Things are getting awkward for the techno utopians.
Their Artificial Intelligence golden child is upending their precious Net Zero energy transition. They may be forced to choose between the two.
What’s the trade off about? Electricity.
According to one estimate, AI’s data centres in the US will need as much electricity as the county’s electrified car fleet! We’re talking about adding a small country’s worth of demand to the grid.
This new demand is blowing all energy transition estimates and plans out of the water.
Total energy demand in the US has been unchanged for decades.
Energy transition plans rest on this presumption holding for decades more.
Hello!
The parts of the county hosting AI data centres are proving them wrong. A JP Morgan Asset Management report highlighted one example:
‘The PJM (mid-Atlantic) region has made sharp increases to projections of future power demand for Dominion Resources, a utility serving 6 million customers in 15 states. These increases are entirely due to an increase in data centers which serve advanced computing/AI needs.’
What about all those that lambasted the crypto brigade for their energy use? Now, AI promoters are discovering their own obsession is going to drain the grid dry for real.
How much more electricity will we need to produce to fuel AI? How much bigger will the electricity grid need to be to ensure reliable power supplies? How much more will it all cost?
Nobody wants to know…
The nuclear option
The good news is that there’s an obvious solution: build a lot of nuclear power plants. Indeed, leading AI innovators also share a suspicious interest in the nuclear space. This is no coincidence.
I highlight this in an upcoming issue of Strategic Intelligence Australia.
A long list of billionaires and tech conglomerates are punting on a nuclear revival to try and power their AI services. Here’s the most obvious example I mentioned:
Talen Energy conveniently built a data centre and nuclear power plant next to each other. In March, Amazon’s Web Services division bought the data centre and secured rights to the power plant’s power at fixed prices in a long term contract. The data centre is expected to use about 40% of the power plant’s capacity once expanded.
Yes, a data centre that uses 40% of a nuclear power plant’s capacity!
That’s just one of many examples where nuclear power is being paired with AI data centre projects. But investors in these projects are making a terrible mistake.
There’s a rather big fly in the ointment of using nuclear power to secure enough electricity for AI. Actually, a pair of flies working together to sink the AI-nuclear hopes of billionaires, from Bill Gates to Sam Altman.
Electricity is fungible – a shortage is
everyone’s problem
Firstly, the energy ‘transition’ is failing so badly that there simply won’t be enough electricity to go around, let alone for AI.
We don’t have enough electricity generation coming online. We don’t have a good enough transmission network to move electricity far enough to overcome intermittency.
We don’t have an electricity distribution network which can handle the load of delivering electricity to users. We don’t have enough baseload power to overcome intermittency.
It takes decades to solve these problems. While fossil fuel power plants are being shut down in a matter of years.
Not that the problems are being solved to begin with. The rate of expansion of the US grid is falling, not rising rapidly as it needs to.
Solar and wind projects are getting abandoned. Wind and solar energy production in the first quarter of each year has been outright falling in Germany, for example.
The future lack of electricity is becoming so obvious that even politicians and their auditors are openly declaring it. When the German Court of Auditors warned the country’s electricity supply is not secure, the country’s economy minister didn’t disagree or deny it. He furiously declared this was already obvious and therefore unhelpful!
But surely nuclear power fixes this by providing a lot of baseload power?
Well, just how much nuclear it would take is an open question. And it takes a long time to build nuclear power. In Western countries, that is.
More importantly, I don’t know of many places planning on using nuclear power as a primary source of electricity, outside of France. Everyone is on the intermittent energy bandwagon.
And that creates a deeper problem which even building your data centre next to a nuclear power plant cannot overcome.
Tied at the hip
The second fly in the ointment is that nuclear power projects would be part of the failing grid. They don’t stand alone. And so you can’t just claim a share of their electricity.
Those hoping to secure nuclear power to secure their AI data centres are going to discover this the hard way. Do these billionaires really think people will endure electricity shortages while AI data centres keep their lights on?
Do you think politicians will let Amazon’s data centre use 40% of a nuclear plant’s power if the rest of the grid needs it? Imagine shutting down basic services during a cloudy windless week so that people can use AI image generators to entertain themselves.
I doubt it will fly.
As soon as the wider grid is failing due to too much intermittent energy, AI data centres will face the chop.
They’ll be the first source of energy demand to get cut by political decree. It won’t matter that they’re next to a nuclear power plant. It won’t matter what their contract with that power plant says.
But perhaps I’m being dramatic. It’s perfectly plausible that AI is merely priced out of the market. Energy shortages will spike prices. Unaffordable power prices hit major users first, as German industry has discovered. AI would be top of that list.
My real point is that nuclear power plants are simply too big for their own good. They function as a crucial part of a much larger network. In our energy scarce future, everyone will want a piece of their output. And AI will be far down the list of political priorities.
The billionaire’s plans to make their AI projects impervious to the failure of the electricity grid will fail.
Skynet will be defeated by Net Zero. No Hollywood script writer could’ve seen it coming…
There is only one ray of hope for the
techno utopians
They need a form of energy generation that is big enough to power AI data centres, but small enough for energy users like AI data centres to buy and build independently.
It has to be small enough to avoid becoming part of the wider grid, like the off-grid solar panel we see on homes. But also provide reliable and stable enough power to be a viable alternative to the failing grid.
What fits the bill? I’m keeping that for subscribers’ eyes only. Unlike my old friend Callum Newman, who is revealing his ideas to profit from AI here.
Until next time,
 |
Nick Hubble,
Editor, Strategic Intelligence Australia
Comments