When AI Goes Nuclear

Google is constructing small nuclear reactors to power its AI data centres.

It is only one example of a larger trend of tech companies turning to nuclear power to meet the voracious energy needs of AI.

The first reactor is expected to be operational by 2030, with more to follow by 2035.

Not for Google is the correct world of solar power or recyclable green energy.

When it comes to AI, let the nuclear turbines spin!

Google is partnering with start-up Kairos Power for the small modular reactors (SMRs). Its reactors will use molten fluoride salt instead of water for cooling. They will generate a total 500 megawatts of power. That’s equivalent to the power consumption of roughly 400,000 to 900,000 average-sized homes. Google says that the small nuclear reactors would be far easier and quicker to deploy than conventional reactors, and are "inherently" safe.

A solar farm capable of generating the equivalent 5 MW of power would cover 25 acres of land…about the same as 25 football fields.

According to Michael Terrell, Google’s senior director for energy and climate, this partnership sends an “important demand signal to the market.” By investing in multiple SMRs, Google aims to accelerate the commercialization of this technology.

A company spokesman said that “Google has played a pivotal role in accelerating clean energy solutions, including the next generation of advanced clean technologies."

Amazon and Microsoft are also turning to nukes. Microsoft is reviving the Three Mile Island plant - what could go wrong? - to supply power for its data centres, and Amazon is opening a nuclear-powered data centre adjacent to a plant in Pennsylvania.

Why do they need so much power?

Google stated that AI is the driver, already "powering major scientific advances, improving services for businesses and customers, and driving national competitiveness and economic growth.” It argues that nuclear power will "unlock [its] full potential for everyone."

AI does draw a lot of electric energy.

A single ChatGPT query uses around ten times as much power as a Google search query, while the AI industry as a whole is predicted to draw about as much electricity per year as the entire country of the Netherlands by 2027.

AI will need a five-fold increase in power over the coming three years, according to Morgan Stanley. Data centers may see their electricity consumption more than double by 2030, reaching 9% of total electricity demand in the US. By the end of the decade, AI data centers could consume as much as 20% to 25% of US power.

The human brain can power itself for six hours off the energy in a breakfast burrito.

Makes AI look pretty stodgy.

A concern at some level, is that energy generation used to be a public service (except, of course, in Texas). Now it is functionally becoming a private sector consumable.

At first glance, this next point jumps off the rails a bit, but in a parallel development, a rock band that doesn’t exist is taking the music world by storm.

Substack commentator Stephen McPherson draws our attention to the advance of AI in the music industry: “The music industry has rapidly been moving towards adoption of technology to eliminate human input. Drum machines were just the start. Eliminating humans means the record companies grab the entire pie. Some smart cookie, who has yet to be identified, used AI to create a rock band, and also all the supporting marketing baggage to go along with it; photos, bios, etc. He used AI to create hits based on music from the 1980 through to about 2010.

“The AI band has become one of the top streaming bands on Spotify, and it's just data. It doesn't exist anywhere, other than on the net. Millions of people listen to this music every day, thinking they are listening to a real band. The disturbing part is that all those listeners believed they were listening to a band of humans making music.”

The real person behind the invented musical artist ‘Andrew Frelon’ is an expert on web platform safety and policy issues, with extensive experience using generative AI. He was born in the United States but lives in Canada. He spotted a band online called the Velvet Sundown, and decided that it was generated by AI.

Indeed, it turns out that the band was a synthetic music project “guided by human creative direction, and composed, voiced, and visualized with the support of artificial intelligence."

To get to this point of usefulness, the AI needs to be “trained” – it has to absorb tons of data and make comparisons and store the results. This is the big drain - the big need for electric power.

So we have nuclear-powered generators churning out generative AI.

That will end well, as a commentator noted.

At the core, AI allows computers to learn and solve problems in ways that can seem human. Ideally, it automates tasks, optimizes decision-making, and fosters innovation. It allows machines to learn from experience.

McKinsey global institute says that by 2030, some 70% of companies will have embraced the AI revolution and adopted at least one type of AI technology. Forbes says that AI has the potential to be among the most disruptive technologies across global economies that we will ever develop.
Goldman Sachs says that AI could replace the equivalent of 300 million full-time jobs, and could replace a quarter of work tasks in the US and Europe -- but may also mean new jobs and a productivity boom.

On the front lines of human creativity, in the Art world, its impact is complex. While it offers new tools and methods of expression, it also delivers copy-cat and low-grade repetition of existing ideas.

I personally think that AI has a long way to go in terms of technological development before it is ready to be considered as a companion for creative thought.

But I have been wrong before.

Perhaps I should listen to the Velvet Sundown again. Another fully AI-generated band, The Devil Inside, has more than 1.7 million plays on Spotify. Currently, Spotify has no rules preventing the upload of AI-generated music, so listeners can be unaware they’re streaming content created by machines.

French music streaming platform Deezer reported that 18 per cent of the new songs uploaded to its service were fully AI-generated—about 20,000 songs a day. It Deezer became the first major platform to launch an AI detection tool that flags such content for users. Deezer CEO Alexis Lanternier said “AI is not inherently good or bad, but we believe a responsible and transparent approach is key to building trust with our users and the music industry.”

A wave of lawsuits is accompanying the trend, with musicians and songwriters alleging their material is being used without permission to train AI models.

As far as artistic content is concerned, it’s the Wild West out there for AI.

But people are listening.

And companies are building nuclear reactors to feed the AI brains.

I bet that wasn’t in the job descriptions five years ago.

Neither was the electrical demand for heat pumps and EVs, which are actually bigger shifts than AI.

It’s a changing world out there; the only certainty is that the need for electric power is voracious.

And no amount of coal-fired generators could possibly meet that demand.

Probably bad news for Trump’s attempt to keep us in the fossil fuel industry era, but they are so yesterday!

Maybe the Velvet Underground will do a song about them…

Thank you for following Barry’s Substack, focusing on the meaning behind the headlines. Each week a summary of a topical book helps full subscribers stay ahead of the conversation.

Next week will be NOT MY TYPE: One Woman v.s. a President…The story of E. Jean Carroll’s law suit against Donald Trump

Recent features have delved into:

PS - If you have enjoyed this article, please consider buying me a TACO!