data centres that will use the electricity has not yet been announced. Constellation says that the deal will create more than 3000 jobs, add $16 billion to Pennsylvania GDP, and bring in at least $3 billion in state and federal taxes while providing safe, reliable, carbon free electricity that helps reduce emissions.
Not to be outdone, Amazon last week announced that they would be spending more than $500 million to support development of four next generation small modular reactors. They took a stake in X Energy and SMR Developer Google additionally agreed to buy 500 megawatts of power from six to seven small modular reactors, which will be built by Cairo's Power. None of these companies revealed the financial details of the agreements or where in the US
the plants will be built. The idea behind small modular reactors, known as SMRS, is that the components of each unit can be small enough to be mass produced on an assembly line, making them cheaper to manufacture than full size reactors. Each power plant could start with just one reactor and add more as needed. SMRS can, in theory, be cited in locations not suitable for
larger nuclear power plants. the US Department of Energy defines SMRS as advanced nuclear reactors with a power capacity of up to 300 megawatts per unit, which is about 1/3 of the generating capacity of traditional nuclear power reactors. Rolls Royce's SMR design generates 470 megawatts, which is actually bigger than many of the 1st generation nuclear reactors that were built in the UK. According to Goldman Sachs data center, power demands will grow 160% between now and 2030.
They say that when this is combined with the electrification of transport and the reassuring of manufacturing, electricity demand can be expected to double. Since 2021, American companies have announced plans to spend more than $500 billion on factories for semiconductors, batteries, solar panels and more. As great as this sounds, there's not sufficient generating capacity to power all of these factories today.
The climate author Solomon Goldstein Rose argued in a 2001 Ted talk that while the world is on track to replace all fossil fuel power plants by 2050, that this would not be good enough as we're likely to need five times more power in 2050 as we use today if the world is actually switching to electric vehicles, Electric heating and electrifying various industrial processes like steel making and concrete manufacturing.
He points out that over 700 million people today have no access to electricity worldwide, and that billions more have access to only small amounts of intermittent power, and that providing these people access to energy would lift them out of poverty. His estimate of needing five times as much power came before these AI data centres were being built. Nuclear power is well suited to data centres because it provides a reliable source of power around the clock, unlike wind and solar.
There are a few excellent YouTube videos by Practical Engineering explaining the importance of a base load on power grids and why it's much harder than you would expect to connect solar and wind power to our existing electrical grids. I was surprised to learn that many grid tight solar systems shut down whenever the grid goes down, which is not really what
you would want. This happens both for the safety of utility workers and because many of the inverters are grid following inverters which need to sense the voltage of the grid so that they can synchronize their output to match the grids frequency and phase. A more obvious barrier to a 100% wind and solar grid is the intermittency of these power sources. The wind doesn't always blow and
the sun doesn't always shine. On top of that, the windiest and sunniest places are not all that close to most major population centers. Grid scale battery backup has been put forth as a solution to this problem, but it's by no means a perfect solution either, as rapid cycling degrades the lifespan of utility scale batteries.
A 2021 wind drought in Europe meant that windmills generated significantly less electricity than was expected, sending power prices to record highs as utilities had to scramble to find alternative fossil fuel power sources. This map shows the potential for solar power in different parts of the world, and you'll notice that Canada and Western Europe are particularly I'll suited to solar generation.
British businesses, as an example, already face the highest energy costs in the developed world, and if the new government is hoping to boost growth, dealing with the weak energy infrastructure might be a good start. Unfortunately, Britain is not only I'll suited for solar and wind generation, it's also the most expensive place in the world to build a new nuclear power station.
According to Sam Dimitro Hinckley, Point C, which is currently being built, will cost six times more to build per MW than it's South Korean equivalent. France and Finland, which used the same reactor design, managed to build them for about 50% less per MW. South Korea have kept their costs lower by building fleets of reactors benefiting from economies of scale. The same reactors is being built in the UK, is cheaper to build in Finland and France because of a simpler approvals process.
UK rate regulations require a more complex planning process and requested 7000 changes to the original design, essentially making it entirely bespoke. One of the alterations required to make it through the planning process involves spending 10s of millions of pounds on underwater loud speakers to scare fish away from the cooling intake. This fish disco is not required in other countries. Even outside the UK, nuclear power can be extremely expensive and time consuming to build.
It's much more expensive than wind or solar. And while the climate case for nuclear power is strong, the challenge of safely storing nuclear waste has not been entirely resolved. There is, of course, plenty of controversy over whether nuclear power can be considered clean energy or not, mostly driven by concerns around nuclear waste, nuclear weapon proliferation, and the risk of accidents. A number of studies show that coal power plants output more radiation than nuclear power plants do.
The fly ash emitted when burning coal for electricity carries 100 times more radiation than a nuclear power plant producing the same amount of power. According to the US Office of Nuclear Energy, the use of nuclear power in the United States has reduced CO2 emissions, equivalent to removing 100 million cars from the road, and nuclear power plants produce more electricity on less land than any other
clean energy source. They say that all of the used nuclear fuel produced by the US nuclear energy industry could fit on an American football field at a depth of less than 3 yards. Because this video is about AI and energy, I thought it might make sense to use Microsoft Copilot to convert those measurements into something more understandable for an international audience.
And I'm not actually joking. It came back with the reply that this volume is equivalent to 18 Olympic sized swimming pools filled up with Tesla's. So yeah, it's a big box, I guess. The US Office of Nuclear Energy goes on to say in their report that a lot of that waste could be reprocessed and recycled and that many of the new reactor designs can operate on used fuel.
Now there are not enough decommissioned nuclear reactors like 3 Mile Island for Big Tech to resurrect, but they have shown a growing interest in using small modular reactors. The problem with that is that there are only three operational SM Rs in the world today, one in China, one in Russia, and 1:00 in India. Three more are under construction and 65 are still in the design stage.
GE, Vernova, Rolls Royce and Westinghouse all have SMR designs, as do a long list of startups like the Bill Gates backed Terror Power, which broke ground on its first plant in Wyoming this summer, X Energy, which is backed by Amazon, and Oaklow, which counts Sam Altman of Open AI among its backers. Bill Gates has invested more than a billion dollars in Terror Power, which is working to develop SMRS in partnership with Warren Buffett's utility company, Pacificorp.
France gets around 70% of its electricity from nuclear energy, which is the highest percentage in the world. But the United States generates the most nuclear power overall, mostly through plants that began construction in the 1960s and 70s. There are more SMRS being planned in the United States today than anywhere else, largely because of the tech industry. While promising, the technology behind SMRS is still unproven at
scale. And the recent Wall Street Journal article asked what customer would pay the higher price for a first of kind project and who would order enough to get an assembly line started. The answer to that question today is quite obvious.
Big tech. And while power plants can be thought of as boring infrastructure which might be difficult to finance, the fact that these data centres are needed for AI, which venture capital and retail investors are very excited about, means that AI companies can raise loads of cheap money from VC funds and pump it into building not only the data centres that they need, but also the nuclear power to run them on. As without sourcing your own
power, you might might be out of luck waiting for grid upgrades. Last month, Microsoft teamed up with BlackRock to launch a $30 billion fund to invest in artificial intelligence infrastructure. They told the press that the partnership would mobilize up to $100 billion in total investment potential once debt financing was included, and that the money would be invested to enhance AI, supply chains and energy
sourcing. Microsoft said that they expect this trend to catch on with other tech firms and said that Microsoft might participate in all of them. Governments, after years of ignoring it, are now getting excited about nuclear power too. During UN Climate Week, officials reaffirmed their commitment to triple the global output of nuclear power by 2050, and at the same event, 14 large banks said that they were willing to finance these
projects. Joe Biden recently signed a law with bipartisan support, which is almost unheard of in the United States today, to hasten the development of nuclear energy projects. Historically, many Democrats opposed new nuclear plants, citing safety, environmental and economic concerns. Today, they see it as being critical to their goals of reducing greenhouse gas emissions and meeting the needs of a growing economy.
The Department of Energy Loan Programs Office has a $600 billion allocation of funds for renewables and advanced manufacturing, and this can be used to fund new nuclear projects too. As an example of how much funding there seems to be, Bloomberg reported this summer that Sam Altman of Open AI was staking a sizeable chunk of his personal wealth on a nuclear fusion start up, which he says
is key to an AI enabled future. Nuclear fusion attempts to create energy not by splitting atoms, but by smashing them together, which requires extreme temperatures and lots of power, but hypothetically could output with more energy than goes into it without creating the radioactive waste associated with today's nuclear fission reactors.
While currently unproven as a workable option for power generation, nuclear fusion advocates claim that if successful, fusion has the potential to be a sustainable and abundant energy source. Skeptics, on the other hand, say that fusion is worse than Tesla robotaxis, as it's always just 20, 20 years away. Most experts say that we're unlikely to be able to generate large scale energy from nuclear
fusion anytime soon. But Microsoft still signed a power purchase agreement with Altman's company Helion. Earlier this year. Bloomberg reported that this deal cut some Helion employees by surprise, and they expressed doubts about the company's promise to get a prototype
running by year end. Like most VC backed start-ups, Elian seems to be aiming to move fast and break things, but is unfortunately being slowed down not just by the science, but by numerous complaints of harassment and gender discrimination. The fusion industry, which as I said is not hugely promising in the short term, has still managed to attract more than $7 billion in venture capital
investment. A nuclear power resurgence may not be essential for AI to get off the ground, but when combined with the climate commitments big tech firms have made, along with the power demands of electrified transportation and reassuring of manufacturing, it does seem to be happening. From the perspective of big tech firms, nuclear power success would help calm the climate driven fears about the extreme power demands of generative AI.
And after that it'll be up to Big Tech to develop the AGI, which they for some strange reason are telling investors will kill us all. Which I have to say, I'm firmly against. Thanks for tuning into this week's podcast, with a special thanks to my supporters on Patreon who make this all happen. If you want to support the show, I'll put a link in the show notes. Have a great week and talk to you again soon. Bye.