Microsoft is hiring a nuclear engineer…
It wants to hook up nuclear reactors to its power-hungry artificial intelligence (AI) data centers.
But these reactors are a little different from the traditional reactors you’re probably used to.
They’re called small modular reactors (SMRs).
They’re closer in size and power output to what’s onboard nuclear submarines.
The “modular” part means they’re made in factories and then transported to the sites where they’ll be installed. This makes them cheaper and faster to get up and running.
Here’s a graphic comparing the size of an SMR to a traditional nuclear plant.
SMRs are less than one-quarter the size of traditional nuclear plants
Source: Idaho National Laboratory
Why is Microsoft looking into hooking up its data centers to SMRs instead of the regular grid?
And why should you care as an investor?
We’ll get into it all in today’s dispatch… including why investing in cutting-edge energy companies is one of the best ways to profit from the AI revolution.
The first thing you need to know is that AI is a power hog.
OpenAI’s ChatGPT and Google’s Bard require warehouses full of specialist computers – called data centers – to work.
And they require a lot of power.
To give you some idea of what scale we’re talking about, the data centers that power today’s internet consume about 1% of the electricity we generate globally.
According to a study published recently in the energy research journal Joule, by 2027, annual worldwide AI-related electricity consumption will increase 56% from 85.4 to 134 terawatt hours (TWh).
That’s roughly the same amount of power Sweden, a country of 10.5 million people, uses in a year.
And figures from consultancy firm McKinsey & Company show that, by 2030, the power requirements from data centers will be double today’s level.
That’s down to the intensive computing that’s needed to make these AI systems work at scale.
A server rack for regular computing uses five kilowatts (kW) of power per hour. That’s about four times as much as a family house uses.
A rack for AI computing consumes about 80kW of power an hour. That’s 16 times more than traditional servers. And you could have hundreds, if not thousands, of these racks within a single data center.
And it’s not as though we’re going to stop rolling out AI systems in 2027 or 2030.
When Microsoft, Amazon, and Apple – along with thousands of other companies around the world – roll out AI systems, energy consumption will skyrocket.
And already the power issue is causing us to throttle back on our AI ambitions.
Take northern Virginia.
It has the highest concentration of data centers in the world. Between 30% and 70% of internet traffic flows through data centers there.
And it’s easy to see why the state is such a popular spot for data centers.
It has blazing-fast internet connections, thanks to all the government agencies that are located in the state. It also has few regulations and generous tax benefits.
More importantly, it has cheap and abundant power. Or rather, it had cheap and abundant power.
Data centers in northern Virginia have a problem. They can’t access enough electricity.
According to Data Center Knowledge, an industry news website, energy bottlenecks in northern Virginia could delay new development into 2026.
Power demand has outgrown the capacity of existing power lines. This is choking the supply of electricity to data centers.
And that’s before you factor in the extra power for data centers we need to train and run AI systems.
It’s not as easy as simply firing up a new coal-powered energy plant.
The Environmental Protection Agency has slapped strict emission limits on power plants. As a result, fossil-fuel power plants are retiring much faster than alternative energy sources are being developed.
That’s why Microsoft is exploring using an SMR to power its AI data centers.
It knows that we can’t rely on the regular grid. And neither solar nor wind is a reliable source of power.
Remember, data centers are mission-critical infrastructure for our digitally-dependent society. Hospitals, GPS, traffic lights, as well as water and wastewater treatment, rely on them. So do defense and intelligence agencies.
So, data centers need reliable power 24/7, 365 days a year. And as we’ve seen in Texas lately, the grid can go down.
That’s where SMRs come in…
SMRs have the benefits of large nuclear plants – they put out affordable, reliable, carbon-free power – but they’re faster and much less expensive to build.
SMRs are the safest nuclear plants ever designed.
For example, they use natural convection for cooling instead of relying on external power-driven cooling systems. They also automatically shut down when they detect abnormal conditions.
Known as “passive safety” features, these operate without human intervention or external power.
That means the type of accidents that happened at Chornobyl, Fukushima, and Three Mile Island are physically impossible with SMRs.
Even better, you can put SMRs almost anywhere.
They have a small footprint – about two acres for the smallest reactors. The average larger reactor requires about 800 acres of land. It also needs to be sited near a lake, river, or ocean for access to water for cooling purposes.
An SMR can fit on a two-acre site. And it doesn’t use water for cooling. So you can install it on the site of a data center – wherever that may be.
And nuclear power isn’t the only energy solution for AI as it scales up globally. On-site battery storage will also play a key role.
Many data centers already use batteries, mostly as a form of backup power.
Batteries also allow data centers to reduce peak power consumption and optimize their operations.
By reducing reliance on the grid, data centers can ensure uninterrupted services and minimize the risk of downtime.
That makes companies that can provide these solutions mission-critical AI plays.
As I’ve been showing you in these pages, AI will transform the world in ways we haven’t yet even begun to fathom.
It’s also our best chance to make life-changing profits as tech investors since the rollout of the internet.
But some of the obvious ways to play it – think AI chipmaker Nvidia – have become crowded plays.
In many cases, sky-high valuations mean investments in these AI darlings have little to no chance of delivering truly life-changing returns.
Instead, because they’re priced for perfection, they’re vulnerable to steep declines if they slip behind – in however minor a way – on their revenue and profit projections.
That’s why my team and I are broadening the scope of our research to include energy solutions for AI. These are ways to profit from the rise of AI without getting suckered into overhyped tech stocks.
We’re just starting that research process now. We haven’t zeroed in on specific stocks yet. But I’ll be updating you on our progress in these pages.
So stay tuned for more on that…
Regards,
Colin Tedards
Editor, The Bleeding Edge
The Bleeding Edge is the only free newsletter that delivers daily insights and information from the high-tech world as well as topics and trends relevant to investments.
The Bleeding Edge is the only free newsletter that delivers daily insights and information from the high-tech world as well as topics and trends relevant to investments.
